Intelligent shelter equipment operation state monitoring method and system based on machine learning

Abstract

The invention provides an intelligent shelter equipment operation state monitoring method and system based on machine learning, and the method comprises the steps: firstly obtaining a synchronous operation real-time signal set which comprises the real-time parameters of all operation parts, the interaction between equipment and the comprehensive influence of a shelter environment, and carrying out the mutual feedback correlation processing of the synchronous operation real-time signal set to extract a mutual feedback correlation feature set; inputting the set into an operation state deduction model to generate a short-term operation state evolution sequence, identifying a potential abnormal evolution trend and determining an abnormal association influence range based on the short-term operation state evolution sequence, and generating collaborative regulation and control demand information; then, inputting the collaborative regulation demand information and the mutual feedback correlation feature set into a collaborative regulation model to generate a multi-device collaborative operation regulation instruction, after the multi-device collaborative operation regulation instruction is sent, collecting an execution feedback signal, updating a feature extraction rule, and adjusting model parameters; the comprehensive, dynamic and accurate monitoring and regulation of the operation state of the square cabin equipment are realized, and the stability and the reliability of the equipment operation are improved.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and more specifically, to a method and system for monitoring the operational status of smart mobile cabin equipment based on machine learning. Background Technology

[0002] In the field of operational management of mobile medical facilities, traditional monitoring methods mainly focus on the independent monitoring of individual equipment components. For example, for electrical equipment within the facility, only basic parameters such as voltage and current are typically monitored; for ventilation equipment, only indicators such as wind speed and air volume are considered. These isolated monitoring methods cannot comprehensively consider the mutual influence between various components during equipment operation.

[0003] In actual operation, the various devices within the shelter are not isolated but interconnected and work collaboratively. The interaction signals between devices, such as the impact of electrical equipment on the operating power of ventilation equipment, and the cross-influence of comprehensive environmental factors (such as temperature, humidity, and air pressure) on the operating status of the equipment, have not been effectively integrated and analyzed. This makes it difficult to accurately determine the root cause and scope of the anomaly when it occurs, hindering timely and effective coordinated control measures. Consequently, the risk of equipment failure increases, impacting the overall operational stability and reliability of the shelter. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for monitoring the operational status of intelligent mobile cabin equipment based on machine learning, the method comprising: Acquire a set of real-time synchronous operation signals, which includes real-time parameter signals of each operating component, inter-equipment interaction signals, and comprehensive environmental impact signals of the container. Perform mutual feedback correlation processing on the set of synchronous real-time signals to extract the set of mutual feedback correlation features in the operating signals. The set of mutual feedback correlation features includes component parameter coupling features, signal transmission features between devices, and environmental parameter cross features. The set of mutually fed-back correlation features is input into the operation status inference model, and a short-term operation status evolution sequence is generated through time-series feature progressive calculation. The short-term operation status evolution sequence includes the predicted values ​​of the parameters of the operating components and the predicted values ​​of the overall operation coordination in each time period. Based on the short-term operational status evolution sequence, potential abnormal evolution trends are identified, and the scope of abnormal correlation impact is determined by combining the mutual feedback correlation feature set, thereby generating information on coordinated regulation and control requirements. The collaborative control demand information and the set of mutual feedback correlation features are input into the collaborative control model to generate multi-device collaborative operation control instructions. These instructions are then sent to the corresponding control modules. The set of synchronous operation feedback signals after execution is collected and input into the mutual feedback correlation processing stage to update the extraction rules of the mutual feedback correlation feature set. At the same time, the timing operation parameters of the operation status inference model and the instruction generation logic of the collaborative control model are adjusted.

[0005] Furthermore, embodiments of the present invention also provide a machine learning-based intelligent mobile cabin equipment operation status monitoring system, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned machine learning-based smart shelter equipment operation status monitoring method by executing the machine-executable instructions.

[0006] Based on the above, by acquiring a set of synchronous real-time operation signals including real-time parameter signals of each operating component, inter-equipment interaction signals, and comprehensive environmental impact signals of the shelter, mutual feedback correlation processing is performed on the set of synchronous real-time operation signals to extract component parameter coupling characteristics, inter-equipment signal transmission characteristics, and environmental parameter cross-characteristics. This deeply explores the intrinsic connections between various elements during equipment operation. The set of mutual feedback correlation features is input into the operation status prediction model to generate a short-term operation status evolution sequence. This allows for the prediction of operating component parameters and overall operation coordination at different time periods. Based on the short-term operation status evolution sequence, potential abnormal evolution trends are identified, and the scope of abnormal correlation impact is determined by combining the set of mutual feedback correlation features, generating coordinated control demand information, making control measures more targeted and effective. Finally, the coordinated control model generates multi-equipment coordinated operation control commands, and updates feature extraction rules and adjusts model parameters based on the feedback signals after execution. This achieves dynamic and precise monitoring and control of equipment operation status, greatly improving the stability, reliability, and intelligence level of shelter equipment operation. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the machine learning-based smart shelter equipment operation status monitoring method provided in the embodiments of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a machine learning-based smart shelter equipment operation status monitoring method according to an embodiment of the present invention. The following is a detailed description of this machine learning-based smart shelter equipment operation status monitoring method.

[0009] Step S110: Obtain a set of real-time synchronous operation signals, which includes real-time parameter signals of each operating component, inter-equipment interaction signals, and comprehensive environmental impact signals of the container.

[0010] In this embodiment, the application scenario is the monitoring of the operational status of a smart medical modular unit. This smart medical modular unit contains multiple key devices, such as a medical diagnostic equipment group, a life support equipment group, and an environmental control equipment group. Each device consists of multiple operating components. For example, a CT scanner in the medical diagnostic equipment group includes rotating components, X-ray generating components, data acquisition components, and other operating components.

[0011] To obtain a set of real-time signals for synchronous operation, a multi-type sensor network needs to be deployed within the shelter. For the real-time parameter signals of each operating component, vibration sensors are installed on the rotating parts of the CT scanner to collect parameters such as vibration frequency and amplitude; temperature sensors are installed on the X-ray generating components to collect operating temperature parameters; and data transmission rate sensors are installed on the data acquisition components to collect data transmission speed parameters. These sensors continuously collect data at a preset sampling frequency, which is set according to the characteristics of different components. For example, the sampling frequency of the vibration sensor is higher than that of the temperature sensor to ensure that rapidly changing vibration signals can be captured.

[0012] Inter-device communication signals are collected through communication interfaces between the devices. Within the mobile medical facility, devices exchange data using dedicated medical device communication protocols. For example, a CT scanner transmits scanned image data and control commands to an image storage server, while life support equipment transmits patient vital signs data and equipment operating status commands to a central monitoring system. By deploying signal acquisition modules on the communication bus, these interactive signals are captured in real time, including information such as signal transmission time, reception time, data length, and checksum.

[0013] The comprehensive environmental impact signals of the makeshift hospital are collected through environmental sensors deployed in different areas of the hospital, including temperature sensors, humidity sensors, air pressure sensors, and electromagnetic interference sensors. These sensors are distributed in key locations such as areas with dense equipment, patient areas, and air circulation areas to comprehensively reflect the environmental conditions inside the makeshift hospital. For example, temperature and humidity sensors are deployed near the CT scanner, air pressure sensors are deployed at the entrance of the makeshift hospital, and electromagnetic interference sensors are deployed near the power supplies of various equipment.

[0014] The collected signals are aggregated through a data aggregation gateway, which performs time synchronization processing on the signals transmitted from various sensors and communication interfaces. Time synchronization employs a network time protocol, unifying the timestamps of all signals with the standard clock system within the shelter, ensuring consistency across signals in the time dimension, thus forming a synchronized real-time signal set. This set is stored in the form of data frames, each containing information such as signal type identifier, acquisition timestamp, signal value, and device component identifier.

[0015] Step S120: Perform mutual feedback correlation processing on the set of synchronous real-time signals, and extract the mutual feedback correlation feature set in the operating signals. The mutual feedback correlation feature set includes component parameter coupling features, signal transmission features between devices, and environmental parameter cross features.

[0016] Step S121: Separate the real-time parameter signals of each operating component from the set of synchronous real-time signals, and extract the time-domain variation characteristics of the parameters of each operating component. The time-domain variation characteristics include parameter fluctuation period characteristics and parameter change rate characteristics.

[0017] In this embodiment, the synchronous real-time signal set is first separated. Based on the device component identifier contained in the signal, the real-time parameter signals belonging to the same operating component are extracted. For example, for the rotating component of a CT scanner, all signals with the device component identifier "CT-rotating component" are filtered from the synchronous real-time signal set to obtain the real-time vibration frequency signal and amplitude signal of the rotating component.

[0018] For each extracted parameter signal of the operating component, time-domain variation features are extracted. Taking the vibration frequency signal of the rotating component as an example, the parameter fluctuation period characteristics are obtained by periodic analysis of the vibration frequency signal. Using the sliding window method, the vibration frequency signal is divided into multiple continuous time windows. Autocorrelation analysis is performed on the signal within each time window to determine the main fluctuation period of the signal. For example, by calculating the autocorrelation coefficient of the signal at different time intervals, the time interval corresponding to the maximum autocorrelation coefficient is the fluctuation period of the signal within that window. Statistical analysis of the fluctuation periods of multiple windows yields the parameter fluctuation period characteristics of the operating component parameter. This characteristic is a sequence containing multiple period values, reflecting the fluctuation pattern of the vibration frequency over different time periods.

[0019] The parameter change rate characteristics are obtained by calculating the first derivative of the parameter signal. The vibration frequency signal is subjected to time series differencing, i.e., the ratio of the vibration frequency difference between two adjacent sampling points to the sampling time interval is calculated to obtain the change rate at each sampling point. The resulting change rate sequence is smoothed to remove noise interference, and then features such as the maximum, minimum, average, and trend of the change rate are extracted to form the parameter change rate characteristics. For example, the change trend is determined by linearly fitting the smoothed change rate sequence and judging whether the change rate is increasing, decreasing, or remaining stable based on the slope of the fitted line.

[0020] Step S122: Separate the device-to-device interaction signals from the synchronous real-time signal set, and extract the transmission characteristics of the device-to-device interaction signals. The transmission characteristics include signal transmission delay characteristics and interaction parameter matching deviation characteristics.

[0021] Inter-device interaction signals are separated from the set of synchronously running real-time signals and filtered based on the source device identifier and destination device identifier in the signals. For example, all interaction signals with the source device identifier "CT scanner" and the destination device identifier "image storage server" are filtered out to obtain the interaction signal sequence between these two devices.

[0022] Signal transmission delay characteristics are obtained by calculating the difference between the transmission time and the reception time of the interactive signal. The data frame of the inter-device interactive signal contains a transmission timestamp and a reception timestamp. Subtracting the transmission timestamp from the reception timestamp yields the transmission delay time of each interactive signal. Statistical analysis of the transmission delay times of multiple interactive signals extracts features such as the average, variance, maximum, and frequency of delay occurrences, forming the signal transmission delay characteristics. For example, the number of times the transmission delay exceeds a preset threshold within a certain period can be counted to reflect the severity of the signal transmission delay.

[0023] Interaction parameter matching deviation features are used to measure the degree of deviation between data parameters and expected parameters in an interaction signal. Taking the transmission of image data from a CT scanner to an image storage server as an example, the expected image data format includes parameters such as image resolution, pixel depth, and data compression rate. The actual image data parameters are extracted from the interaction signal and compared with the preset expected parameters. For example, the percentage difference between the actual and expected image resolutions, or the difference between the actual and expected pixel depths, are considered. After normalizing these deviation values, they are combined to form the interaction parameter matching deviation feature. This feature reflects the consistency and accuracy of data exchanged between devices.

[0024] Step S123: Separate the comprehensive influence signal of the cabin environment from the real-time signal set of synchronous operation, and extract the differential effect characteristics of the environmental signal on the parameters of different operating components. The differential effect characteristics include the influence amplitude characteristics and influence lag characteristics of the environmental parameters on the component parameters.

[0025] The comprehensive environmental impact signals of the mobile cabin are separated from the real-time signal set of synchronous operation, and environmental signals such as temperature, humidity, air pressure, and electromagnetic interference are filtered out according to the signal type identifier. For example, signals with signal type identifiers such as "temperature" and "humidity" are extracted to obtain the comprehensive environmental impact signals of the mobile cabin.

[0026] This study analyzes the differentiated effects of environmental signals on various operating component parameters. Taking the influence of temperature signals on the vibration frequency of a CT scanner's rotating component and the operating temperature of its X-ray generating component as an example, the magnitude of the influence is determined by analyzing the correlation between changes in environmental parameters and changes in component parameters. Using a sliding window method, the temperature signal and the vibration frequency signal of the rotating component are divided into multiple time windows. Within each window, the ratio of temperature change to vibration frequency change is calculated; this ratio reflects the magnitude of the temperature's influence on the vibration frequency. The magnitude values ​​across multiple windows are statistically analyzed to obtain the average, maximum, and distribution characteristics of the influence magnitude, thus forming the magnitude characteristics of the influence of environmental parameters on component parameters.

[0027] The lag characteristic is used to measure the time delay in the impact of environmental parameter changes on component parameters. It is calculated by the difference between the start time of the environmental parameter change and the start time of the component parameter change. For example, when the temperature inside the shelter begins to rise from a certain value, the time point at which the temperature begins to rise is recorded, and simultaneously, the time point at which the vibration frequency of a rotating component begins to change is observed. The difference between these two time points is the lag time of the temperature's impact on the vibration frequency. Statistical analysis of the lag times corresponding to multiple environmental parameter change events yields the lag characteristics, including the mean and variance of the lag time.

[0028] Step S124: Perform a first-level correlation operation on the time-domain variation features and the conduction features, calculate the coupling coefficient of different operating component parameters and inter-device interaction signals within the same device, and generate the first correlation feature.

