An artificial intelligence-based instrument device remote monitoring method

Abstract

The application provides an instrument and equipment remote monitoring method based on artificial intelligence, and relates to the technical field of equipment monitoring network, comprising: obtaining runtime sequence data sets of a plurality of heterogeneous devices distributed in a target area; generating a uniform dimension feature vector set corresponding to each heterogeneous device; inputting the uniform dimension feature vector set into a pre-trained space-time coupling analysis model to extract spatial graph correlation features between the plurality of heterogeneous devices in the target area and time evolution correlation features of a single heterogeneous device; outputting a global abnormal risk index vector of the target area and an energy consumption modulation instruction for an abnormal heterogeneous device through a fully connected mapping layer in the space-time coupling analysis model; and based on the global abnormal risk index vector, the energy consumption modulation instruction is issued to the corresponding abnormal heterogeneous device through an edge computing node to perform device state adjustment; the application improves the matching degree of the energy consumption scheduling instruction and the real load state of the global system.

Description

Technical Field

[0001] This invention relates to the technical field of equipment monitoring networks, and in particular to a remote monitoring method for instruments and equipment based on artificial intelligence. Background Technology

[0002] With the accelerated advancement of digital transformation, various large organizations have built distributed device management systems that integrate hardware platforms, software platforms, and various expansion units. These systems support multiple communication interfaces and protocols, aiming to achieve interconnection, expansion, and centralized management of massive heterogeneous devices.

[0003] In such highly heterogeneous device network architectures, existing technologies have a significant technical deficiency when performing remote management and energy consumption monitoring. Specifically, existing solutions cannot accurately identify implicit state coupling anomalies among multiple devices under complex business cycles in highly heterogeneous distributed device networks. The operating status of devices is the result of the combined effects of electrical, environmental, and network factors. These implicit state coupling anomalies not only exist in the correlation between devices with similar data types but also prevalent in cross-dimensional correlations among these three types of data. For example, anomalies in device electrical power may be caused by excessively high ambient temperature or network packet loss. Traditional monitoring logic... Fixed operating parameter thresholds are typically set for each individual device, and an alarm is triggered when the value exceeds the threshold. However, in real-world applications, there are complex spatiotemporal coupling relationships between heterogeneous devices. Even if the parameters of a certain device do not exceed the absolute threshold, it may actually be in a hidden abnormal high-energy-consumption state when combined with the status of its surrounding environmental devices and the current business rhythm. This single threshold determination method makes it impossible for the system to capture the hidden state coupling anomalies between multiple devices, nor can it identify anomalies caused by cross-dimensional factor correlations, resulting in a large number of false alarms or missed alarms. Consequently, the output energy consumption scheduling instructions cannot accurately match the actual load status of the global system.

[0004] Therefore, it is necessary to provide an artificial intelligence-based method for remote monitoring of instruments and equipment to solve the above-mentioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a remote monitoring method for instruments and equipment based on artificial intelligence.

[0006] This invention provides a remote monitoring method for instruments and equipment based on artificial intelligence, comprising: S101, Obtain the runtime sequence dataset of multiple heterogeneous devices distributed in the target area. Multiple edge computing nodes are configured in the target area, and the heterogeneous devices establish data connections with the edge computing nodes through communication interfaces. S102, through the pre-configured feature embedding module, performs dimension mapping on the runtime sequence dataset to generate a unified set of dimension feature vectors corresponding to each heterogeneous device; S103, input the unified dimension feature vector set into the pre-trained spatiotemporal coupling analysis model to extract the spatial map correlation features between multiple heterogeneous devices in the target area and the temporal evolution correlation features of a single heterogeneous device; S104, based on the spatial map correlation characteristics and temporal evolution correlation characteristics, outputs the global anomaly risk index vector of the target area and the energy consumption modulation command for the anomaly heterogeneous device through the fully connected mapping layer in the spatiotemporal coupling analysis model. S105, based on the global anomaly risk index vector, sends energy consumption modulation commands to the corresponding abnormal heterogeneous devices through edge computing nodes to perform device status adjustment.

[0007] Preferably, the step of obtaining the runtime sequence dataset of multiple heterogeneous devices distributed in the target area includes: obtaining the underlying electrical consumption sequence, the surrounding environment sensing sequence, and the network link status sequence of the heterogeneous devices; and aligning the underlying electrical consumption sequence, the surrounding environment sensing sequence, and the network link status sequence with timestamps using the hardware clock chip in the edge computing node to generate the aligned runtime sequence dataset.

[0008] Preferably, the step of performing dimensional mapping on the runtime sequence dataset through a pre-configured feature embedding module includes: converting the underlying electrical consumption sequence into electrical feature vectors through a linear mapping matrix; converting the surrounding environment sensing sequence and network link state sequence into environmental feature vectors and link feature vectors through a multilayer perceptron; and concatenating the electrical feature vectors, environmental feature vectors, and link feature vectors to generate a unified dimensional feature vector set through orthogonal dimensionality reduction operations.

[0009] Preferably, the spatiotemporal coupling analysis model includes a graph attention layer and a temporal convolution layer. The steps for extracting spatial graph association features among multiple heterogeneous devices within the target area and temporal evolution association features of a single heterogeneous device include: using the graph attention layer to process a set of feature vectors of a unified dimension to generate spatial graph association features; and using the temporal convolution layer to perform sliding window convolution processing on the spatial graph association features in the temporal dimension to generate temporal evolution association features.

[0010] Preferably, the step of processing a set of unified-dimensional feature vectors using a graph attention layer includes: constructing a spatial adjacency matrix based on the topological distance between heterogeneous devices within the target area; using the spatial adjacency matrix as a mask to calculate the attention weight coefficient between any two feature vectors in the set of unified-dimensional feature vectors; and performing weighted aggregation of the feature vectors based on the attention weight coefficient to output spatial graph association features.

[0011] Preferably, the step of using a temporal convolutional layer to perform sliding window convolution processing on the spatial map association features in the temporal dimension includes: setting multiple one-dimensional convolutional kernels with different receptive field sizes; using multiple one-dimensional convolutional kernels to scan the historical time step data of the spatial map association features in parallel; extracting sequence change features at different time scales and splicing them into temporal evolution association features.

[0012] Preferably, the step of outputting the global anomaly risk index vector of the target region through the fully connected mapping layer in the spatiotemporal coupling analysis model includes: performing a tensor product operation on the spatial map correlation features and the temporal evolution correlation features to obtain spatiotemporal fusion features; inputting the spatiotemporal fusion features into the fully connected mapping layer with a nonlinear activation function; and outputting the global anomaly risk index vector in floating-point format.

[0013] Preferably, the step of outputting energy consumption modulation instructions for abnormal heterogeneous devices includes: multiplying the global abnormal risk index vector with the device management strategy matrix to obtain the control strategy coefficient corresponding to each heterogeneous device; matching the target control code in the instruction code library according to the calculation result; and encapsulating the target control code into an energy consumption modulation instruction.

[0014] Preferably, the step of sending power consumption modulation instructions to the corresponding abnormal heterogeneous devices through edge computing nodes includes: the edge computing node parsing the power consumption modulation instructions to obtain the target control code; converting the target control code into the corresponding underlying register operation message according to the communication protocol type of the abnormal heterogeneous device; and sending the underlying register operation message for execution through the data transmission bus of the heterogeneous device.

[0015] Preferably, the step of calculating the attention weight coefficient between any two feature vectors in the unified dimension feature vector set further includes: obtaining business scheduling data records of the target area; encoding the business scheduling data records into business rhythm feature vectors; and introducing the business rhythm feature vectors as bias terms into the calculation process of the attention weight coefficients, thereby dynamically updating the attention weight coefficients between various heterogeneous devices.

[0016] Preferably, the step of sending power consumption modulation instructions to the corresponding abnormal heterogeneous devices through the edge computing node based on the global anomaly risk index vector further includes: real-time monitoring of network link latency parameters between the edge computing node and the cloud server; when the network link latency parameters exceed a preset security latency threshold and an incomplete power consumption modulation instruction is received, the edge computing node intercepts the power consumption modulation instruction; and the edge computing node calls the degradation protection strategy of local storage to switch the abnormal heterogeneous device to a low-power safe standby state.

[0017] Compared with related technologies, the remote monitoring method for instruments and equipment based on artificial intelligence provided by this invention involves machine learning and deep learning technologies, and has the following beneficial effects: This invention breaks through the limitations of traditional single physical quantity monitoring by introducing deep mapping of multimodal features and graph attention mechanism. It deeply mines the cross-dimensional correlation features of electrical, environmental and network data, and can keenly discover the hidden state coupling anomalies between heterogeneous devices in the target area. It can accurately identify coupling anomalies between devices with the same type of data and more effectively capture the hidden anomalies caused by cross-dimensional correlation of the three types of data. It significantly reduces the false alarm rate of monitoring in complex business scenarios and provides a high-precision dynamic energy consumption scheduling and anomaly management solution for large equipment networks.

[0018] This invention utilizes a spatiotemporal coupling analysis model to integrate cross-dimensional features from electrical, environmental, and network perspectives. First, it extracts spatial graph association features between heterogeneous devices, defined by spatial proximity and functional synergy, using a graph attention layer. Then, based on these spatial features, it extracts temporal evolution association features of the devices within the coupled network using multi-scale one-dimensional causal convolution kernels. This forms a cascaded analysis architecture that prioritizes space over time, accurately identifying implicit coupling relationships and temporal changes in operational status among multiple devices. This overcomes the limitation of traditional single-threshold judgment methods, which can only identify explicit anomalies. Furthermore, the calculation of functional association strength is based on cross-dimensional fusion features and dynamically adjusts the association weights between devices by incorporating business rhythm features. The multi-scale one-dimensional causal convolution kernels can cover temporal features at different time granularities and rely solely on historical data to ensure the causal rationality of the analysis. This allows the anomaly identification results to not only align with actual business scenarios but also accurately match the actual operational status of devices with cross-dimensional data associations, effectively reducing false alarms and false negatives.