[0029] In this embodiment, a CT scanner is used as an example. It includes operating components such as a rotating component, a radiation generating component, and a data acquisition component, and there are device-to-device interaction signals between it and the image storage server. First, the time-domain variation characteristics of the rotating component (fluctuation period characteristics and rate of change characteristics of vibration frequency), the time-domain variation characteristics of the radiation generating component (fluctuation period characteristics and rate of change characteristics of operating temperature), and the time-domain variation characteristics of the data acquisition component (fluctuation period characteristics and rate of change characteristics of data transmission rate) are correlated with the transmission characteristics of the device-to-device interaction signals between the CT scanner and the image storage server (signal transmission delay characteristics and interaction parameter matching deviation characteristics).

[0030] For calculating the coupling coefficient between parameters of different operating components within the same equipment and the interaction signals between equipment, a correlation analysis method is used. For example, the correlation between the vibration frequency change rate characteristics of a rotating component and the signal transmission delay characteristics of the interaction signals between equipment is analyzed. The vibration frequency change rate sequence and the signal transmission delay sequence are time-aligned, and then the correlation coefficient between the two sequences is calculated. This correlation coefficient is one of the coupling coefficients between the rotating component parameters and the interaction signals between equipment.

[0031] Following the same method, the correlation coefficients between the fluctuation period characteristics of the operating temperature of the radiation generating component and the matching deviation characteristics of the interaction parameters, and the correlation coefficients between the change rate characteristics of the data transmission rate of the data acquisition component and the signal transmission delay characteristics, were calculated separately. These coupling coefficients were then combined to form the first correlation feature, which reflects the degree of correlation between the parameters of each operating component within the same device and the interaction signals between devices.

[0032] Step S125: Perform a second-level association operation on the first association feature and the differential action feature, calculate the indirect influence coefficient of the environmental signal on the parameters of the operating components through the interaction between devices, and generate the second association feature.

[0033] Taking the temperature environment signal inside the mobile CT scanner as an example, its differential effects include the amplitude and hysteresis characteristics of its influence on the vibration frequency of the rotating components of the CT scanner, as well as the amplitude and hysteresis characteristics of its influence on the hard drive temperature of the image storage server. The first correlation feature includes the coupling coefficient between the parameters of each operating component inside the CT scanner and the interactive signals between the CT scanner and the image storage server.

[0034] The calculation of the indirect influence coefficient of environmental signals on the parameters of operating components through inter-device interaction considers the process by which environmental signals first affect the parameters of one device's components, and the changes in these parameters then affect the parameters of another device's components through inter-device interaction signals. For example, an increase in temperature affects the temperature of the hard drive in an image storage server. Changes in hard drive temperature cause changes in the matching deviation characteristics of the interaction parameters between the hard drive and the CT scanner, thereby affecting the data transmission rate of the CT scanner's data acquisition components.

[0035] To calculate the indirect influence coefficient, first, the magnitude of the environmental signal's influence on the first device component parameter is determined (obtained from the differential action feature). Then, the coupling coefficient between this component parameter and the inter-device interaction signal is determined (obtained from the first correlation feature). Next, the magnitude of the inter-device interaction signal's influence on the second device component parameter is determined (combining the differential action feature and the first correlation feature). Multiplying these three values ​​yields the indirect influence coefficient. The indirect influence coefficients under different environmental signals and different inter-device interaction paths are calculated and combined to generate the second correlation feature.

[0036] Step S126: After unifying the dimensions of the first and second associated features, based on the feature importance evaluation results, assign dynamic weights to the unified first and second associated features, integrate the weighted first and second associated features, and generate a set of mutually fed-back associated features that includes component parameter coupling features, signal transmission features between devices, and cross-feedback features of environmental parameters.

[0037] The first and second correlation features may have different dimensions. For example, the first correlation feature may contain multiple coupling coefficients, each a numerical value, while the second correlation feature may contain multiple indirect influence coefficients, each also a numerical value, but the number of coefficients may differ. Dimensionality unification is achieved using feature vectorization, converting both the first and second correlation features into fixed-length feature vectors. For instance, if the first correlation feature contains 5 coupling coefficients, it is converted into a 5-dimensional vector; if the second correlation feature contains 4 indirect influence coefficients, it is converted into a 4-dimensional vector. Then, through feature expansion or dimensionality reduction, the two vectors are adjusted to the same dimension, for example, both to 10-dimensional vectors. Feature expansion can use zero-padding, adding zero elements to the end of the vector to achieve the target dimension; dimensionality reduction can use principal component analysis to extract principal components and reduce the vector dimension.

[0038] Feature importance assessment employs a tree-based feature importance assessment method. A decision tree model is constructed, using the device operating status (normal or abnormal) corresponding to the set of synchronously running real-time signals as labels, and the first and second associated features as input features to train the decision tree model. After training, the importance weight of each feature is determined based on its contribution to node splitting in the decision tree model. For example, if a coupling coefficient plays a key role in multiple node splits in the decision tree, its importance weight is higher.

[0039] Based on the feature importance assessment results, dynamic weights are assigned to each feature component in the unified first and second correlation features. Feature components with higher importance weights are assigned larger weight values, while those with lower importance weights are assigned smaller weight values. Then, the weighted first and second correlation feature vectors are concatenated and integrated, that is, the two vectors are connected sequentially to form a higher-dimensional feature vector, which is the set of mutual feedback correlation features. Specifically, the portion of the original first correlation feature reflecting the correlation of parameters between different components within the same device constitutes the component parameter coupling feature, the portion reflecting the correlation of interactive signals between devices constitutes the inter-device signal transmission feature, and the portion of the original second correlation feature reflecting the correlation between environmental signals and device component parameters constitutes the environmental parameter cross feature.

[0040] Step S127: Associate the mutual feedback correlation feature set with the acquisition time period information of the synchronous operation real-time signal set, and extract the core correlation feature subset in the mutual feedback correlation feature set. The core correlation feature subset is a feature combination whose influence weight on the prediction of the operating state exceeds a preset weight threshold.

[0041] The acquisition period information of the synchronously running real-time signal set includes the start and end times of acquisition for each signal. The generation time of the mutual feedback associated feature set is matched with the acquisition period information, thus associating the mutual feedback associated feature set with the corresponding acquisition period. For example, if a certain mutual feedback associated feature set is generated based on the synchronously running real-time signal set from 8:00 AM to 8:30 AM, then this feature set will be associated and stored with the acquisition period information of "8:00 AM to 8:30 AM".

[0042] The extraction of the core associated feature subset is based on the previously calculated feature importance weights. A preset weight threshold is established, for example, set as the average of all feature importance weights. Each feature component in the feedback associated feature set is iterated over, and its importance weight is checked against the preset weight threshold. Feature components exceeding the threshold are extracted and combined to form the core associated feature subset. For example, if the feedback associated feature set has 10 feature components, and 6 of them have importance weights exceeding the preset threshold, then these 6 feature components constitute the core associated feature subset. This subset contains key features that significantly impact operational state prediction and can be used in subsequent operational state extrapolation processes.

[0043] Step S130: Input the set of mutually fed-back correlation features into the operation state inference model, and generate a short-term operation state evolution sequence through time-series feature progressive operation. The short-term operation state evolution sequence includes the predicted values ​​of the parameters of the operating components in each time period and the predicted value of the overall operation coordination.

[0044] Step S131: Input the mutual feedback correlation feature set into the feature preprocessing layer of the running state inference model, perform temporal alignment processing on the mutual feedback correlation feature set, pass the temporally aligned mutual feedback correlation feature set to the first temporal operation layer of the running state inference model, extract the local temporal correlation information of the mutual feedback correlation feature set through convolution operation, and generate local temporal feature vector.

[0045] In this embodiment, the feature components in the feed-through feature set are extracted at different time points, which may result in temporal misalignment. The feature preprocessing layer first performs timestamp calibration on all feature components in the feed-through feature set to ensure they are consistent in the time dimension. For example, the time series of each feature component is interpolated or resampled to a uniform time interval, so that all feature components have corresponding values ​​at the same time point.

[0046] The temporally aligned, mutually correlated feature set is represented as a three-dimensional tensor. The first dimension is the time step, the second dimension is the number of feature channels (i.e., the number of feature components in the mutually correlated feature set), and the third dimension is the number of samples (1 in real-time processing). This three-dimensional tensor is input into the first temporal computation layer, which employs a one-dimensional convolutional neural network structure. The size of the convolutional kernel is determined based on the temporal range of the local temporal correlation; for example, setting the temporal dimension of the convolutional kernel to 5 means that each convolution operation considers features from 5 consecutive time steps.

[0047] Convolution operations involve sliding the convolution kernel along the time dimension, performing weighted summation and non-linear activation on the features within each sliding window. For example, for a window with a time step of t, each element in the convolution kernel is multiplied by the feature value of the corresponding time step and feature channel within that window. All products are then summed, a bias term is added, and finally, the ReLU activation function is applied to obtain the convolution output value for that window. After performing convolution operations on windows of all time steps, a local temporal feature map is obtained. Flattening this feature map into a vector form generates a local temporal feature vector. This vector contains the correlation information of the mutually fed-back related feature sets within the local time range.

[0048] Step S132: Pass the local time series feature vector to the second time series operation layer of the running state inference model. The local time series feature vector is modeled in a long-term correlation through a loop unit to capture the dependency relationship of the local time series feature vector within a continuous time step and generate a long-term time series feature vector.

[0049] The second temporal computation layer employs a long short-term memory recurrent unit (LSM) structure. Local temporal feature vectors are input into the recurrent unit in chronological order, with the local temporal feature vector at each time step serving as the input to the recurrent unit. The recurrent unit contains input gates, forget gates, and output gates, using these gating mechanisms to selectively remember and forget historical information.

[0050] At each time step, the input gate determines how much information from the current local temporal feature vector is retained in the cell state; the forget gate determines how much information from previous cell states is forgotten; and the output gate generates the output for the current time step based on the current cell state and the hidden state. In this way, the recurrent unit can capture the dependencies of local temporal feature vectors over a longer time range, such as the impact of changes in a feature at multiple time steps on the feature at the current time step.

[0051] After processing at all time steps, the hidden state sequence of the recurrent unit constitutes a long-term temporal feature vector. Each element of this vector corresponds to the hidden state at one time step, containing long-term temporal association information from the initial time step to that time step.

[0052] Step S133: Pass the long-term time series feature vector to the feature abstraction layer of the running state inference model, and perform high-order feature abstraction on the long-term time series feature vector through a multilayer perceptron to generate a high-order running state feature vector.

[0053] Long-term temporal feature vectors still contain a significant amount of original temporal information. The feature abstraction layer is constructed using a multilayer perceptron, which maps the long-term temporal feature vectors to a higher-level abstract feature space. The multilayer perceptron contains multiple fully connected layers. The number of input neurons in the first layer is equal to the length of the long-term temporal feature vector, and the number of neurons in subsequent layers gradually decreases to achieve feature dimensionality reduction and abstraction.

[0054] For example, the first fully connected layer has 256 neurons, transforming a long-term temporal feature vector of length 512 into a 256-dimensional feature vector; the second fully connected layer has 128 neurons, transforming the 256-dimensional feature vector into a 128-dimensional feature vector; and the third fully connected layer has 64 neurons, transforming the 128-dimensional feature vector into a 64-dimensional higher-order running state feature vector. The output of each fully connected layer undergoes a non-linear transformation using the ReLU activation function, enhancing the model's expressive power. The higher-order running state feature vector contains an abstract representation of the running state, better reflecting its essential characteristics.

[0055] Step S134: Pass the high-order operating state feature vector to the prediction output layer of the operating state inference model, and convert the high-order operating state feature vector into the predicted values ​​of the operating component parameters for each time period through linear mapping operation. At the same time, calculate the coordination coefficient of the predicted values ​​of the different operating component parameters for each time period to generate the overall operating coordination degree prediction value for each time period.

[0056] The prediction output layer contains two parallel output branches: one branch predicts the parameters of operating components for each time period, and the other branch predicts the overall operational coordination for each time period. For the operating component parameter prediction branch, the higher-order operational state feature vector is linearly mapped through a fully connected layer. The number of output neurons in this fully connected layer is equal to the number of operating component parameters to be predicted. For example, if the operating component parameters to be predicted in the mobile shelter include eight parameters such as the vibration frequency of the CT scanner's rotating component, the operating temperature of the X-ray generator, and the data transmission rate of the data acquisition component, then this fully connected layer has eight output neurons, and the output value of each neuron is the predicted value of the corresponding operating component parameter for a future time period.

[0057] For the overall operational synergy prediction branch, the synergy coefficients of the predicted values ​​of different operational component parameters in each time period are first calculated. The synergy coefficients are obtained by calculating the correlation coefficient matrix between the predicted values ​​of different operational component parameters, where each element represents the degree of linear correlation between two predicted values ​​of operational component parameters. Then, the correlation coefficient matrix is ​​decomposed into eigenvalues, and the eigenvector corresponding to the largest eigenvalue is taken as the synergy weight vector. The predicted values ​​of each operational component parameter are then weighted and summed with the synergy weight vector to obtain the overall operational synergy prediction value. Alternatively, a dedicated fully connected layer can be used to process the higher-order operational state feature vectors, directly outputting the overall operational synergy prediction value. This fully connected layer has only one output neuron.

[0058] Step S135: Arrange the predicted values ​​of the operating component parameters for each time period and the predicted values ​​of the overall operating coordination for the corresponding time period in chronological order to form an initial short-term operating state evolution sequence.

[0059] The predicted values ​​of operating component parameters and overall operational synergy for each time period are output sequentially in future time order. For example, the predicted operating status is given for each 5-minute period within the next hour, for a total of 12 time periods. The predicted values ​​of 8 operating component parameters and 1 overall operational synergy for each time period are combined into a state vector. Then, the state vectors of the 12 time periods are arranged in chronological order to form a two-dimensional matrix, where rows represent time periods and columns represent the predicted values ​​of operating component parameters and overall operational synergy. This matrix is ​​the initial short-term operational status evolution sequence.

[0060] Step S136: Perform trend smoothing on the initial short-term operating state evolution sequence, extract key time node features from the smoothed short-term operating state evolution sequence, and associate and store the key time node features with the smoothed short-term operating state evolution sequence. The key time node features are the features corresponding to the time nodes when the predicted values ​​of the operating component parameters or the predicted values ​​of the overall operating coordination degree change significantly.