[0019] This invention outputs a global anomaly risk index vector and energy consumption modulation instructions through a fully connected mapping layer, enabling precise location and targeted control of abnormal devices; it improves instruction response speed and execution reliability by completing instruction parsing, protocol conversion, and execution through edge computing nodes; and it activates a degradation protection strategy when the network is abnormal, switching abnormal heterogeneous devices to a low-power safe standby state, further enhancing the stability and security of system operation.

[0020] By collecting runtime sequence data from multiple heterogeneous devices within the target area and aligning it with timestamps, unified processing of multi-source heterogeneous device data was achieved, ensuring data consistency and reliability in the time dimension. Through the feature embedding module, dimension mapping and orthogonal dimensionality reduction were performed on multi-type time-series data to generate a unified set of feature vectors, eliminating processing obstacles caused by the inconsistency of heterogeneous device data dimensions and improving the efficiency and accuracy of subsequent state analysis. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the steps of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0023] Please refer to the following: Figure 1 A remote monitoring method for instruments and equipment based on artificial intelligence, comprising: S101, Obtain the runtime sequence dataset of multiple heterogeneous devices distributed in the target area. Multiple edge computing nodes are configured in the target area, and the heterogeneous devices establish data connections with the edge computing nodes through communication interfaces.

[0024] By strategically deploying edge computing nodes within the target area, near-field data acquisition and preliminary processing of multiple heterogeneous devices within the area can be achieved, reducing the pressure of remote data transmission and improving the real-time performance of data acquisition. Each heterogeneous device establishes a stable data connection with the nearest edge computing node through its own communication interface, ensuring that device operation data is uploaded to the edge computing node in real time, providing reliable data support for subsequent feature processing and anomaly analysis. The runtime sequence dataset covers various key parameters during the operation of heterogeneous devices, comprehensively reflecting the device operating status and providing a sufficient data foundation for subsequent artificial intelligence analysis.

[0025] In some embodiments, step 101, obtaining the runtime sequence dataset of multiple heterogeneous devices distributed within the target area, includes: By using the sensors and communication modules built into the heterogeneous devices, the underlying electrical consumption sequence and the surrounding environment sensing sequence of the devices are collected; by using the network monitoring module built into the edge computing node, the network link status sequence between the edge computing node and the heterogeneous devices, and between the edge computing node and the cloud server is collected. By using the hardware clock chip in the edge computing node, a unified standard timestamp is added to the underlying electrical consumption sequence, the surrounding environment sensing sequence, and the network link status sequence, respectively. The three types of sequences are aligned based on timestamps to eliminate timing discrepancies in data collection and generate an aligned runtime sequence dataset.

[0026] Specifically, the underlying electrical consumption sequence is collected by the electrical sensors built into the device, including core electrical parameters during device operation. These core electrical parameters include the device's real-time operating current, operating voltage, active power, reactive power, and power factor. The surrounding environment sensing sequence is collected by environmental sensors deployed around the device, including core environmental parameters of the device's operating environment. These core environmental parameters include ambient temperature, relative humidity, atmospheric pressure, dust concentration in the air, and ambient noise level. The network link status sequence is collected by the network monitoring module of the edge computing node, including the relationship between the edge computing node and the device. The core link parameters for cloud communication include link transmission bandwidth, data transmission latency, packet loss rate, link jitter value, and link connection stability parameters. This invention uses edge computing nodes as communication and data processing hubs for heterogeneous devices. Communication between heterogeneous devices and the cloud, as well as between heterogeneous devices themselves, is achieved through relaying via edge computing nodes. Therefore, collecting link parameters between edge computing nodes and devices, and between the cloud, can accurately reflect the network foundation for the operation of heterogeneous devices. At the same time, it can characterize the collaborative relationship between heterogeneous devices from the perspective of data interaction effectiveness, and can achieve centralized monitoring of network status, reducing the monitoring cost of distributed heterogeneous device networks. The link connection stability is calculated as (normal connection time per unit time / total time per unit time) x 100%. The parameter value ranges from 0 to 1, and the closer the value is to 1, the more stable the link connection is.

[0027] The hardware clock chip within the edge computing node adds a timestamp to each data point of the three types of sequences according to a unified time standard, ensuring consistency of time stamps. Based on the timestamps, data at the same time point in the three types of sequences are correlated. Data with timestamp deviations exceeding a preset value (e.g., 10ms) are identified as mismatched and removed. Simultaneously, linear interpolation is used to fill in the missing data after removal, ensuring the integrity of the time-series dataset. Finally, an aligned runtime sequence dataset is generated, ensuring consistency in the time dimension and providing a unified time reference for subsequent data processing. The hardware clock chip within the edge computing node has high-precision time synchronization capabilities, adding unified standard timestamps to the underlying electrical consumption sequence, the surrounding environment sensing sequence, and the network link status sequence. Through timestamp alignment, the consistency of the three types of sequences in the time dimension is ensured, avoiding errors in subsequent feature analysis caused by acquisition time-series deviations. The final aligned runtime sequence dataset provides standardized and unified data input for subsequent feature embedding processing.