[0061] Trend smoothing employs a moving average method to smooth the predicted parameter sequences of each operating component and the overall operational synergy sequence in the initial short-term operational state evolution sequence. For example, using a moving average window of size 3, the predicted value for time period t is replaced by the average of the predicted values ​​for time periods t-1, t, and t+1 (a single-sided window is used for the time periods at the ends of the sequence). This smoothing process removes noise and random fluctuations from the predicted values, making the trend of the sequence more apparent.

[0062] In the smoothed short-term operation state evolution sequence, the identification of key time nodes is achieved by detecting the change rate of the predicted values. For each predicted value sequence of the operation component parameters and the predicted value sequence of the overall operation coordination degree, calculate the ratio of the difference between the predicted values of two adjacent time periods to the time interval to obtain the change rate. Preset a change rate threshold. When the change rate of a certain time period exceeds this threshold, this time period is considered a key time node.

[0063] For key time nodes, extract all the predicted values of the operation component parameters and the predicted value of the overall operation coordination degree corresponding to this node, as well as the change trend of the predicted values in the time periods before and after this node (such as an upward trend, a downward trend, or a stable trend). These information together constitute the characteristics of the key time node. Store the characteristics of the key time node and the smoothed short-term operation state evolution sequence in the same data structure. The characteristics of the key time node are associated with the corresponding time period in the smoothed sequence through timestamps, so as to quickly locate the key time points when identifying potential abnormal evolution trends later.

[0064] Step S140: Identify potential abnormal evolution trends based on the short-term operation state evolution sequence, combine the mutual feedback association feature set to determine the scope of abnormal association influence, and generate coordinated control demand information.

[0065] Step S141: Extract the predicted values of the operation component parameters in each time period of the short-term operation state evolution sequence, compare them with the preset normal operation parameter range, mark the predicted values of the operation component parameters that exceed the normal operation parameter range and the corresponding time periods, and count the number of time periods in which each operation component exceeds the normal parameter range in the short-term operation state evolution sequence, and calculate the abnormal time period ratio. The abnormal time period ratio is the ratio of the number of time periods that exceed the normal parameter range to the total number of predicted time periods.

[0066] In this embodiment, the preset normal operation parameter range is determined according to the design specifications and historical operation data of each operation component in the field hospital. For example, the normal vibration frequency range of the rotating component of the CT scanner is [f_min, f_max], and the normal working temperature range of the ray generating component is [T_min, T_max], etc. The above normal ranges are stored in the device parameter database and can be adjusted according to the device model and service life.

[0067] Extract the predicted values of each operation component parameter in each time period from the short-term operation state evolution sequence. For example, the predicted vibration frequency f1 of the rotating component in time period 1, the predicted vibration frequency f2 in time period 2, etc. Compare each predicted value with the corresponding normal operation parameter range. If f1 < f_min or f1 > f_max, mark f1 as a predicted value that exceeds the normal range, and record the corresponding time period 1 at the same time.

[0068] For each operating component, count the number of time periods during which its parameters exceed the normal range in all prediction time periods. For example, if the predicted vibration frequency of a rotating component exceeds the normal range in 3 out of 12 prediction time periods, then the number of time periods during which the operating component exceeds the normal parameter range is 3. The total number of prediction time periods is 12, so the proportion of abnormal time periods is 3 / 12 = 0.25.

[0069] Step S142: Extract the predicted values of the overall operation coordination degree for each time period in the short-term operation state evolution sequence, compare them with the preset normal coordination degree range, mark the predicted values of the overall operation coordination degree and the corresponding time periods that are lower than the normal coordination degree range, count the number of time periods during which the predicted values of the overall operation coordination degree are lower than the normal coordination degree range, and calculate the proportion of coordination abnormal time periods.

[0070] The preset normal coordination degree range is determined according to the collaborative work requirements of the equipment in the shelter. For example, the normal coordination degree range is [C_min, C_max], and this range is obtained by analyzing the distribution of coordination degree values during the normal collaborative operation of the equipment in history. Extract the predicted values of the overall operation coordination degree for each time period from the short-term operation state evolution sequence, such as the predicted coordination degree C1 for time period 1, the predicted coordination degree C2 for time period 2, etc.

[0071] Compare each predicted coordination degree value with the normal coordination degree range. If C1 < C_min, then mark C1 as a predicted value lower than the normal range, and record the corresponding time period 1 at the same time. Count the number of time periods during which the predicted values of the overall operation coordination degree are lower than the normal coordination degree range. For example, if the predicted coordination degree values in 2 out of 12 prediction time periods are lower than the normal range, then the number of coordination abnormal time periods is 2, and the proportion of coordination abnormal time periods is 2 / 12 ≈ 0.17.

[0072] Step S143: If the proportion of abnormal time periods of any operating component exceeds the preset component abnormal threshold, or the proportion of coordination abnormal time periods of the overall operation coordination degree exceeds the preset coordination abnormal threshold, then it is determined that there is a potential abnormal evolution trend.

[0073] The preset component abnormal threshold and the preset coordination abnormal threshold are set according to the importance and fault tolerance of the shelter equipment. For example, the preset component abnormal threshold is set to 0.2, and the preset coordination abnormal threshold is set to 0.15. In this embodiment, the proportion of abnormal time periods of the rotating component is 0.25, which exceeds the preset component abnormal threshold of 0.2. Therefore, it is determined that there is a potential abnormal evolution trend. Even if the proportion of abnormal time periods of all operating components does not exceed the component abnormal threshold, but if the proportion of coordination abnormal time periods of the overall operation coordination degree exceeds the coordination abnormal threshold, it will also be determined that there is a potential abnormal evolution trend.

[0074] Step S144: Extract the component parameter prediction value change curves corresponding to the operating components with potential abnormal evolution trends, analyze the slope and acceleration of the component parameter prediction value change curves, and determine the rate characteristics of abnormal evolution.

[0075] The rotating component of the CT scanner is identified as a potential source of abnormal evolution. The predicted vibration frequency changes in its short-term operational sequence are extracted. This curve is plotted with time period on the x-axis and predicted vibration frequency on the y-axis. The slope is obtained by linearly fitting this curve, for example, using the least squares method to fit a straight line. The slope of this line represents the change slope; a positive slope indicates an increasing vibration frequency trend, while a negative slope indicates a decreasing trend. The absolute value of the slope indicates the steepness of the trend.

[0076] The rate of change of acceleration is obtained by calculating the rate of change of the slope. The curve is divided into multiple continuous sub-intervals. Within each sub-interval, the slope of the linear fit is calculated. Then, the ratio of the difference in slope between adjacent sub-intervals to the time length of the sub-interval is calculated to obtain the rate of change of acceleration. A positive rate of change of acceleration indicates that the slope is increasing, i.e., the rate of change of vibration frequency is accelerating; a negative rate of change of acceleration indicates that the slope is decreasing, i.e., the rate of change is slowing down. The slope and acceleration together constitute the rate characteristic of anomaly evolution, reflecting the development speed and changing trend of the anomaly.

[0077] Step S145: Combining the component parameter coupling features in the mutual feedback correlation feature set, find other operating components that are coupled with operating components that have potential abnormal evolution trends, and form a list of directly related components.

[0078] The component parameter coupling features in the mutual feedback correlation feature set include the coupling coefficients between parameters of different operating components within the same device. In this embodiment, the component parameter coupling features of the rotating component include the coupling coefficients with other operating components such as the ray generating component and the data acquisition component. The magnitude of the coupling coefficient reflects the tightness of the association between components. A preset coupling coefficient threshold is set; when the coupling coefficient between two operating components exceeds this threshold, they are considered to have a coupling relationship.

[0079] For example, the coupling coefficient between the rotating component and the X-ray generating component is 0.7, exceeding the preset threshold of 0.5. Therefore, the X-ray generating component is a directly associated component that is coupled with the rotating component. The coupling coefficient between the rotating component and the data acquisition component is 0.6, also exceeding the threshold, so the data acquisition component is also included in the list of directly associated components. The identifiers of all operating components that are coupled with the rotating component (such as "CT-X-ray Generating Component" and "CT-Data Acquisition Component") are added to the list of directly associated components.

[0080] Step S146: Combining the signal transmission characteristics between devices in the mutual feedback correlation feature set, find other devices that are associated with devices that have potential abnormal evolution trends through signal transmission, and form a list of indirectly related devices.

[0081] The CT scanner is a device with a potential for anomalous evolution. The signal transmission characteristics between devices in the mutual feedback correlation feature set include signal transmission delay characteristics and interaction parameter matching deviation characteristics between the CT scanner and other devices. By analyzing these characteristics, it is possible to determine which devices have signal transmission correlations with the CT scanner.

[0082] For example, frequent image data interaction occurs between CT scanners and image storage servers. The signal transmission delay and interaction parameter matching deviation characteristics have high weight in the inter-device signal transmission characteristics. Therefore, the image storage server is a device associated with the CT scanner through signal transmission. Furthermore, the CT scanner transmits patient scan plans and device status information with the central monitoring system, also involving signal transmission. The central monitoring system is also included in the list of indirectly associated devices. All device identifiers associated with the CT scanner through signal transmission are added to the list of indirectly associated devices.

[0083] Step S147: Integrate the list of directly associated components and the list of indirectly associated devices to determine the scope of impact of abnormal associations. The scope of impact of abnormal associations includes operating components and devices that are potentially affected by abnormalities.

[0084] The directly associated components list includes CT scanners, while the indirectly associated devices list consists of other independent devices that have signal transmission connections with the CT scanner. When merging these two lists, it's necessary to distinguish the hierarchical relationship between the operating components and the devices. Merging the devices (i.e., CT scanners) belonging to the operating components in the directly associated components list with the devices in the indirectly associated devices list yields a set of devices potentially affected by anomalies. This set includes CT scanners, image storage servers, central monitoring systems, etc.

[0085] Within each affected device, in addition to the operating components listed in the directly associated component list, there may be other operating components coupled to these directly associated components, which may also be affected by the potential anomaly. For example, the X-ray generator of a CT scanner is a directly associated component, and it is coupled to the cooling component of the CT scanner, which may also be affected. Therefore, it is necessary to expand the list of operating components within each affected device to include other operating components coupled to the directly associated components within the scope of the anomaly's impact. Ultimately, the scope of the anomaly's impact is represented by a device-component hierarchical structure, clearly listing all devices potentially affected by the anomaly and the relevant operating components within each device.

[0086] Step S148: Based on the rate characteristics of abnormal evolution, the size of the scope of abnormal correlation and the degree of abnormality of overall operational coordination, determine the emergency level of coordinated control, and based on the emergency level of coordinated control and the scope of abnormal correlation and impact, determine the number of operating components and equipment that need to participate in control, and generate a list of control objects.

[0087] The emergency level of coordinated control is divided into three levels: high, medium, and low. Among the rate characteristics of abnormal evolution, the larger the absolute value of the slope of change and the positive acceleration of change (i.e., the abnormal trend is accelerating), the higher the emergency level; the larger the scope of the abnormal correlation and the more equipment and operating components involved, the higher the emergency level; the degree of abnormality of the overall operational coordination is measured by the proportion of abnormal coordination periods and the degree of deviation between the abnormal coordination value and the normal range. The higher the proportion and the greater the deviation, the higher the emergency level.

[0088] For example, the absolute value of the slope of the abnormal evolution of the rotating component is large and the acceleration of the change is positive. The scope of the abnormal impact includes 3 pieces of equipment and 8 operating components. The proportion of the abnormal period of the overall operation coordination degree is 0.17, and the coordination degree value of some periods is far below the lower limit of the normal range. Taking these factors into account, the emergency level of coordinated control is determined to be high.

[0089] Based on the high level of emergency and the scope of impact of the anomaly, the operational components and equipment requiring intervention should include the anomaly source component (rotating component), directly related components (ray generation component, data acquisition component), indirectly related equipment (image storage server, central monitoring system), and their related critical operational components. For example, the hard disk storage component of the image storage server and the data processing component of the central monitoring system both need to be involved in intervention. List the identifiers of the aforementioned operational components and equipment to generate a list of intervention targets.

[0090] Step S149: Based on the emergency level, the list of control objects, and the rate characteristics of abnormal evolution, generate control targets, which include parameter values ​​that need to be adjusted to the normal range and synergy values ​​that need to be restored.

[0091] For situations classified as high-level emergency, the control objective should be set to adjust the abnormal parameters to the normal range within a short period and restore overall operational coordination. The parameter values ​​to be adjusted to the normal range are determined based on preset normal operating parameter ranges. For example, the vibration frequency of rotating components needs to be adjusted to the range of [f_min, f_max], and the operating temperature of the radiation generating components needs to be adjusted to the range of [T_min, T_max]. For each operational component parameter requiring adjustment, a phased adjustment objective should be set based on the rate characteristics of the abnormal evolution. For example, in the first control period, the vibration frequency could be reduced from the current predicted value by a certain margin, and adjustments could continue in subsequent periods until it reaches the normal range.

[0092] The required coordination degree value to be restored is determined based on the preset normal coordination degree range, with the goal of adjusting the overall operational coordination degree prediction value to the range of [C_min, C_max]. Simultaneously, considering the coordination relationships between different devices and operating components, target parameters for the interaction signals between each device must be set. For example, the signal transmission delay between the CT scanner and the image storage server needs to be controlled within the preset normal delay range, and the interaction parameter matching deviation needs to be adjusted to zero or close to zero.

[0093] Step S1410: Integrate the emergency level, the list of control objects, and the control objectives to generate coordinated control demand information, which also includes the time window requirements for control implementation.

[0094] The coordinated control demand information is a structured data object containing the following fields: Emergency Level (value is "High"), Control Object List (containing the identifiers of all equipment and operating components in the control object list), Control Target (containing the phased adjustment target values ​​of the parameters of each operating component and the recovery target value of the overall operational coordination), and Time Window Requirement.

[0095] The time window requirement is determined based on the urgency level. For high urgency levels, the time window is a short period starting from the current moment, such as one hour to complete the control. Within the time window, multiple control periods need to be defined, each corresponding to a period in the short-term operational status evolution sequence, specifying the control tasks to be completed within each control period. This information is then integrated into the coordinated control demand information to form a complete basis for generating control instructions.