[0028] S102 uses a pre-configured feature embedding module to perform dimensional mapping on the runtime dataset, generating a unified set of dimensional feature vectors corresponding to each heterogeneous device.

[0029] Because the heterogeneous devices within the target area are of different types, their runtime sequence datasets have significant differences in dimensions and scales. Directly performing subsequent artificial intelligence analysis would lead to large deviations in the results. The pre-configured feature embedding module can standardize and map the runtime sequence datasets of different types of heterogeneous devices, transforming runtime data with different dimensions and scales into unified dimension feature vectors, eliminating the influence of data differences, and simultaneously achieving data dimensionality reduction and feature extraction. This improves the efficiency and accuracy of subsequent model analysis and ultimately generates a unified dimension feature vector set that corresponds one-to-one with each heterogeneous device.

[0030] In some embodiments, step 102, which involves dimensional mapping of the runtime dataset using a pre-configured feature embedding module, includes: A fixed-dimensional linear mapping matrix is ​​preset. The underlying electrical consumption sequence is input into this matrix and transformed into a fixed-dimensional electrical feature vector through linear transformation. By using a pre-defined multilayer perceptron structure, nonlinear transformations are performed on the surrounding environment sensing sequence and network link state sequence, respectively, to convert them into environmental feature vectors and link feature vectors with the same dimension as electrical feature vectors. Electrical feature vectors, environmental feature vectors, and link feature vectors are concatenated in a preset fixed order to obtain a fused high-dimensional feature vector; An orthogonal dimensionality reduction algorithm is used to process the fused high-dimensional feature vectors, remove redundant information, and map them to a unified low-dimensional feature space to generate a set of unified-dimensional feature vectors.

[0031] Specifically, the parameters of the linear mapping matrix are determined through calibration of the equipment operation data in the early stage. Each data point of the underlying electrical consumption sequence is substituted into the matrix, and a fixed-dimensional electrical feature vector is obtained by matrix multiplication. The multilayer perceptron receives the environmental sensing sequence and the network link status sequence, and outputs the environmental feature vector and link feature vector with the same dimension as the electrical feature vector through the nonlinear operation of the hidden layer. The parameters of the linear mapping matrix were calibrated using the least squares method based on prior equipment operation data. The calibration dataset consisted of 30 consecutive days of normal operation time-series data from heterogeneous equipment within the target area, with a dataset completeness of at least 99% and no abnormal operation data. A feature reconstruction error of less than 5% was used as the baseline. The kernel evaluation index for matrix shifting is to ensure that the electrical eigenvectors after linear transformation can completely retain the kernel information of the original sequence.

[0032] The three types of feature vectors are concatenated in a predetermined fixed order: electrical feature vector, environmental feature vector, and link feature vector. The elements of these three types of feature vectors are then chained together to obtain a high-dimensional feature vector. This order is based on the importance of the correlation between the features and the equipment's operating status, prioritizing the retention of information at the core operating feature level. Principal Component Analysis (PCA), an orthogonal dimensionality reduction algorithm, is used to process the fused high-dimensional feature vector, removing redundant information and mapping it to a unified low-dimensional feature space. Specifically, the covariance matrix of the fused high-dimensional feature vector is calculated, and the eigenvalues ​​and eigenvectors of the covariance matrix are extracted. Core feature vectors with a cumulative contribution rate of no less than 5% after sorting the eigenvalues ​​from largest to smallest are selected to construct a dimensionality reduction matrix. The fused high-dimensional feature vector is multiplied by the dimensionality reduction matrix to obtain a feature vector of unified dimension. The unified-dimensional feature vectors of all heterogeneous devices are integrated and arranged in ascending order according to the unique physical ID of the heterogeneous devices to generate a set of unified-dimensional feature vectors. The arrangement order of the feature vectors in the set corresponds one-to-one with the unique physical ID of the heterogeneous devices, ensuring that the feature vectors can accurately match the target devices when input into the subsequent spatiotemporal coupling analysis model.

[0033] S103 inputs a unified dimensional feature vector set into a pre-trained spatiotemporal coupling analysis model to extract spatial graph correlation features between multiple heterogeneous devices within the target area and temporal evolution correlation features of a single heterogeneous device. Among them, the temporal evolution correlation features are generated based on the spatial graph correlation features, aiming to establish the temporal change analysis of a single device on the basis of spatial collaboration and functional coupling between devices, capture the temporal change patterns of device states under coupling correlation, and avoid the limitations of isolated analysis of device temporal sequences. The spatial graph correlation features provide a static coupling correlation framework for the temporal evolution correlation features, and the temporal evolution correlation features give the static coupling structure temporal dynamism. The two form complementary spatiotemporal coupling features, which together provide core feature support for subsequent identification of hidden state coupling anomalies of devices.