[0096] Step S150: Input the collaborative control demand information and the mutual feedback correlation feature set into the collaborative control model, generate multi-device collaborative operation control instructions, send the multi-device collaborative operation control instructions to the corresponding control modules, collect the synchronous operation feedback signal set after execution, input the synchronous operation feedback signal set into the mutual feedback correlation processing stage, update the extraction rules of the mutual feedback correlation feature set, and at the same time adjust the timing operation parameters of the operation status inference model and the instruction generation logic of the collaborative control model.

[0097] Step S151: Input the coordinated regulation demand information into the demand parsing layer of the coordinated regulation model to obtain the emergency level, the list of regulation objects, the regulation target and the time window requirements.

[0098] The demand parsing layer of the coordinated control model employs structured information extraction technology from natural language processing to parse the textual descriptions in the coordinated control demand information. For example, it extracts the "high-level" information from the "emergency level field" of the coordinated control demand information; it extracts the identification strings of each device and operating component from the "control object list field" and converts them into standardized device-component codes; it extracts the target value range and phased adjustment requirements of each operating component parameter, as well as the target value range of the overall operational coordination degree, from the "control target field"; and it extracts the total time window length and the start and end times of each control period from the "time window requirement field".

[0099] The parsed information is stored in a structured data structure, such as a dictionary, where the keys are information categories (e.g., "emergency level," "target of regulation," etc.) and the values ​​are the specific content obtained from the parsing. The requirement parsing layer also verifies the parsing results to ensure consistency between different pieces of information, such as whether the division of regulation periods matches the time window length and whether the regulated targets are within the scope of abnormal correlation impact. If inconsistencies are found, an error message will be returned and the collaborative regulation requirement information will be regenerated.

[0100] Step S152: Input the set of mutually fed-back correlation features into the correlation feature analysis layer of the collaborative control model, extract the component parameter coupling features, signal transmission features between devices, and environmental parameter cross features corresponding to the list of control objects, and generate a subset of control correlation features.

[0101] The correlation feature analysis layer filters relevant features from the mutual feedback correlation feature set based on the equipment and operating component identifiers in the list of controlled objects. For component parameter coupling features, it filters out all coupling coefficients that include operating components in the list of controlled objects; for signal transmission features between equipment, it filters out signal transmission delay features and interaction parameter matching deviation features between equipment in the list of controlled objects; for environmental parameter cross features, it filters out the magnitude and hysteresis features of the influence of environmental parameters on the parameters of operating components in the list of controlled objects.

[0102] For example, if the list of regulated objects includes the rotating component and the ray generating component of a CT scanner, as well as the hard disk storage component of an image storage server, then the coupling coefficients between the rotating component and the ray generating component, and between the rotating component and the data acquisition component (the data acquisition component is a directly associated component) are extracted from the component parameter coupling characteristics; the signal transmission delay characteristics and the interaction parameter matching deviation characteristics between the CT scanner and the image storage server are extracted from the signal conduction characteristics between devices; the influence amplitude characteristics and influence lag characteristics of temperature on the vibration frequency of the rotating component, and the influence amplitude characteristics of temperature on the operating temperature of the ray generating component are extracted from the environmental parameter cross characteristics. These screened characteristics are combined to generate a subset of regulated association characteristics.

[0103] Step S153: Input the subset of regulated association characteristics and the parsed regulation target into the parameter adjustment calculation layer of the collaborative regulation model, calculate the initial parameter adjustment amount for each regulated object, and, in combination with the coupling coefficients in the mutual feedback association feature set, calculate the indirect influence value of the initial parameter adjustment amount on the associated regulated object, and analyze whether the indirect influence value exceeds the normal parameter range.

[0104] The parameter adjustment calculation layer first calculates the initial parameter adjustment amount according to the target values of the parameters of each operating component in the regulation target and the predicted values in the current short-term operating state evolution sequence. The initial parameter adjustment amount is the difference between the target value and the predicted value. For example, if the currently predicted vibration frequency of the rotating component is f_pred and the target value is f_target (f_target is within the normal range), then the initial parameter adjustment amount is f_target - f_pred.

[0105] Then, for the initial parameter adjustment amount of each regulated object, according to the component parameter coupling characteristics in the subset of regulated association characteristics, find the associated regulated object that has a coupling relationship with this regulated object. For example, the initial parameter adjustment amount of the rotating component will affect the operating temperature of the ray generating component through the coupling relationship. According to the coupling coefficient k between the rotating component and the ray generating component, the indirect influence value is calculated as the initial parameter adjustment amount × k.

[0106] Compare the calculated indirect influence value with the normal parameter range of the associated regulated object to determine whether it exceeds the range. For example, the normal operating temperature range of the ray generating component is [T_min, T_max], its current predicted value is T_pred, and the indirect influence value is ΔT, then the adjusted predicted value is T_pred + ΔT. If T_pred + ΔT < T_min or T_pred + ΔT > T_max, it means that the indirect influence value exceeds the normal parameter range.

[0107] Step S154: If the indirect impact value exceeds the normal parameter range, adjust the initial parameter adjustment amount and recalculate the indirect impact value until the indirect impact value is within the normal parameter range, thus obtaining the final parameter adjustment amount.

[0108] When the indirect impact value exceeds the normal parameter range, the initial parameter adjustment amount needs to be corrected. The direction of adjustment is determined based on the direction in which the indirect impact value exceeds the normal range. For example, if the indirect impact value causes the predicted value of the related controlled object to be higher than the upper limit of the normal range, the initial parameter adjustment amount should be reduced; if it causes it to be lower than the lower limit of the normal range, the initial parameter adjustment amount should be increased.

[0109] The adjustment range is determined based on the degree of exceedance and the magnitude of the coupling coefficient. The greater the exceedance, the larger the adjustment range; the larger the coupling coefficient, the more sensitive the initial parameter adjustment is to the influence of the related controlled objects, and the larger the adjustment range. The adjusted initial parameter adjustment is recorded as the new initial parameter adjustment. Its indirect impact on the related controlled objects is recalculated, and it is checked whether it still exceeds the normal range. This process may require multiple iterations until the indirect impact values ​​of all related controlled objects are within the normal parameter range. The parameter adjustment at this point is the final parameter adjustment.

[0110] Step S155: Based on the urgency level in the coordinated control demand information, assign control priorities to the final parameter adjustment amount of each control object.

[0111] The urgency level in the coordinated control demand information is high. Under the high urgency level, the allocation of control priorities mainly considers the following factors: the importance of the control object in the scope of the abnormal correlation, the degree of influence of the rate characteristics of the abnormal evolution on the control object, and the contribution of the parameter adjustment of the control object to the recovery of the overall operational coordination.

[0112] For example, the rotating component is the source of the anomaly, and its parameter adjustment is most critical for controlling the evolution of the anomaly, thus it is assigned the highest priority; the X-ray generating component is closely coupled to the rotating component, and its abnormal operating temperature may directly affect the quality of CT scan images, thus it is assigned the second highest priority; although the hard disk storage component of the image storage server is an indirectly related device, it is responsible for storing scan images, and data security is of paramount importance, thus it is assigned a relatively high priority. Based on these factors, a priority level (such as level 1, level 2, level 3, etc., with level 1 being the highest) is assigned to the final parameter adjustment amount of each control object.

[0113] Step S156: Based on the time window requirements in the coordinated control demand information, allocate control execution periods for the final parameter adjustment amount of each control object.

[0114] The time window requirement includes the total control time length and the division of control periods. Based on the control priority and the complexity of parameter adjustments, a control execution period is allocated to the final parameter adjustment amount for each control object. High-priority control objects should be allocated to earlier control periods to control abnormal trends as quickly as possible; control objects with complex parameter adjustments requiring longer execution times (such as those involving physical adjustments of mechanical components) should be allocated longer control execution periods or multiple consecutive control periods.

[0115] For example, the total time window is 1 hour, divided into 12 control periods of 5 minutes each. The control of the rotating component has the highest priority and is assigned to the first and second control periods; the X-ray generating component is assigned to the second and third control periods; the hard disk storage component of the image storage server is assigned to the third and fourth control periods, and so on. The control execution period for each control object has clearly defined start and end time numbering to ensure that all control tasks are completed within the time window.

[0116] Step S157: Generate the control execution sequence according to the control priority and control execution period, and integrate the identifier, final parameter adjustment amount, control execution period and control execution sequence of each control object to generate a single object control instruction.

[0117] The generation of the control execution order is first arranged according to the order of the control execution time periods, and the control objects within the same time period are arranged from high to low control priority. For example, the first time period only has the control task of the rotating component, and the execution order is 1; the second time period has the control tasks of the rotating component (incomplete) and the ray generating component, with the rotating component having a higher priority than the ray generating component, and the execution orders are 2 and 3 respectively; the third time period has the control tasks of the ray generating component (incomplete) and the hard disk storage component of the image storage server, and the execution orders are 4 and 5 respectively, and so on.

[0118] A single-object control command is a specific operational command for each controlled object. It includes the device-component identifier of the controlled object, the final parameter adjustment amount (such as vibration frequency adjustment amount, temperature adjustment amount, etc.), the control execution period (start period and end period), and the execution sequence within that period. For example, the single-object control command for a rotating component might be: device-component identifier "CT-rotating component", final parameter adjustment amount "-Δf" (indicating a reduction in vibration frequency Δf), control execution period "period 1-period 2", and execution sequence "1".

[0119] Step S158: Arrange all single-object control instructions in the control execution order to generate multi-device collaborative operation control instructions, and add control batch identifier and time validity identifier to the multi-device collaborative operation control instructions. The time validity identifier is used to indicate the effective execution time period of the multi-device collaborative operation control instructions.

[0120] Multi-device collaborative operation control instructions are a sequence of instructions, with all single-object control instructions arranged sequentially according to the control execution order. The position of each single-object control instruction in the sequence is determined by its execution order. The control batch identifier is generated based on the total control time window and the urgency level of the control. For example, the control batch identifier for a high urgency level is "E-YYYYMMDD-HHMM", where "E" indicates urgency, followed by the current timestamp. The timeliness identifier is the effective execution period of the control instruction, i.e., the time window requirement in the collaborative control demand information, such as "Effective period: YYYYMMDD-HHMM to YYYYMMDD-HHMM+1 hour".

[0121] Multi-device collaborative operation control instructions also include overall control information, such as an overview of the control objectives and emergency level prompts, located at the beginning of the instruction sequence. All single-object control instructions, control batch identifiers, timeliness identifiers, and overall control information are combined into a complete instruction document. This document uses a standardized format so that each control module can correctly parse and execute it.

[0122] Step S159: Extract key control parameters from the multi-device collaborative operation control instructions and generate a control parameter summary, which includes core adjustment amounts and key execution periods.

[0123] Core adjustment quantities refer to parameter adjustment quantities that play a decisive role in the overall control effect. They are usually the final parameter adjustment quantities of the highest priority control objects, as well as the parameter adjustment quantities that have the greatest impact on the overall operational coordination. For example, the final parameter adjustment quantities of rotating components and the adjustment quantities of interaction parameters between the CT scanner and the image storage server (calculated based on the interaction parameter matching deviation characteristics) are selected as core adjustment quantities.

[0124] Key execution periods refer to the control periods corresponding to critical time nodes in the control process, such as the first period (the start of control), the period containing multiple high-priority control objects, and the last period before the control ends. The core adjustment quantities and key execution periods are extracted and presented in a concise tabular format (but in list form in the text description) to generate a summary of control parameters, allowing operators to quickly understand the core content of the control instructions.

[0125] Step S1510: Send multi-device collaborative operation control instructions to the corresponding control module and collect the set of synchronous operation feedback signals after execution.

[0126] Multi-device collaborative operation control commands are sent to the control modules of the respective devices via the control network within the control facility. Each control module identifies the operating component requiring control based on the device-component identifier in the command and executes the control operation according to the final parameter adjustment amount, control execution period, and execution sequence. For example, after receiving a control command for a rotating component, the control module of a CT scanner adjusts the drive motor current or control signal of the rotating component within the specified control execution period to reduce the vibration frequency.

[0127] After the control command is executed, the parameters of each operating component change, and the interaction signals between equipment and the comprehensive influence signals of the shelter environment may also change accordingly. These changed signals are re-acquired using the previously deployed sensor network and signal acquisition modules, and after time synchronization processing, a set of synchronized operation feedback signals is formed. The structure of this set is the same as the real-time synchronized operation signal set obtained in step S110, including the real-time parameter signals of each operating component, the interaction signals between equipment, and the comprehensive influence signals of the shelter environment; however, the acquisition period is the time period after the control command is executed.

[0128] Step S1511: Input the synchronous operation feedback signal set into the mutual feedback correlation processing stage, update the extraction rules of the mutual feedback correlation feature set, and adjust the timing operation parameters of the operation state inference model and the instruction generation logic of the collaborative control model.

[0129] Step S1512: Input the synchronous running feedback signal set into the mutual feedback correlation processing stage to update the extraction rules of the mutual feedback correlation feature set.

[0130] For example, step S15121: Analyze the set of synchronous operation feedback signals, extract the deviation data between the predicted value of the short-term operation state evolution sequence of the operation state inference model and the actual operation state value in the set of synchronous operation feedback signals, and generate a set of prediction deviation features. The set of prediction deviation features includes the prediction deviation values ​​of the parameters of each operating component and the prediction deviation value of the overall operation coordination.

[0131] The synchronous operation feedback signal set contains the actual parameter values ​​of each operating component and the actual operational coordination value of the equipment after the execution of the control command. These actual values ​​are compared with the predicted values ​​for the corresponding time periods in the short-term operational state evolution sequence generated by the operational state simulation model to calculate the deviation data. The predicted deviation value for each operating component's parameter is the actual parameter value minus the predicted parameter value; the predicted deviation value for the overall operational coordination is the actual coordination value minus the predicted coordination value. The predicted deviation values ​​for all operating components' parameters and the overall operational coordination are arranged in chronological order to form a predicted deviation feature set.

[0132] Step S15122: Perform feature decomposition on the prediction deviation feature set to extract the fluctuation cycle features, change rate features, and time series change trend features of the overall operational coordination prediction deviation values ​​of the component parameters.

[0133] For each operating component parameter prediction deviation value sequence, the fluctuation period characteristics are extracted, and the main fluctuation period of the deviation value is determined by using an autocorrelation analysis method similar to that in step S121; the rate of change characteristics are obtained by calculating the first derivative of the deviation value sequence, reflecting how fast the deviation value changes with time.

[0134] The time-series trend features of the predicted deviation value sequence of overall operational coordination are extracted. A linear fitting method within a sliding window is used to determine the trend (rising, falling, or stable) of the deviation value within each window, and the frequency and duration of different trends are statistically analyzed.

[0135] Step S15123: Perform correlation analysis between the fluctuation period characteristics and change rate characteristics of the predicted deviation values ​​of component parameters and the real-time parameter signals in the synchronous operation feedback signal set, and calculate the deviation sensitivity factor of each operating component parameter. The deviation sensitivity factor is used to quantify the contribution of parameter fluctuations to the prediction deviation.

[0136] The fluctuation period characteristics of the predicted deviation values ​​of component parameters are compared with the fluctuation period characteristics of the real-time parameter signals of the corresponding operating components in the synchronous operation feedback signal set. The similarity between the two is calculated. The higher the similarity, the stronger the correlation between the fluctuation of the operating component parameters and the fluctuation of the predicted deviation. At the same time, correlation analysis is performed on the change rate characteristics of the predicted deviation values ​​of component parameters and the change rate characteristics of the real-time parameter signals, and the correlation coefficient is calculated.

[0137] The deviation sensitivity factor is obtained by comprehensively considering the similarity of fluctuation cycles and the correlation coefficient of change rates, and then summing the two in a weighted manner. For example, deviation sensitivity factor = α × similarity of fluctuation cycles + (1-α) × correlation coefficient of change rates, where α is a weighting coefficient determined based on experience or cross-validation. The larger the value of the deviation sensitivity factor, the greater the contribution of the fluctuation of the operating component parameter to the prediction deviation.

[0138] Step S15124: Perform correlation analysis between the temporal variation trend characteristics of the overall operational coordination degree prediction deviation value and the equipment interaction signals and the comprehensive influence signals of the container environment in the synchronous operation feedback signal set, and extract the coordination deviation transmission path characteristics. The coordination deviation transmission path characteristics include the delay characteristics of the equipment interaction signals and the indirect influence characteristics of environmental parameters on the coordination degree.

[0139] For each trend segment (such as the upward trend segment above) in the time-series variation trend characteristics of the overall operational coordination prediction deviation value, analyze the changes in the inter-equipment interaction signals corresponding to the trend segment, and extract the change characteristics of signal transmission delay (such as delay increase, delay decrease, or delay fluctuation); at the same time, analyze the changes in the comprehensive influence signal of the cabin environment in the trend segment, and extract the change characteristics of the influence amplitude and lag of environmental parameters on the operating component parameters of each equipment.

[0140] The characteristics of the coordination deviation transmission path are represented by a directed graph. Nodes in the graph represent equipment or environmental parameters, edges represent signal transmission or influence relationships, and attributes on the edges represent the delay characteristics of signals exchanged between equipment or the indirect influence characteristics of environmental parameters on coordination (such as the rate of change of influence amplitude or the rate of change of influence hysteresis). For example, the edge from the CT scanner to the image storage server is marked with the characteristic of increased signal transmission delay, and the edge from the temperature parameter to the overall operational coordination is marked with the characteristic of increased influence amplitude.

[0141] Step S15125: Based on the deviation sensitivity factor and the characteristics of the collaborative deviation transmission path, construct a feature correlation strength evaluation matrix, which is used to describe the degree of correlation between different types of signal features.

[0142] The rows and columns of the feature correlation strength evaluation matrix represent different types of signal features. For example, the rows represent the signal features of operating component parameters (such as vibration frequency, operating temperature, etc.), the signal features of interaction between equipment (such as signal transmission delay, interaction parameter matching deviation, etc.), and the signal features of environmental parameters (such as temperature, humidity, etc.). The columns represent the prediction deviation features (such as component parameter prediction deviation, overall operation coordination prediction deviation, etc.).

[0143] The element values ​​in the matrix represent the correlation strength between the signal characteristics of the corresponding row and the prediction deviation characteristics of the corresponding column. For the correlation strength between the signal characteristics of operating component parameters and the prediction deviation characteristics of component parameters, the deviation sensitivity factor of that operating component is used. For the correlation strength between the signal characteristics of inter-equipment interaction and the prediction deviation characteristics of overall operational coordination, it is determined based on the degree of change in the delay characteristics of inter-equipment interaction signals in the coordination deviation transmission path characteristics; the greater the degree of change, the higher the correlation strength. For the correlation strength between the signal characteristics of environmental parameters and the prediction deviation characteristics of overall operational coordination, it is determined based on the degree of change in the indirect influence characteristics of environmental parameters on coordination in the coordination deviation transmission path characteristics.

[0144] Step S15126: Based on the feature association strength evaluation matrix, adjust the extraction rules of the mutual feedback association feature set, including: for the extraction rules of component parameter coupling features, increase the calculation dimension of the coupling coefficient between operating component parameters whose deviation sensitivity factors exceed the preset threshold; for the extraction rules of signal transmission features between devices, strengthen the weight allocation of the interaction signal delay features between devices in the collaborative deviation transmission path features; for the extraction rules of environmental parameter cross features, introduce a dynamic evaluation mechanism for the indirect influence of environmental parameters on the collaborative degree features.

[0145] A threshold for a deviation sensitivity factor is preset. When the deviation sensitivity factor of a certain operating component parameter exceeds this threshold, it indicates that the parameter is closely related to the prediction deviation. Therefore, the extraction rules for component parameter coupling features need to include the calculation dimension of the coupling coefficient between this parameter and other related operating component parameters. For example, if the deviation sensitivity factor of a rotating component exceeds the threshold, then when extracting component parameter coupling features, not only the coupling coefficient between the rotating component and existing related components should be calculated, but also the coupling coefficient between the rotating component and previously unconsidered cooling components.

[0146] For the extraction rules of signal transmission features between devices, the weight of the delay feature in feature extraction is adjusted based on the correlation strength between the signal delay feature between devices and the prediction deviation feature of overall operational synergy in the feature correlation strength evaluation matrix. The higher the correlation strength, the greater the weight, and the more likely the feature will be selected in the subsequent feature importance evaluation. For example, if the correlation strength of the signal transmission delay feature between a CT scanner and an image storage server is high, then the weight of this delay feature will be increased when extracting signal transmission features between devices.

[0147] Previously, the extraction rules for cross-features of environmental parameters may have only considered the direct impact of environmental parameters on the parameters of operating components. However, by introducing a dynamic evaluation mechanism for the indirect impact of environmental parameters on coordination, the extraction weights of cross-features between environmental parameters and operating component parameters are dynamically adjusted based on the indirect impact of environmental parameters on coordination in the coordination deviation transmission path characteristics. For example, when the indirect impact of temperature on overall operational coordination increases, the weight of the cross-feature between temperature and the vibration frequency of rotating components is increased.

[0148] Step S15127: Associate and store the extraction rules of the adjusted mutual feedback correlation feature set with the collection time period information of the synchronous operation feedback signal set to form a feature extraction rule library that is dynamically updated over time.

[0149] The extraction rules for the adjusted mutual feedback correlation feature set are stored in a versioned format. Each version of the rule is associated with the collection period information (such as the start and end times of collection) of the synchronous operation feedback signal set. The feature extraction rule base adopts a database table structure, with fields including rule version number, collection period start time, collection period end time, component parameter coupling feature extraction rules, inter-device signal transmission feature extraction rules, and environmental parameter cross-feature extraction rules. Each time the extraction rules are updated, a new record is added to the rule base, retaining historical rule versions for backtracking and comparative analysis when needed.

[0150] Step S15128: Use the historical synchronous running data set to verify the extraction rules of the updated mutual feedback correlation feature set, calculate the prediction deviation change rate before and after the feature extraction rule adjustment, and if the prediction deviation change rate does not reach the preset optimization threshold, reconstruct the feature correlation strength evaluation matrix and adjust the extraction rules until the preset optimization threshold is met.

[0151] The historical synchronous operation dataset contains synchronous operation data from a past period when no adjustments were made or when old extraction rules were used. The updated extraction rules are used to extract a set of mutually correlated features from the historical synchronous operation dataset. These extracted features are then input into the operational status projection model to obtain predicted values ​​for the historical data. The deviation between these predicted values ​​and the actual values ​​in the historical data is calculated; this deviation represents the prediction bias of the adjusted rules.

[0152] Simultaneously, feature extraction and prediction are performed on the same historical synchronous dataset using the old extraction rules, yielding the prediction bias of the rules before adjustment. The prediction bias change rate is (prediction bias before adjustment - prediction bias after adjustment) / prediction bias before adjustment. A preset optimization threshold, such as 0.1, is set, indicating that the prediction bias needs to be reduced by at least 10%. If the calculated prediction bias change rate is greater than or equal to the optimization threshold, the adjusted extraction rules are considered effective; otherwise, the process returns to step S15125 to reconstruct the feature association strength evaluation matrix, and the extraction rules are readjusted according to step S15126. The verification is then repeated until the prediction bias change rate reaches the optimization threshold.

[0153] Step S1513: Adjust the timing operation parameters of the running state deduction model and the instruction generation logic of the collaborative control model.

[0154] Step S15131: Analyze the set of synchronous operation feedback signals, extract the deviation data between the predicted value of the short-term operation state evolution sequence of the operation state inference model and the actual operation state value in the set of synchronous operation feedback signals, and generate a set of prediction deviation features. The set of prediction deviation features includes the prediction deviation values ​​of the parameters of each operating component and the prediction deviation value of the overall operation coordination.

[0155] This step is the same as step S15121, which involves parsing the set of synchronous operation feedback signals and generating a set of prediction deviation features, which are used to adjust the timing operation parameters of the operation state inference model in the future.

[0156] Step S15132: Based on the prediction deviation feature set, calculate the time series operation parameter adjustment demand index of the operating state inference model. The time series operation parameter adjustment demand index is determined comprehensively based on the magnitude of the prediction deviation value and the duration of the deviation.

[0157] The timing parameter adjustment demand index measures the urgency of adjusting the timing parameters of the operational state simulation model. For each operational component parameter prediction deviation, the average of its absolute values ​​is calculated as the deviation magnitude index; the number of consecutive periods where the deviation exceeds the preset allowable deviation range is counted as the deviation duration index. The timing parameter adjustment demand index = β × deviation magnitude index + (1-β) × deviation duration index, where β is a weighting coefficient. A larger deviation magnitude index and a larger deviation duration index indicate a higher timing parameter adjustment demand index, signifying a greater urgency for adjusting that parameter.

[0158] Step S15133: Adjust the demand index of the timing operation parameters according to the timing operation parameters, and determine the priority order of the timing operation parameters in the operation state simulation model. The higher the adjustment demand index of the timing operation parameters, the higher the corresponding adjustment priority.

[0159] The time-series operational parameters of the operational state extrapolation model (such as the convolution kernel parameters of the first time-series operational layer, the recurrent unit parameters of the second time-series operational layer, and the perceptron parameters of the feature abstraction layer) are correlated with the prediction deviation values ​​in the prediction deviation feature set. For example, the convolution kernel parameters of the first time-series operational layer mainly affect the extraction of local time-series feature vectors and are correlated with the prediction deviation values ​​of operational component parameters that have strong local time-series correlations. Based on the time-series operational parameter adjustment demand index of the prediction deviation feature set associated with each time-series operational parameter, the time-series operational parameters are sorted, with parameters having higher adjustment demand indices having higher adjustment priority.

[0160] Step S15134: According to the adjustment priority order, adjust the convolution kernel parameters of the first time-series operation layer, the recurrent unit parameters of the second time-series operation layer, and the perceptron parameters of the feature abstraction layer of the running state inference model in sequence. The adjustment range is dynamically calculated based on the deviation values ​​of the corresponding running component parameters in the prediction deviation feature set.

[0161] For the convolution kernel parameters of the first temporal computation layer, the adjustment magnitude is proportional to the magnitude of the prediction deviation value of the corresponding runtime component parameters. For example, if the prediction deviation value associated with a certain convolution kernel parameter is large, then the adjustment magnitude of that convolution kernel parameter will also be large. The adjustment direction is determined according to the sign of the deviation: if the predicted value is consistently higher than the actual value, the value of the corresponding weight in the convolution kernel parameter is decreased; if the predicted value is consistently lower than the actual value, the value of the corresponding weight is increased.

[0162] The parameters of the loop units in the second temporal operation layer include the weights and biases of the input gate, forget gate, and output gate. When adjusting these parameters, the parameters of the forget gate are adjusted according to the deviation duration index to enhance or weaken the memory of historical information in the time region where the deviation continues to occur. The parameters of the input gate and output gate are adjusted according to the deviation magnitude index to adjust the contribution ratio of the current input information and the hidden state.

[0163] The adjustment of the perceptron parameters in the feature abstraction layer is similar to that in the first time-series computation layer. The adjustment magnitude is based on the magnitude of the corresponding prediction deviation value, and the adjustment direction is determined according to the positive or negative sign of the deviation, so that the high-order operating state feature vector output by the perceptron is closer to the real operating state features.

[0164] Step S15135: After adjusting the time series operation parameters of the running state inference model, use the historical synchronous running data set to verify the prediction accuracy of the adjusted running state inference model. If the prediction accuracy does not reach the preset accuracy threshold, recalculate the time series operation parameter adjustment requirement index and iteratively adjust the time series operation parameters until the prediction accuracy meets the standard.

[0165] Using the same historical synchronous running data set as in step S15128, input it into the running state inference model after adjusting the time series operation parameters to obtain the predicted value. Calculate the mean absolute error or root mean square error between the predicted value and the historical actual value as the prediction accuracy index. Preset an accuracy threshold, for example, the mean absolute error is less than a certain value. If the adjusted prediction accuracy reaches this threshold, the adjustment stops; otherwise, return to step S15132 to recalculate the time series operation parameter adjustment demand index, determine the new adjustment priority order according to step S15133, perform parameter adjustment in step S15134, and verify again until the prediction accuracy meets the standard.