[0034] A spatial adjacency matrix is ​​constructed based on the topological distance of heterogeneous devices within the target area to represent the spatial correlation between devices. Using the spatial adjacency matrix as a mask, the attention weight coefficient between any two feature vectors in the unified dimension feature vector set is calculated. This coefficient represents the functional correlation strength between devices. The functional correlation strength and spatial correlation are independent of each other, have no constraints, and have no subordinate relationship. The functional correlation strength serves as the core basis for weighted aggregation, which can break through the limitations of physical space to explore the actual business coupling relationship between devices. It can also be dynamically adjusted in combination with business rhythm to make the characterization of the correlation between devices more in line with the actual business scenario. The temporal evolution correlation characteristics of a single heterogeneous device can capture the changing patterns of the operating parameters of a single device over time under the background of spatial coupling and reflect the temporal evolution of the coupling relationship between devices. This enables dynamic tracking and anomaly prediction of the device's operating status, providing core feature support for subsequent anomaly risk assessment and energy consumption control.

[0035] The spatiotemporal coupling analysis model is trained based on a large amount of heterogeneous device operation data and has the ability to capture spatial correlation and temporal evolution characteristics. By inputting a unified dimensional feature vector set into the model, two core features can be mined simultaneously: First, the spatial map correlation features among multiple heterogeneous devices in the target area. This feature can reflect the correlation between the spatial layout and functional coordination of each heterogeneous device, helping to identify potential hazards caused by abnormal coordination between devices. Second, the temporal evolution correlation features of a single heterogeneous device. This feature can capture the changing patterns of the operating parameters of a single device over time, enabling dynamic tracking and anomaly prediction of the device's operating status, and providing core feature support for subsequent anomaly risk assessment and energy consumption control.

[0036] In some embodiments, the spatiotemporal coupling analysis model includes a graph attention layer and a temporal convolutional layer. The steps of extracting spatial graph correlation features among multiple heterogeneous devices within a target region and temporal evolution correlation features of a single heterogeneous device include: A graph attention layer is used to process a set of feature vectors of the same dimension to generate spatial graph association features; Temporal convolutional layers are used to perform sliding window convolution on spatial map association features in the temporal dimension to generate temporally evolved association features.

[0037] Specifically, the graph attention layer is a network layer specifically designed to handle spatial relationships. Based on the feature vectors of each heterogeneous device, it automatically learns the relationship weights between devices, focusing on the mutual influence between devices with high correlation. Through weighted aggregation, it generates spatial graph relationship features, clearly representing the spatial relationship structure of multiple heterogeneous devices within the target area. The temporal convolutional layer has the ability to capture the changing patterns of time-series data. By setting a sliding window, it performs parallel convolution processing on the spatial graph relationship features in the time dimension, which can effectively extract the changing features of device operating status at different time scales, thereby generating temporal evolution relationship features, accurately reflecting the temporal change trend of the operating parameters of a single heterogeneous device and the dynamic evolution of the relationship between devices.

[0038] In some embodiments, the step of processing a set of feature vectors of uniform dimension using a graph attention layer includes: Construct a spatial adjacency matrix based on the topological distances of heterogeneous devices within the target area; Using the spatial adjacency matrix as a mask, calculate the attention weight coefficient between any two feature vectors in a set of feature vectors of the same dimension; The feature vectors are weighted and aggregated based on attention weight coefficients to output spatial map association features.

[0039] Specifically, firstly, the actual deployment locations of multiple heterogeneous devices within the target area are obtained, and the topological distance between any two heterogeneous devices is calculated. The smaller the topological distance, the closer the two devices are in space, and the higher the correlation is usually. A spatial adjacency matrix is ​​constructed based on the topological distance. The matrix elements are used to characterize the spatial correlation strength between two heterogeneous devices. The smaller the topological distance, the larger the corresponding element value, and vice versa.

[0040] Using the spatial adjacency matrix as a mask, the association between any two feature vectors in the set of feature vectors of the same dimension is constrained, focusing only on device pairs with high spatial correlation to avoid interference from irrelevant devices; by using a preset attention calculation function, the attention weight coefficient between any two feature vectors is calculated. This coefficient is used to characterize the functional association strength between two devices. The larger the coefficient, the more significant the mutual influence between the operating states of the two devices.

[0041] Based on the calculated attention weight coefficients, all feature vectors in the unified dimension feature vector set are weighted and aggregated. Under the pre-constraint of spatial correlation strength, the features of devices with high functional correlation strength are mainly integrated. The final output is a spatial map correlation feature that clearly reflects the business function collaborative correlation structure under the physical space constraints between devices. The spatial map correlation feature between multiple heterogeneous devices in the target area can reflect the fusion correlation relationship of each heterogeneous device in spatial layout and functional collaboration, and help identify potential problems caused by abnormal functional collaboration between devices.

[0042] In some embodiments, the step of calculating the attention weight coefficient between any two feature vectors in a set of feature vectors with uniform dimension further includes: Retrieve business scheduling data records for the target region; Encode business scheduling data records into business rhythm feature vectors; By incorporating the business rhythm feature vector as a bias term into the calculation of the attention weight coefficient, the attention weight coefficients between various heterogeneous devices are dynamically updated.