[0166] Step S15136: Analyze the set of synchronous operation feedback signals, extract the execution effect data of the multi-device collaborative operation control instructions generated by the collaborative control model, and generate a control effect feature set. The control effect feature set includes the matching degree between the change value of the operating component parameters after control and the expected control target, and the matching degree between the change value of the overall operation coordination degree after control and the expected coordination degree target.

[0167] The change in operating component parameters after regulation is the difference between the actual parameter values ​​in the synchronous operation feedback signal set and the predicted values ​​in the short-term operating state evolution sequence before regulation; the expected regulation target is the difference between the target value in the regulation target and the predicted value before regulation. The matching degree is the ratio of the actual change value to the expected change value; if the ratio is close to 1, the matching degree is high. The matching degree between the overall operating coordination change value and the expected coordination target is calculated in a similar way.

[0168] Step S15137: Based on the set of regulatory effect features, calculate the instruction generation logic adjustment demand index of the collaborative regulation model. The instruction generation logic adjustment demand index is determined comprehensively based on the degree of matching and the stability of matching.

[0169] The matching degree is measured by the average matching degree value; the closer the average value is to 1, the higher the matching degree. Matching stability is measured by the variance of the matching degree value; the smaller the variance, the higher the matching stability. The instruction generation logic adjustment demand index = γ × (1 - average matching degree) + (1 - γ) × matching degree variance, where γ is a weighting coefficient. The lower the average matching degree and the larger the variance, the higher the instruction generation logic adjustment demand index, indicating a higher urgency for adjusting the instruction generation logic.

[0170] Step S15138: Adjust the demand index according to the instruction generation logic, determine the adjustment priority order of the instruction generation logic in the coordinated control model, and adjust the parameter adjustment algorithm of the parameter adjustment calculation layer, the priority allocation rule of the priority allocation layer, and the time period allocation strategy of the time period allocation layer in sequence according to the adjustment priority order.

[0171] The various instruction generation logic modules (parameter adjustment calculation layer, priority allocation layer, and time period allocation layer) of the collaborative control model are associated with various matching indices in the control effect feature set. For example, the parameter adjustment algorithm of the parameter adjustment calculation layer mainly affects the matching degree of the parameter changes of operating components, and the priority allocation rule of the priority allocation layer mainly affects the overall matching stability of multi-device collaborative control. Based on the instruction generation logic adjustment demand index of the control effect feature set associated with each module, the instruction generation logic modules are ranked, and the module with the higher the adjustment demand index has the higher adjustment priority.

[0172] For the parameter adjustment algorithm of the parameter adjustment calculation layer, if the matching degree of the parameter change value of the running component is low, it may be that there is a problem with the calculation method of the initial parameter adjustment amount or the calculation method of the indirect influence value. The adjustment algorithm should more accurately consider the dynamic change of the coupling coefficient or the normal parameter range boundary of the associated control object.

[0173] When adjusting the priority allocation rules of the priority allocation layer, if the matching stability is poor, it may be because the priority division is not refined enough or the mutual influence between the controlled objects is not considered. Adjust the rules to add more priority division dimensions or dynamically adjust the priority according to the correlation strength of the controlled objects.

[0174] When adjusting the time allocation strategy of the time allocation layer, if the control execution time does not match the actual needs, it may be because the time period length or the time period division method is unreasonable. The adjustment strategy should dynamically allocate time periods according to the difficulty of adjusting the parameters of the control object and the required time.

[0175] Step S15139: After adjusting the instruction generation logic of the coordinated control model, use the historical coordinated control data set to verify the control effect of the adjusted coordinated control model. If the control effect does not reach the preset effect threshold, recalculate the instruction generation logic adjustment demand index and iteratively adjust the instruction generation logic until the control effect meets the standard.

[0176] The historical coordinated control dataset contains information on coordinated control needs, a set of mutually related features, and corresponding control effect data from the past when using the old command generation logic. The adjusted command generation logic processes the coordinated control needs and mutually related feature sets in the historical coordinated control dataset to generate new multi-device coordinated operation control commands. These commands are then compared with the actual control effect data in the historical coordinated control dataset to calculate control effect indicators (such as the average and variance of the matching degree). A preset effect threshold is set, for example, an average matching degree greater than 0.9 and a variance less than 0.05. If the adjusted control effect indicators reach this threshold, the adjustment stops; otherwise, the process returns to step S15137 to recalculate the command generation logic adjustment demand index, readjusts the command generation logic according to step S15138, and verifies again until the control effect meets the target.

[0177] Step S151310: Integrate the timing operation parameter adjustment process of the operational status deduction model with the instruction generation logic adjustment process of the collaborative control model, and establish the correlation mapping relationship between parameter adjustment and logic adjustment.

[0178] Record the adjustments made to the time-series operational parameters of the operational state deduction model and the adjustments made to the instruction generation logic modules of the collaborative control model, as well as the changes in the prediction deviation feature set and control effect feature set before and after the adjustments. Through correlation analysis, identify the relationship between the adjustment of time-series operational parameters and the adjustment of instruction generation logic. For example, the adjustment of the convolution kernel parameters of the first time-series operational layer may affect the adjustment direction of the parameter adjustment computation layer algorithm.

[0179] Step S151311: Associate and store the timing operation parameters of the adjusted running state deduction model and the instruction generation logic of the collaborative control model with the adjustment timestamp.

[0180] The adjusted time-series computation parameters and instruction generation logic are stored in a versioned format, with each version associated with a timestamp of the adjustment operation. These are stored in a model parameter library, which includes fields such as model version number, adjustment timestamp, time-series computation parameters of the running state model, collaborative control model instruction generation logic, corresponding summary of prediction bias feature sets, and summary of control effect feature sets, to track the model's evolution and trace back to historical versions.

[0181] Furthermore, the construction process of the operational state simulation model includes: Step S211: Collect a sample synchronous operation data set, which includes operating component parameter signals, inter-device interaction signals, and comprehensive environmental impact signals under different operating scenarios.

[0182] In this embodiment, the collection of the sample synchronous operation data set covers various operating scenarios of the smart medical mobile cabin equipment, including normal operation scenarios, abnormal operation scenarios of varying degrees, and simulated fault scenarios. In normal operation scenarios, operational data of each device within the mobile cabin is collected under normal workload and standard environmental conditions. Abnormal operation scenarios of varying degrees include situations where some operating component parameters are close to the normal range boundary, or where there is significant delay or parameter matching deviation in the interaction between some devices. Simulated fault scenarios are generated by artificially setting fault conditions, such as simulating a situation where the cooling system efficiency of a certain device decreases, leading to an abnormal increase in the operating temperature of related components. The time span of sample data collection covers different time periods, including the initial stage of equipment startup, peak continuous operation period, and low-load operation period. All sample data is acquired through the same sensor network and data aggregation gateway as the real-time data acquisition, and undergoes strict time synchronization and data verification to remove invalid and abnormal data points.

[0183] Step S212: Divide the sample synchronous running data set into time sequences, and divide each continuous sample data segment into sample input segments and corresponding sample output segments according to a fixed time length. The sample output segments are the running data of the subsequent time periods of the sample input segments.

[0184] A fixed time period is set, and each continuous data segment in the synchronous sample dataset is divided into sample input segments and sample output segments. For example, a continuous sample data segment can be divided into multiple sample pairs, where the sample input segment of one sample pair is the operational data for a certain time period, and the corresponding sample output segment is the operational data for subsequent time periods. A sliding window method is used for partitioning, with the window sliding step size set according to the sample size. Both the sample input and sample output segments contain complete data including operational component parameter signals, inter-device interaction signals, and comprehensive environmental influence signals, and their timestamps are continuous. After partitioning, each sample pair is labeled, and its corresponding operational scenario category is recorded.

[0185] Step S213: Perform mutual feedback correlation processing on the sample input segment and extract the sample mutual feedback correlation feature set, which includes sample component parameter coupling features, signal transmission features between sample devices, and sample environmental parameter cross features.

[0186] For each sample input segment, the same feedback correlation processing procedure as in step S120 is performed. First, the parameter signals of each operating component in the sample input segment are separated, and time-domain variation features are extracted; the interaction signals between devices are separated, and conduction features are extracted; the comprehensive environmental influence signals are separated, and differential effect features are extracted. Then, a first-level correlation operation is performed to generate the first correlation feature, and a second-level correlation operation is performed to generate the second correlation feature. The first and second correlation features are then processed to unify dimensions, and dynamic weights are assigned and integrated based on the feature importance evaluation results to generate a sample feedback correlation feature set. Throughout this process, the parameter settings for all feature extraction and correlation operations remain consistent with those used in real-time processing.

[0187] Step S214: Use the sample feedback correlation feature set as the input sample of the operation state inference model, and use the operation component parameter values ​​and the overall operation coordination value in the sample output segment as the output sample of the operation state inference model.

[0188] The sample feedback correlation feature set is directly used as the input sample for the operational state inference model, and its data format is the same as the feedback correlation feature set input to the model in real-time processing. The parameter values ​​of the operating components in the sample output segment need to be extracted to correspond to the parameters of the operating components in the input sample. The overall operational coordination degree is calculated in the same way as in step S134, by calculating the coordination coefficient of different operating component parameter values ​​at different time periods in the sample output segment, and this is used as part of the output sample. Input samples and output samples correspond one-to-one, forming the sample pairs required for model training.

[0189] Step S215: Divide the input samples and output samples of the running state inference model into training sample set, verification sample set and test sample set according to a preset ratio.

[0190] The preset ratio is set according to the model training requirements. The partitioning process uses stratified sampling to ensure that the proportion of samples from different running scenario categories in each sample set is consistent with the proportion in the original sample synchronous running dataset. After partitioning, each sample set is stored independently and the data format is checked to ensure that the dimensions of the input and output samples match.

[0191] Step S216: Construct the initial network structure of the running state inference model. The initial network structure includes a feature preprocessing layer, a first time-series operation layer, a second time-series operation layer, a feature abstraction layer, and a prediction output layer.

[0192] The feature preprocessing layer performs temporal alignment and standardization on the input sample feedback correlation feature set. The first temporal computation layer adopts a one-dimensional convolutional neural network structure, setting multiple convolutional kernels to capture local temporal correlation information at different time scales. Each convolutional layer is followed by a batch normalization layer and an activation function. The second temporal computation layer adopts a recurrent unit structure, setting multiple hidden layer neurons to capture long-term temporal dependencies, and employs dropout technology to prevent overfitting. The feature abstraction layer is a multilayer perceptron structure containing multiple fully connected layers, mapping long-term temporal feature vectors to a higher-order feature space. Each layer is followed by a batch normalization layer and an activation function. The prediction output layer contains two parallel fully connected sub-layers: one sub-layer outputs the predicted values ​​of the running component parameters at each time period, and the other sub-layer outputs the predicted value of the overall running coordination degree, both using linear activation functions.

[0193] Step S217: Set the initial values ​​of the parameters of each layer of the initial network structure, including the initial parameters of the convolution kernels of the first temporal operation layer, the initial parameters of the recurrent units of the second temporal operation layer, and the initial parameters of the perceptron of the feature abstraction layer.

[0194] The initial parameters of the convolutional kernels in the first temporal computation layer are initialized using a method suitable for the ReLU activation function, and the bias parameters of the convolutional kernels are initialized to zero. In the initial parameters of the recurrent units in the second temporal computation layer, the weight matrices are initialized using a method that helps maintain gradient stability during the loop, the bias parameters are initialized to zero, and the cell states are initialized as all-zero vectors. In the initial parameters of the perceptron in the feature abstraction layer, the weight matrices are initialized using a method suitable for the activation function, and the bias parameters are initialized to zero.

[0195] Step S218: Input the training sample set into the initial network structure, calculate the error between the predicted output of the running state inference model and the output samples of the training sample set through the backpropagation algorithm, and adjust the parameters of each layer.

[0196] Mean squared error is used as the loss function to calculate the error between the model's predicted output and the output samples of the training sample set. The weighted sum of the component parameter errors and coordination errors is used as the total loss function. An appropriate optimizer is selected, and the initial learning rate and learning rate decay strategy are set. During training, the training sample set is input into the model in batches. For each batch of data, forward propagation is used to calculate the predicted output, and then the gradient of each layer's parameters is calculated using the backpropagation algorithm. The optimizer is then used to update the parameters.

[0197] Step S219: During the training process, after each preset training round, the prediction accuracy of the running state inference model is evaluated using the validation sample set, and the change curve of the validation accuracy is recorded. If the validation accuracy does not improve for a consecutive preset round, the training is stopped by adopting an early stop strategy to avoid overfitting of the running state inference model.

[0198] The preset number of training epochs is set according to the model complexity. After each preset number of training epochs, the model is evaluated using a validation sample set. Validation accuracy is measured by calculating the mean squared error of the model's predictions on the validation sample set. The validation accuracy of each evaluation is recorded and a change curve is plotted. A preset number of consecutive epochs without improvement is set. When the validation accuracy of consecutive preset epochs does not exceed the previous highest validation accuracy, training stops, and the model parameters at the point of highest validation accuracy are saved. A model checkpointing mechanism is used during training, saving the current model parameters after each training epoch.

[0199] Step S220: Use the test sample set to perform performance testing on the trained running state inference model, calculate the average prediction error of the running state inference model on the test sample set, calculate the prediction accuracy, and if the average prediction error exceeds the preset test error threshold or the prediction accuracy is lower than the preset test accuracy threshold, adjust the network structure of the running state inference model, including increasing the number of convolution kernels in the first temporal operation layer and adjusting the number of recurrent units in the second temporal operation layer, and retrain and test.

[0200] The test sample set is used to evaluate the model's generalization ability. When calculating the average prediction error, the average absolute error of the predicted values ​​for each operating component parameter and the overall operational synergy is calculated separately, and then the average is taken. The prediction accuracy is calculated as follows: for each time period, if the predicted value is within the preset error range of the actual value, the prediction is considered accurate. The prediction accuracy is the proportion of the number of time periods with accurate predictions to the total number of test time periods. The preset test error threshold and preset test accuracy threshold are set according to application requirements. If the model performance does not meet the standards, the network structure is adjusted, such as increasing the number of convolutional kernels in the first temporal operation layer or adjusting the number of recurrent units in the second temporal operation layer, and retraining and testing are performed until the model performance meets the threshold requirements.

[0201] Step S221: Optimize the computational efficiency of the tested running state inference model so that the running state inference model can generate a short-term running state evolution sequence within a preset computation time. Then, store the optimized running state inference model together with the training parameters, verification results and test results in the running state inference model construction process.

[0202] Computational efficiency optimization includes model pruning and quantization. Model pruning reduces the number of model parameters and computational cost by removing connections or neurons with small absolute weights. Quantization converts model parameters from higher-order floating-point numbers to lower-order floating-point numbers or integers, reducing computational resource consumption. Preset computation times are set based on real-time monitoring requirements. The optimized model is run in a test environment, and its computation time is measured. If the preset requirements are not met, the pruning threshold or quantization bit depth is further adjusted. The optimized model is stored as a binary file, and hyperparameters, validation results, and test results from the training process are associated with the model file and stored in a model repository to establish a model version management record.

[0203] The process of constructing a coordinated regulation model includes: Step S311: Collect a sample collaborative regulation data set, which includes sample collaborative regulation demand information, sample mutual feedback correlation feature set, sample multi-device collaborative operation regulation instructions and corresponding sample synchronous operation feedback signal set.

[0204] The collection of sample collaborative control data focuses on the collaborative control process of smart medical mobile cabin equipment, covering multiple control scenarios, including control needs at different urgency levels, control objects with different abnormal correlation impact ranges, and scenarios with different control objectives. Sample collaborative control requirement information includes different urgency levels, lists of control objects, control objectives, and time window requirements. The sample mutual feedback correlation feature set consists of mutual feedback correlation features under the corresponding control scenario. The sample multi-device collaborative operation control instructions are control instructions generated based on the sample collaborative control requirement information. The sample synchronous operation feedback signal set consists of feedback data after executing the sample control instructions. All sample data are collected through standardized processes to ensure data integrity and accuracy.

[0205] Step S312: Organize the sample collaborative regulation data set, use the sample collaborative regulation demand information and the corresponding sample mutual feedback correlation feature set as input samples for the collaborative regulation model, and use the sample multi-device collaborative operation regulation command as output samples for the collaborative regulation model.

[0206] The various data types in the sample coordinated regulation dataset are correlated and organized to ensure a one-to-one correspondence between sample coordinated regulation demand information, sample mutual feedback correlation feature sets, sample multi-device coordinated operation regulation commands, and sample synchronous operation feedback signal sets. The sample coordinated regulation demand information and the corresponding sample mutual feedback correlation feature sets are combined to form the input samples for the coordinated regulation model, and the sample multi-device coordinated operation regulation commands serve as the output samples. The input and output samples are cleaned and formatted for use in model training.

[0207] Step S313: Perform structured processing on the sample collaborative regulation demand information in the input samples of the collaborative regulation model, extract the sample emergency level, sample regulation object list, sample regulation target and sample time window requirements, and generate structured input features.

[0208] The sample coordinated regulation demand information may be unstructured or semi-structured data. This data is transformed into features recognizable by the model through structuring processing. This involves extracting the sample urgency level and converting it into numerical or categorical features; extracting the sample regulation object list and converting equipment and component identifiers into coded forms; extracting the sample regulation targets, including the normal range values ​​to which each parameter needs to be adjusted and the required degree of coordination to be restored; and extracting the sample time window requirements, including the total time window length and the division of each regulation period. The extracted information is then combined into structured input features.

[0209] Step S314: Perform feature screening on the sample feedback correlation feature set in the input samples of the collaborative control model, extract the sample control correlation feature subset corresponding to the sample control object list, use it as the correlation input feature, and integrate the structured input feature with the correlation input feature to form the input sample of the collaborative control model. Use the parameter adjustment amount, control priority, and control execution period in the multi-device collaborative operation control command of the sample as the key features of the output sample of the collaborative control model.

[0210] Based on the equipment and component identifiers in the sample control object list, relevant features are selected from the sample mutual feedback correlation feature set to generate a sample control correlation feature subset. The structured input features and correlation input features are integrated along the feature dimension to form the input samples for the collaborative control model. Key information such as parameter adjustment amounts, control priorities, and control execution periods are extracted from the multi-device collaborative operation control instructions in the samples, serving as key features of the collaborative control model's output samples.

[0211] Step S315: Divide the complete input samples and output samples of the collaborative regulation model into a model training set, a model validation set, and a model test set according to a preset ratio.

[0212] The preset proportions are determined based on the training requirements of the collaborative regulation model, and stratified sampling is used to divide the datasets to ensure that the proportions of samples from different regulation scenario categories are consistent across all sample sets. After the division is completed, each sample set is stored and validated independently.

[0213] Step S316: Construct the initial structure of the collaborative control model, which includes a demand analysis layer, a correlation feature analysis layer, a parameter adjustment calculation layer, a priority allocation layer, and a time period allocation layer.

[0214] The demand parsing layer parses the coordinated control demand information and outputs structured information such as urgency level and a list of control targets. The correlation feature analysis layer extracts a subset of control-related features from the mutual feedback correlation feature set. The parameter adjustment calculation layer calculates the initial and final parameter adjustment amounts for the control targets. The priority allocation layer assigns control priorities to the control targets. The time period allocation layer assigns control execution time periods to the control targets. All layers are connected via data interfaces to ensure smooth information transmission.

[0215] Step S317: Set the initial parameters for each layer of the initial structure, including the initial value of the coupling coefficient of the parameter adjustment calculation layer, the initial value of the weight of the priority allocation layer, and the initial parameters of the algorithm of the time segment allocation layer.

[0216] The initial values ​​of the coupling coefficients in the parameter adjustment calculation layer are set based on historical data or domain knowledge, and can be initialized to the average coupling coefficient obtained statistically from the sample data. The initial values ​​of the weights in the priority allocation layer are set according to factors such as the importance of the regulated object, and can be initialized to equal weights or weights allocated based on experience. The initial parameters of the time period allocation layer algorithm include the time interval for dividing the time period, the initial rules for the order of regulation execution, etc., and are set according to conventional regulation strategies.

[0217] Step S318: Input the model training set into the initial structure, calculate the error between the control instructions generated by the collaborative control model and the output samples of the model training set using the gradient descent algorithm, and adjust the parameters of each layer.

[0218] Using appropriate loss functions, such as the mean squared error of parameter adjustments or the ranking loss based on adjustment priorities, the error between the control commands generated by the model and the output samples of the model training set is calculated. Gradient descent and its variants are selected as the optimizer, with initial learning rate and training batch size set. The model training set is input into the initial structure in batches, forward propagation generates control commands, and after calculating the error, backpropagation adjusts the parameters of each layer.

[0219] Step S319: During the training process, after each preset training batch is completed, the matching degree between the control instructions generated by the collaborative control model and the output samples of the validation set is evaluated using the matching degree evaluation algorithm on the model validation set. The trend of matching degree change is recorded. If the matching degree does not improve for consecutive preset batches, the weight of the loss function of the collaborative control model is adjusted and retraining is performed until the matching degree reaches the preset validation matching threshold.

[0220] Preset training batches are set according to the model training progress. After each preset training batch is completed, the model is evaluated using a validation set. The matching degree evaluation algorithm is used to measure the similarity between the control instructions generated by the model and the output samples of the validation set, and can be evaluated from multiple dimensions such as parameter adjustment amount, priority, and execution time. The matching degree of each evaluation is recorded and the trend of change is observed. A preset batch with no improvement is set. When the matching degree of consecutive preset batches does not improve, the weights of each part of the loss function are adjusted, such as increasing the weight of parameter adjustment error, and training is re-run until the matching degree reaches the preset validation matching threshold.

[0221] Step S320: Test the trained collaborative regulation model using the model test set, calculate the synergy compliance rate and regulation effect compliance rate of the regulation instructions generated by the collaborative regulation model on the test set. If the synergy compliance rate is lower than the preset synergy threshold or the regulation effect compliance rate is lower than the preset effect threshold, adjust the network structure of the collaborative regulation model, including increasing parameters to adjust the computational dimension of the computation layer, optimizing the computational logic of the priority allocation layer, and retraining and testing.

[0222] The model test set is used to evaluate the generalization performance of the collaborative regulation model. The synergy achievement rate is the proportion of coordinated actions among the regulated objects in the generated regulation instructions, such as reasonable execution order and no conflicting parameter adjustments. The regulation effect achievement rate is the proportion of the regulation objectives achieved after the regulation instructions are executed. Preset synergy thresholds and preset effect thresholds are set according to actual regulation needs. If the model fails to meet the test criteria, the network structure is adjusted, such as increasing the computational dimension of the parameter adjustment layer to consider more factors, optimizing the computational logic of the priority allocation layer to improve the accuracy of priority allocation, and then retraining and testing.

[0223] Step S321: Optimize the logic of the tested collaborative control model, add an exception handling branch, so that the collaborative control model can generate reasonable control instructions when the control demand information is incomplete or the mutual feedback correlation feature set is abnormal. Then, store the optimized collaborative control model together with the training parameters, verification results and test results in the collaborative control model construction process.

[0224] Logical optimization includes adding anomaly handling mechanisms. For example, when the list of control targets in the control demand information is incomplete, the model can automatically supplement possible control targets based on the mutual feedback correlation feature set; when there are outliers or missing values ​​in the mutual feedback correlation feature set, the model can use a fault-tolerant mechanism for feature processing. Anomaly handling branches are implemented by adding corresponding judgment logic and processing algorithms to the model. The optimized collaborative control model is stored in association with parameters, validation results, and test results from the training process, establishing a model archive.

[0225] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for monitoring the operational status of smart mobile cabin equipment based on machine learning, characterized in that, The method includes: Acquire a set of real-time synchronous operation signals, which includes real-time parameter signals of each operating component, inter-equipment interaction signals, and comprehensive environmental impact signals of the container. Perform mutual feedback correlation processing on the set of synchronous real-time signals to extract the set of mutual feedback correlation features in the operating signals. The set of mutual feedback correlation features includes component parameter coupling features, signal transmission features between devices, and environmental parameter cross features. The set of mutually fed-back correlation features is input into the operation status inference model, and a short-term operation status evolution sequence is generated through time-series feature progressive calculation. The short-term operation status evolution sequence includes the predicted values ​​of the parameters of the operating components and the predicted values ​​of the overall operation coordination in each time period. Based on the short-term operational status evolution sequence, potential abnormal evolution trends are identified, and the scope of abnormal correlation impact is determined by combining the mutual feedback correlation feature set, thereby generating information on coordinated regulation and control requirements. The collaborative control demand information and the set of mutual feedback correlation features are input into the collaborative control model to generate multi-device collaborative operation control instructions. These instructions are then sent to the corresponding control modules. The set of synchronous operation feedback signals after execution is collected and input into the mutual feedback correlation processing stage to update the extraction rules of the mutual feedback correlation feature set. At the same time, the timing operation parameters of the operation status inference model and the instruction generation logic of the collaborative control model are adjusted.

2. The method for monitoring the operational status of smart mobile cabin equipment based on machine learning according to claim 1, characterized in that, The step of performing mutual feedback correlation processing on the set of synchronously running real-time signals to extract the set of mutual feedback correlation features from the running signals includes: The real-time parameter signals of each operating component are separated from the set of synchronous real-time signals, and the time-domain variation characteristics of the parameters of each operating component are extracted. The time-domain variation characteristics include parameter fluctuation period characteristics and parameter change rate characteristics. Separate the device-to-device interaction signals from the real-time signal set of synchronous operation, and extract the transmission characteristics of the device-to-device interaction signals. The transmission characteristics include signal transmission delay characteristics and interaction parameter matching deviation characteristics. The comprehensive influence signal of the cabin environment in the real-time signal set of synchronous operation is separated, and the differential effect characteristics of the environmental signal on the parameters of different operating components are extracted. The differential effect characteristics include the influence amplitude characteristics and influence lag characteristics of the environmental parameters on the component parameters. The time-domain variation features and the transmission features are correlated at the first level to calculate the coupling coefficient between the parameters of different operating components within the same device and the interaction signals between devices, thereby generating the first correlation feature. The first correlation feature and the differential effect feature are subjected to a second-level correlation operation to calculate the indirect influence coefficient of environmental signals on the parameters of operating components through inter-device interaction, and the second correlation feature is generated. After unifying the dimensions of the first and second related features, dynamic weights are assigned to the unified first and second related features based on the feature importance evaluation results. The weighted first and second related features are then integrated to generate a set of mutually fed-back related features that includes component parameter coupling features, signal transmission features between devices, and cross-feedback features of environmental parameters. The mutual feedback correlation feature set is associated with the acquisition time period information of the synchronous operation real-time signal set, and the core correlation feature subset in the mutual feedback correlation feature set is extracted. The core correlation feature subset is the feature combination whose influence weight on the operation state prediction exceeds a preset weight threshold.

3. The method for monitoring the operational status of intelligent mobile cabin equipment based on machine learning according to claim 1, characterized in that, The step of inputting the set of mutually fed-back correlation features into the operational state deduction model and generating a short-term operational state evolution sequence through progressive calculation of time-series features includes: The mutual feedback correlation feature set is input into the feature preprocessing layer of the running state inference model. The mutual feedback correlation feature set is then subjected to temporal alignment processing. The temporally aligned mutual feedback correlation feature set is then passed to the first temporal operation layer of the running state inference model. Local temporal correlation information of the mutual feedback correlation feature set is extracted through convolution operation to generate local temporal feature vectors. The local time series feature vector is passed to the second time series operation layer of the running state inference model. The local time series feature vector is modeled in a long-term correlation through a loop unit to capture the dependency relationship of the local time series feature vector within a continuous time step and generate a long-term time series feature vector. The long-term time series feature vector is passed to the feature abstraction layer of the running state inference model. The long-term time series feature vector is then subjected to high-order feature abstraction by a multilayer perceptron to generate a high-order running state feature vector. The high-order operating state feature vector is passed to the prediction output layer of the operating state inference model. The high-order operating state feature vector is converted into the predicted values ​​of the operating component parameters for each time period through linear mapping operation. At the same time, the coordination coefficient of the predicted values ​​of the different operating component parameters for each time period is calculated to generate the overall operating coordination degree prediction value for each time period. Arrange the predicted values ​​of the operating component parameters for each time period and the predicted values ​​of the overall operating synergy for the corresponding time period in chronological order to form an initial short-term operating state evolution sequence. The initial short-term operating state evolution sequence is subjected to trend smoothing processing. Key time node features are extracted from the smoothed short-term operating state evolution sequence. The key time node features are associated with and stored in the smoothed short-term operating state evolution sequence. The key time node features are the features corresponding to the time nodes when the predicted values ​​of operating component parameters or the predicted values ​​of overall operating synergy change significantly.