[0043] Specifically, the business scheduling data records of the target area reflect the operational patterns of business and equipment usage requirements within the area. The operating load and coordination relationships of heterogeneous equipment will change during different business periods. The business scheduling data records are transformed into a fixed-dimensional business rhythm feature vector through an encoding algorithm. This vector can represent the business demand and equipment operating rhythm at different times.

[0044] By incorporating the business rhythm feature vector as a bias term into the attention weight coefficient calculation process, the attention weight coefficients between various heterogeneous devices can be dynamically adjusted according to changes in business rhythm. For example, during peak business periods, the attention weight coefficients between devices related to core business will increase accordingly, focusing on the collaborative operation status of these devices; during off-peak business periods, the correlation between devices decreases, and the attention weight coefficients are adjusted accordingly to ensure that they align with actual business needs and improve the accuracy of spatial map correlation features.

[0045] In some embodiments, the step of performing sliding window convolution processing on spatial map association features in the temporal dimension using a temporal convolutional layer includes: Set up multiple one-dimensional convolutional kernels with different receptive field sizes; Utilize multiple one-dimensional convolutional kernels to scan historical time-step data of spatial map association features in parallel; Extract sequence change features at different time scales and concatenate them into temporal evolution correlation features.

[0046] Specifically, in order to comprehensively capture the changing patterns of spatial map association features at different time scales, multiple one-dimensional convolutional kernels with different receptive field sizes are set. The smaller the receptive field size, the more it can capture the changing features at short-term time scales, while the larger the receptive field size can capture the changing features at long-term time scales. The convolutional kernels with multiple receptive field sizes work in parallel, which can simultaneously extract the sequence change features at different time scales and avoid feature omissions caused by a single receptive field size.

[0047] Multiple one-dimensional convolutional kernels are used to scan the historical time step data of spatial map correlation features in parallel. Each convolutional kernel extracts the change features at a corresponding time scale. The features extracted by all convolutional kernels are spliced ​​together to integrate the sequence change information at different time scales, and finally a time evolution correlation feature is generated. This feature can comprehensively and accurately reflect the time change trend of the operating parameters of a single heterogeneous device and the evolution of the correlation between devices.

[0048] S104, based on the spatial map correlation characteristics and temporal evolution correlation characteristics, outputs the global anomaly risk index vector of the target area and the energy consumption modulation command for the anomaly heterogeneous device through the fully connected mapping layer in the spatiotemporal coupling analysis model.

[0049] The fully connected mapping layer within the spatiotemporal coupling analysis model can perform deep fusion and nonlinear mapping of spatial map correlation features and temporal evolution correlation features, transforming the fused features into output results with practical physical meaning. Among them, the global anomaly risk index vector is used to characterize the overall anomaly risk level of all heterogeneous devices within the target area. Each element in the vector corresponds to the anomaly risk index of a heterogeneous device. The higher the index, the greater the probability of the device malfunctioning. The energy consumption modulation command is used to regulate the energy consumption of heterogeneous devices with anomaly risk indices exceeding the standard, enabling timely intervention in abnormal devices, preventing the expansion of device failures, and optimizing device energy consumption to improve energy utilization efficiency.

[0050] In some embodiments, the global anomaly risk index vector of the target region output in S104 includes: Tensor multiplication is performed on the spatial map correlation features and the temporal evolution correlation features, and element-wise multiplication is performed on the spatial map correlation features and the temporal evolution correlation features to obtain the spatiotemporal fusion features; The spatiotemporal fusion features are input into a fully connected processing layer with a nonlinear activation function for nonlinear mapping processing; Output a global anomaly risk index vector in floating-point format. The vector dimension is the same as the number of heterogeneous devices. Each element in the vector corresponds to the anomaly risk index of a heterogeneous device, which is used to characterize the degree of anomaly risk of the device.

[0051] Specifically, the tensor product operation performs element-wise multiplication of spatial map correlation features and temporal evolution correlation features in a preset dimensional order to obtain spatiotemporal fusion features that combine spatial correlation and temporal evolution information. The fully connected processing layer maps the spatiotemporal fusion features to a preset numerical range through a nonlinear activation function and outputs a global anomaly risk index vector. The larger the value of the vector element, the higher the anomaly risk of the corresponding device.

[0052] In some embodiments, the output of the power consumption modulation command for the abnormal heterogeneous device in S104 includes: A preset equipment management strategy matrix is ​​provided, where rows correspond to heterogeneous equipment numbers and columns correspond to different abnormal risk threshold ranges or control strategies; the matrix elements are the equipment control coefficients corresponding to different risk ranges. Multiply the global anomaly risk index vector with the equipment management strategy matrix to obtain the control strategy coefficients for each heterogeneous device; A pre-defined instruction code library stores equipment control codes corresponding to different control strategy coefficients; Based on the control strategy coefficient, match the corresponding target control code in the instruction code library; According to the preset communication protocol format, the target control code is encapsulated into standardized energy consumption modulation instructions to ensure that the instructions can be recognized and executed by the equipment.

[0053] Specifically, the rows of the equipment management strategy matrix correspond to the equipment number, the columns correspond to different abnormal risk levels, and the matrix elements are the control strategy coefficients. The higher the level, the larger the corresponding coefficient. The inner product calculation follows the multiplication rules of vectors and matrices to obtain the control strategy coefficient for each equipment. Based on the calculated control strategy coefficient, the corresponding target control code is extracted from the instruction code library, encapsulated according to the preset communication protocol, and a standardized energy consumption modulation instruction is generated.

[0054] S105, based on the global anomaly risk index vector, sends energy consumption modulation commands to the corresponding abnormal heterogeneous devices through edge computing nodes to perform device status adjustment.

[0055] Based on the global anomaly risk index vector, heterogeneous devices with anomaly risk indices exceeding a preset threshold are screened out. Edge computing nodes, as the core carriers for data transmission and command issuance, can quickly parse energy consumption modulation commands and accurately send them to the corresponding abnormal heterogeneous devices. After receiving the commands, the abnormal heterogeneous devices perform corresponding state adjustment operations, enabling timely control of the abnormal devices, reducing the risk of device failure, optimizing device energy consumption, and ensuring the stable and efficient operation of all heterogeneous devices in the target area.

[0056] In some embodiments, the step of sending power consumption modulation commands to corresponding abnormal heterogeneous devices via edge computing nodes includes: Edge computing nodes parse energy consumption modulation instructions to obtain target control code; Based on the communication protocol type of the abnormal heterogeneous devices, the target control code is converted into the corresponding underlying register operation message; The underlying register operation messages are sent and executed through the data transmission bus of heterogeneous devices.

[0057] Specifically, after receiving the energy consumption modulation command output by the spatiotemporal coupling analysis model, the edge computing node parses the command, extracts the target control code, and clarifies the control operation to be performed by the abnormal heterogeneous device. Since the types of heterogeneous devices in the target area are different, and the communication protocols they use are also different, the edge computing node converts the target control code into the corresponding underlying register operation message according to the specific communication protocol type of the abnormal heterogeneous device, ensuring that the message can be correctly identified and parsed by the abnormal heterogeneous device.

[0058] Through the data transmission bus of the heterogeneous device itself, the underlying register operation message is sent to the control module of the abnormal heterogeneous device. After receiving the message, the control module executes the corresponding register operation, adjusts the device operating parameters, realizes the precise adjustment of the device status, and completes the energy consumption control and fault intervention of the abnormal device.

[0059] In some embodiments, the step of sending power consumption modulation commands to corresponding abnormal heterogeneous devices via edge computing nodes based on a global anomaly risk index vector further includes: Real-time monitoring of network link latency parameters between edge computing nodes and cloud servers; When the network link delay parameter exceeds the preset security delay threshold and an incomplete power consumption modulation command is received, the edge computing node intercepts the power consumption modulation command. Edge computing nodes invoke local storage degradation protection strategies to switch abnormal heterogeneous devices to a low-power safe standby state.

[0060] Specifically, the stability of the network link between edge computing nodes and cloud servers directly affects the efficiency and integrity of energy consumption modulation commands. Therefore, it is necessary to monitor the network link latency parameters in real time. The preset security latency threshold is the critical value to ensure that commands are issued and executed normally. When the network link latency parameter exceeds this threshold, problems such as command transmission interruption and command loss are likely to occur, resulting in incomplete energy consumption modulation commands.

[0061] When the network link latency parameter is detected to exceed the safe latency threshold and the edge computing node receives an incomplete power consumption modulation command, in order to avoid device malfunction caused by the incomplete command, the edge computing node will immediately intercept the power consumption modulation command to prevent the command from being sent to abnormal heterogeneous devices. At the same time, the edge computing node calls the degradation protection policy of local storage to switch the abnormal heterogeneous device to a low-power safe standby state. This reduces device power consumption and prevents the device from malfunctioning due to continuous operation in an abnormal state, ensuring device safety. After the network link returns to normal, the complete power consumption modulation command will be received and sent again.

[0062] In summary, the artificial intelligence-based remote monitoring method for instruments and equipment provided in this application achieves real-time acquisition and timestamp alignment of runtime sequence data from multiple heterogeneous devices through edge computing nodes, ensuring data accuracy and consistency; it transforms runtime data of different dimensions into feature vectors of a unified dimension through a feature embedding module, eliminating the impact of data differences; it extracts spatial map correlation features between devices and temporal evolution correlation features of the devices themselves through a spatiotemporal coupling analysis model, achieving accurate assessment of device anomaly risks; and it outputs a global anomaly risk index vector and energy consumption modulation instructions through a fully connected mapping layer, combined with the instruction issuance and degradation protection mechanism of edge computing nodes, to achieve timely control and safety protection of abnormal devices.