4. The method for monitoring the operational status of intelligent mobile cabin equipment based on machine learning according to claim 3, characterized in that, The construction process of the operational state simulation model includes: Collect a sample synchronous operation data set, which includes operating component parameter signals, inter-device interaction signals, and comprehensive environmental impact signals under different operating scenarios; The sample synchronous running data set is divided into time series segments. Each continuous sample data segment is divided into sample input segments and corresponding sample output segments according to a fixed time length. The sample output segments are the running data of the subsequent time periods of the sample input segments. Perform mutual feedback correlation processing on the sample input segments and extract the sample mutual feedback correlation feature set, which includes sample component parameter coupling features, signal transmission features between sample devices, and sample environmental parameter cross features. The set of mutual feedback correlation features of samples is used as the input sample of the operation state inference model, and the parameter values ​​of the operating components and the overall operation coordination value in the sample output segment are used as the output sample of the operation state inference model. The input samples and output samples of the operational state simulation model are divided into training sample set, validation sample set and test sample set according to a preset ratio; The initial network structure for constructing the operational state inference model includes a feature preprocessing layer, a first temporal operation layer, a second temporal operation layer, a feature abstraction layer, and a prediction output layer. Set the initial values ​​of the parameters of each layer of the initial network structure, including the initial parameters of the convolution kernels of the first temporal operation layer, the initial parameters of the recurrent units of the second temporal operation layer, and the initial parameters of the perceptron of the feature abstraction layer; The training sample set is input into the initial network structure, and the error between the predicted output of the running state inference model and the output samples of the training sample set is calculated through the backpropagation algorithm, and the parameters of each layer are adjusted accordingly. During the training process, after each preset training round, the prediction accuracy of the running state inference model is evaluated using the validation sample set, and the change curve of the validation accuracy is recorded. If the validation accuracy does not improve for a consecutive preset round, the training is stopped by adopting an early stop strategy to avoid overfitting of the running state inference model. The performance of the trained running state inference model is tested using a test sample set. The average prediction error of the running state inference model on the test sample set is calculated, and the prediction accuracy is calculated. If the average prediction error exceeds the preset test error threshold or the prediction accuracy is lower than the preset test accuracy threshold, the network structure of the running state inference model is adjusted, including increasing the number of convolutional kernels in the first time-series operation layer and adjusting the number of recurrent units in the second time-series operation layer. The training and testing are then repeated. The computational efficiency of the tested operational state simulation model is optimized so that it can generate a short-term operational state evolution sequence within a preset computation time. The optimized operational state simulation model, along with the training parameters, verification results, and test results from the model construction process, are stored together.

5. The method for monitoring the operational status of intelligent mobile cabin equipment based on machine learning according to claim 1, characterized in that, The process of identifying potential abnormal evolution trends based on short-term operational state evolution sequences, determining the scope of abnormal correlation impact by combining a set of mutually fed-back correlation features, and generating coordinated control demand information includes: Extract the predicted values ​​of operating component parameters for each time period in the short-term operating state evolution sequence, compare them with the preset normal operating parameter range, mark the predicted values ​​of operating component parameters that exceed the normal operating parameter range and the corresponding time periods, and count the number of time periods that exceed the normal parameter range for each operating component in the short-term operating state evolution sequence, calculate the percentage of abnormal time periods, and the percentage of abnormal time periods is the ratio of the number of time periods that exceed the normal parameter range to the total number of predicted time periods. Extract the overall operational coordination prediction value for each time period in the short-term operational status evolution sequence, compare it with the preset normal coordination range, mark the overall operational coordination prediction value and corresponding time period that are lower than the normal coordination range, and count the number of time periods where the overall operational coordination prediction value is lower than the normal coordination range, and calculate the proportion of coordination abnormal time periods. If the proportion of abnormal periods for any operating component exceeds the preset component abnormality threshold, or the proportion of abnormal periods for overall operational coordination exceeds the preset coordination abnormality threshold, then it is determined that there is a potential abnormal evolution trend. Extract the component parameter prediction value change curves corresponding to the operating components with potential abnormal evolution trends, analyze the change slope and change acceleration of the component parameter prediction value change curves, and determine the rate characteristics of abnormal evolution. By combining the component parameter coupling characteristics in the mutual feedback correlation feature set, other operating components that are coupled with operating components that have potential abnormal evolution trends are found to form a list of directly related components; By combining the signal transmission characteristics between devices in the mutual feedback correlation feature set, other devices that are associated with devices that have potential abnormal evolution trends through signal transmission are identified, forming a list of indirectly related devices; By integrating the list of directly associated components and the list of indirectly associated devices, the scope of impact of abnormal associations is determined. The scope of impact of abnormal associations includes operating components and devices that are potentially affected by abnormalities. Based on the rate characteristics of abnormal evolution, the size of the scope of abnormal correlation and the degree of abnormality in overall operational coordination, the emergency level of coordinated control is determined, and based on the emergency level of coordinated control and the scope of abnormal correlation and impact, the number of operating components and equipment that need to participate in control is determined, and a list of control objects is generated. Based on the emergency level, the list of control targets, and the rate characteristics of abnormal evolution, control targets are generated. The control targets include parameter values ​​that need to be adjusted to the normal range and synergy values ​​that need to be restored. By integrating the emergency level, the list of control targets, and the control objectives, a coordinated control demand information is generated, which also includes the time window requirements for control implementation.

6. The method for monitoring the operational status of intelligent mobile cabin equipment based on machine learning according to claim 1, characterized in that, The step of inputting the coordinated control demand information and the set of mutually related features into the coordinated control model to generate multi-device coordinated operation control instructions includes: Input the demand information for coordinated regulation into the demand parsing layer of the coordinated regulation model to obtain the emergency level, the list of regulation objects, the regulation target and the time window requirements. The set of mutually fed-back correlation features is input into the correlation feature analysis layer of the collaborative control model to extract the component parameter coupling features, signal transmission features between devices, and environmental parameter cross features corresponding to the list of control objects, and generate a subset of control correlation features; The subset of regulatory features and the regulatory targets obtained by analysis are input into the parameter adjustment calculation layer of the collaborative regulation model to calculate the initial parameter adjustment amount for each regulatory object. In addition, the coupling coefficient in the set of mutual feedback features is used to calculate the indirect impact value of the initial parameter adjustment amount on the related regulatory object, and to analyze whether the indirect impact value exceeds the normal parameter range. If the indirect impact value exceeds the normal parameter range, adjust the initial parameter adjustment amount and recalculate the indirect impact value until the indirect impact value is within the normal parameter range to obtain the final parameter adjustment amount; Based on the urgency level in the coordinated control demand information, control priorities are assigned to the final parameter adjustment amounts of each control object. Based on the time window requirements in the coordinated regulation demand information, allocate regulation execution periods for the final parameter adjustment amount of each regulation object; Based on the control priority and control execution period, a control execution sequence is generated. The identifier, final parameter adjustment amount, control execution period and control execution sequence of each control object are integrated to generate a single-object control instruction. All single-object control commands are arranged in the order of control execution to generate multi-device collaborative operation control commands. Control batch identifiers and time-sensitivity identifiers are added to the multi-device collaborative operation control commands. The time-sensitivity identifier is used to indicate the effective execution time period of the multi-device collaborative operation control commands. Key control parameters are extracted from the multi-device collaborative operation control instructions, and a control parameter summary is generated. The control parameter summary includes the core adjustment amount and the key execution period.

7. The method for monitoring the operational status of intelligent mobile cabin equipment based on machine learning according to claim 6, characterized in that, The construction process of the aforementioned coordinated regulation model includes: Collect a sample collaborative regulation data set, which includes sample collaborative regulation demand information, sample mutual feedback correlation feature set, sample multi-device collaborative operation regulation instructions and corresponding sample synchronous operation feedback signal set. The sample coordinated regulation data set is organized, and the sample coordinated regulation demand information and the corresponding sample mutual feedback correlation feature set are used as input samples of the coordinated regulation model, and the sample multi-device coordinated operation regulation command is used as output samples of the coordinated regulation model. The collaborative regulation demand information of the input samples in the collaborative regulation model is processed in a structured manner to extract the sample emergency level, sample regulation object list, sample regulation target and sample time window requirements, and generate structured input features; Feature filtering is performed on the sample feedback correlation feature set in the input sample of the collaborative control model. The sample control correlation feature subset corresponding to the sample control object list is extracted as the correlation input feature. The structured input feature and the correlation input feature are integrated to form the input sample of the collaborative control model. The parameter adjustment amount, control priority and control execution period in the multi-device collaborative operation control command of the sample are used as the key features of the output sample of the collaborative control model. The complete input samples and output samples of the collaborative regulation model are divided into model training set, model validation set and model test set according to a preset ratio; The initial structure of the collaborative regulation model is constructed, which includes a demand analysis layer, a correlation feature analysis layer, a parameter adjustment calculation layer, a priority allocation layer, and a time period allocation layer. Set the initial parameters for each layer of the initial structure, including the initial value of the coupling coefficient of the parameter adjustment calculation layer, the initial value of the weight of the priority allocation layer, and the initial parameters of the algorithm of the time segment allocation layer; Input the model training set into the initial structure, calculate the error between the output of the collaborative regulation model and the output samples of the model training set using the gradient descent algorithm, adjust the parameters of each layer, and during the training process, after each preset training batch, use the model validation set to evaluate the matching degree between the regulation instructions generated by the collaborative regulation model and the output samples of the validation set using the matching degree evaluation algorithm, and record the trend of matching degree changes. If the matching degree does not improve for a continuous preset batch, adjust the weight of the loss function of the collaborative regulation model and retrain until the matching degree reaches the preset validation matching threshold. The trained collaborative regulation model is tested using a model test set. The collaborative compliance rate and regulation effect compliance rate of the regulation instructions generated by the collaborative regulation model on the test set are calculated. If the collaborative compliance rate is lower than the preset collaborative threshold or the regulation effect compliance rate is lower than the preset effect threshold, the network structure of the collaborative regulation model is adjusted, including adding parameters to adjust the computational dimension of the computation layer, optimizing the computational logic of the priority allocation layer, and retraining and testing. The logic of the tested collaborative control model is optimized by adding an abnormal situation handling branch. This enables the collaborative control model to generate reasonable control instructions when the control demand information is incomplete or the set of mutual feedback correlation features is abnormal. The optimized collaborative control model, along with the training parameters, verification results, and test results from the collaborative control model construction process, are then stored together.

8. The method for monitoring the operational status of smart mobile cabin equipment based on machine learning according to claim 1, characterized in that, The step of inputting the synchronous operation feedback signal set into the mutual feedback correlation processing stage and updating the extraction rules of the mutual feedback correlation feature set includes: The synchronous operation feedback signal set is analyzed, and the deviation data between the predicted value of the short-term operation state evolution sequence of the operation state inference model and the actual operation state value in the synchronous operation feedback signal set is extracted to generate a prediction deviation feature set. The prediction deviation feature set includes the prediction deviation value of each operating component parameter and the prediction deviation value of the overall operation coordination. The feature set of prediction deviations is decomposed to extract the fluctuation cycle features, change rate features, and time series change trend features of the prediction deviation values ​​of component parameters and overall operational coordination. The fluctuation period characteristics and change rate characteristics of the predicted deviation values ​​of component parameters are correlated with the real-time parameter signals in the synchronous operation feedback signal set. The deviation sensitivity factor of each operating component parameter is calculated, and the deviation sensitivity factor is used to quantify the contribution of parameter fluctuations to the prediction deviation. The temporal variation trend of the overall operational coordination degree prediction deviation value is correlated with the inter-equipment interaction signals and the comprehensive impact signals of the container environment in the synchronous operation feedback signal set to extract the coordination deviation transmission path characteristics. The coordination deviation transmission path characteristics include the delay characteristics of inter-equipment interaction signals and the indirect impact characteristics of environmental parameters on coordination degree. Based on the deviation sensitivity factor and the characteristics of the collaborative deviation transmission path, a feature correlation strength evaluation matrix is ​​constructed. The feature correlation strength evaluation matrix is ​​used to describe the degree of correlation between different types of signal features. Based on the feature association strength evaluation matrix, adjust the extraction rules for the mutual feedback association feature set, including: The extraction rules for component parameter coupling features are expanded to include a dimension for calculating the coupling coefficient between operating component parameters where the deviation sensitivity factor exceeds a preset threshold. The extraction rules for signal transmission characteristics between devices are improved, and the weight allocation of the interaction signal delay characteristics between devices in the characteristics of cooperative deviation transmission path is strengthened. The extraction rules for the cross-features of environmental parameters introduce a dynamic evaluation mechanism for the indirect impact of environmental parameters on the degree of synergy. The extraction rules of the adjusted mutual feedback correlation feature set are associated and stored with the collection time period information of the synchronous operation feedback signal set to form a feature extraction rule library that is dynamically updated over time. The extraction rules of the updated mutual feedback correlation feature set are verified using the historical synchronous running data set. The prediction deviation change rate before and after the feature extraction rule adjustment is calculated. If the prediction deviation change rate does not reach the preset optimization threshold, the feature correlation strength evaluation matrix is ​​reconstructed and the extraction rules are adjusted again until the preset optimization threshold is met.

9. A machine learning-based intelligent mobile cabin equipment operation status monitoring system, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the machine learning-based smart shelter equipment operation status monitoring method according to any one of claims 1 to 8 by executing the machine executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. The processor of the computer device reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the computer device to perform the machine-based smart shelter equipment operation status monitoring method as described in any one of claims 1 to 8.

Images