[0063] This method effectively solves the problems of delayed anomaly identification, inaccurate energy consumption control, and unstable command issuance in traditional remote monitoring methods. It realizes collaborative monitoring of multiple heterogeneous devices, intelligent anomaly prediction, and precise energy consumption control, improving the efficiency and reliability of remote monitoring, reducing equipment failure risk and energy consumption. It is applicable to various target areas with multiple heterogeneous devices and has broad application prospects.

[0064] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A remote monitoring method for instruments and equipment based on artificial intelligence, characterized in that, include: Obtain runtime sequence datasets from multiple heterogeneous devices distributed within the target area. Multiple edge computing nodes are configured within the target area, and data connections are established between the heterogeneous devices and the edge computing nodes. The runtime dataset is dimension-mapped by a pre-configured feature embedding module to generate a unified set of dimensional feature vectors corresponding to each heterogeneous device. A unified set of feature vectors is input into a pre-trained spatiotemporal coupling analysis model to extract spatial map correlation features between multiple heterogeneous devices within the target area and temporal evolution correlation features of a single heterogeneous device. Based on the spatial map correlation characteristics and temporal evolution correlation characteristics, the global anomaly risk index vector of the target area and the energy consumption modulation command for anomaly heterogeneous devices are output through the fully connected mapping layer in the spatiotemporal coupling analysis model. Based on the global anomaly risk index vector, energy consumption modulation commands are sent to the corresponding abnormal heterogeneous devices through edge computing nodes to perform device status adjustments.

2. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 1, characterized in that, The steps to obtain the runtime dataset are as follows: Acquire the underlying electrical consumption sequence of heterogeneous devices, the surrounding environment sensing sequence, and the network link status sequence; The underlying electrical consumption sequence, the surrounding environment sensing sequence, and the network link status sequence are timestamped by the hardware clock chip in the edge computing node to obtain the runtime sequence dataset.

3. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 2, characterized in that, The steps to generate a set of feature vectors with uniform dimensions are as follows: The underlying electrical consumption sequence is transformed into an electrical feature vector through a linear mapping matrix; The surrounding environment sensor sequence and network link state sequence are transformed into environmental feature vectors and link feature vectors through a multilayer perceptron. The electrical feature vector, environmental feature vector, and link feature vector are concatenated, and a set of feature vectors with unified dimension is generated through orthogonal dimensionality reduction.

4. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 1, characterized in that, The spatiotemporal coupling analysis model includes a graph attention layer and a temporal convolutional layer. The steps for extracting temporal evolution correlation features are as follows: A graph attention layer is used to process a set of feature vectors of the same dimension to generate spatial graph association features; Temporal convolutional layers are used to perform sliding window convolution on spatial map association features in the temporal dimension to generate temporally evolved association features.

5. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 4, characterized in that, The steps for processing a set of feature vectors of uniform dimension using a graph attention layer include: Construct a spatial adjacency matrix based on the topological distances of heterogeneous devices within the target area; Using the spatial adjacency matrix as a mask, calculate the attention weight coefficient between any two feature vectors in a set of feature vectors of the same dimension; The feature vectors are weighted and aggregated based on attention weight coefficients to output spatial map association features.

6. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 5, characterized in that, The step of calculating the attention weight coefficient between any two feature vectors in a set of feature vectors with uniform dimensions also includes: Retrieve business scheduling data records for the target region; Encode business scheduling data records into business rhythm feature vectors; By incorporating the business rhythm feature vector as a bias term into the calculation of the attention weight coefficient, the attention weight coefficients between various heterogeneous devices are dynamically updated.

7. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 4, characterized in that, The steps of using temporal convolutional layers to perform sliding window convolution processing on spatial map association features in the temporal dimension include: Set up multiple one-dimensional convolutional kernels with different receptive field sizes; Utilize multiple one-dimensional convolutional kernels to scan historical time-step data of spatial map association features in parallel; Extract sequence change features at different time scales and concatenate these features into temporal evolution correlation features.

8. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 1, characterized in that, The steps for outputting the global anomaly risk index vector of the target region through the fully connected mapping layer within the spatiotemporal coupling analysis model include: Tensor multiplication is performed on spatial map correlation features and temporal evolution correlation features to obtain spatiotemporal fusion features; The spatiotemporal fusion features are input into a fully connected mapping layer with a nonlinear activation function; Output a global anomaly risk index vector in floating-point format.

9. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 8, characterized in that, The steps for outputting power consumption modulation commands for abnormal heterogeneous devices include: Multiply the global anomaly risk index vector with the equipment management strategy matrix to obtain the control strategy coefficient corresponding to each heterogeneous device; Match the target control code in the instruction code library according to the control strategy coefficient; The target control code is encapsulated as energy consumption modulation instructions.

10. The method for remote monitoring of instruments and equipment based on artificial intelligence according to claim 1, characterized in that, The steps of sending power consumption modulation commands to corresponding abnormal heterogeneous devices through edge computing nodes include: Edge computing nodes parse energy consumption modulation instructions to obtain target control code; Based on the communication protocol type of the abnormal heterogeneous devices, the target control code is converted into the corresponding underlying register operation message; The underlying register operation messages are sent and executed through heterogeneous devices.

Images