A remote transport monitoring method and device for chemical filters

Abstract

The invention relates to the technical field of remote transport monitoring, and discloses a remote transport monitoring method and device for a chemical filter, wherein the method comprises: sealing and packaging the chemical filter and installing an AMC monitoring module to obtain a target monitoring box; performing initialization and baseline data acquisition to obtain AMC concentration baseline data and warning thresholds; performing real-time data acquisition and hierarchical transmission to obtain AMC concentration real-time data and environmental and vibration real-time data; performing concentration change trend analysis and feature extraction to obtain an AMC concentration change feature set and an environmental and vibration parameter feature set; performing transport state evaluation through a dual-branch time series and anomaly detection fusion model to obtain a target transport state evaluation index; performing comparison, triggering a multi-level alarm and generating anomaly handling solutions and a transport quality evaluation report, wherein the method realizes remote transport monitoring of the chemical filter and improves the transport safety of the chemical filter.

Description

Technical Field

[0001] The present invention relates to the technical field of remote transport monitoring, and in particular to a remote transport monitoring method and device for a chemical filter. Background Art

[0002] Chemical filters are special filtering products that remove chemical pollutants from controlled environments. Currently, manufacturers on the market generally seal the filter products in plastic bags, then pack them in boxes and transport them to the project department for use. During the process from production to project installation and use, sealed packaging must be ensured to prevent air leakage from entering the carton and contaminating the chemical filter, resulting in reduced removal efficiency of the chemical filter or failure of the chemical filter.

[0003] How to ensure the sealing of chemical filters during long-distance transportation, prevent them from being damaged, improve the monitoring and traceability of transportation, and simplify the inspection process before installation, so as to ensure that the chemical filters can maintain their normal performance and quality when they arrive at the project department for use. At present, chemical filters on the market are generally packed in plastic bags and then transported in boxes. However, during long-distance transportation, various factors (such as vibration, temperature changes, packaging damage, etc.) may cause the packaging bags to break or the cartons to be loosely sealed, thereby allowing air to enter and contaminate the chemical filters, reducing their efficiency in removing chemical pollutants, and even causing the chemical filters to fail. During transportation, chemical filters may be physically damaged, such as collisions, extrusions, etc., which will also affect their performance. In the prior art, chemical filters lack effective monitoring and recording methods during transportation, and it is difficult to trace the cause once a problem occurs. At present, chemical filters need to undergo a series of inspections before installation to ensure their normal performance, which is time-consuming and labor-intensive. Summary of the invention

[0004] The invention provides a remote transportation monitoring method and device for a chemical filter, which are used to realize the remote transportation monitoring of the chemical filter and improve the transportation safety of the chemical filter.

[0005] In a first aspect, the present invention provides a remote transport monitoring method for a chemical filter, the remote transport monitoring method for a chemical filter comprising:

[0006] The chemical filter is sealed and packaged and the AMC monitoring module is installed to obtain the target monitoring box;

[0007] Initializing and collecting baseline data of the AMC concentration in the target monitoring box to obtain AMC concentration baseline data, and setting a warning threshold value of the target monitoring box;

[0008] Performing real-time data collection and hierarchical transmission on the target monitoring box to obtain real-time data of AMC concentration as well as real-time data of environment and vibration;

[0009] Performing concentration change trend analysis on the AMC concentration real-time data and the AMC concentration baseline data to obtain an AMC concentration change feature set, and performing feature extraction on the environment and vibration real-time data to obtain an environment and vibration parameter feature set;

[0010] Input the AMC concentration change feature set and the environment and vibration parameter feature set into a preset dual-branch time series and anomaly detection fusion model to perform transport state evaluation to obtain a target transport state evaluation index;

[0011] The target transport status assessment index and the warning threshold are compared to trigger a multi-level alarm and generate an abnormality handling plan and a transport quality assessment report.

[0012] In a second aspect, the present invention provides a remote transport monitoring device for a chemical filter, the remote transport monitoring device for a chemical filter comprising:

[0013] An installation module is used to seal and package the chemical filter and install the AMC monitoring module to obtain a target monitoring box;

[0014] An initialization module, used to initialize and collect benchmark data of the AMC concentration in the target monitoring box, obtain AMC concentration benchmark data, and set the warning threshold of the target monitoring box;

[0015] A collection module is used to collect and grade data of the target monitoring box in real time to obtain real-time data of AMC concentration and real-time data of environment and vibration;

[0016] An analysis module, configured to perform concentration change trend analysis on the AMC concentration real-time data and the AMC concentration baseline data to obtain an AMC concentration change feature set, and to perform feature extraction on the environment and vibration real-time data to obtain an environment and vibration parameter feature set;

[0017] An evaluation module, used for inputting the AMC concentration change feature set and the environment and vibration parameter feature set into a preset dual-branch time series and anomaly detection fusion model to perform transport state evaluation and obtain a target transport state evaluation index;

[0018] The generation module is used to compare the target transportation status evaluation index with the warning threshold, trigger multi-level alarms and generate an abnormality handling plan and a transportation quality evaluation report.

[0019] The third aspect of the present invention provides a remote transport monitoring device for a chemical filter, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the remote transport monitoring device for the chemical filter executes the above-mentioned remote transport monitoring method for the chemical filter.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned remote transport monitoring method for a chemical filter.

[0021] In the technical solution provided by the present invention, by integrating AMC sensors, temperature and humidity sensors and acceleration sensors, multi-dimensional monitoring of the chemical filter during transportation is realized, and the filter status and environmental changes are fully grasped. Adaptive sampling and filtering technology is adopted to dynamically adjust the sampling frequency according to the status evaluation results, and the original data is checked and processed in real time to ensure the accuracy and reliability of the monitoring data. A dual-branch fusion model is constructed by combining LSTM and CNN networks, and the temporal characteristics of AMC concentration and the spatial characteristics of environmental vibration are fully utilized to improve the recognition accuracy of abnormal states. Through a multi-level alarm mechanism and fuzzy reasoning, different levels of alarms are intelligently triggered according to the degree of deviation, and targeted abnormal handling solutions are generated to effectively reduce transportation risks. A hierarchical transmission strategy is adopted to perform adaptive transmission according to data importance and network conditions, reasonably allocate bandwidth resources, and improve data transmission efficiency. Through remote data transmission and cloud analysis, real-time monitoring and remote management of the transportation status of chemical filters are realized, and supervision efficiency is improved. By generating a transportation quality assessment report, including trend analysis and prediction results.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of an embodiment of a remote transport monitoring method for a chemical filter in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of an embodiment of a remote transport monitoring device for a chemical filter according to an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of an embodiment of a remote transport monitoring device for a chemical filter in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0029] To facilitate understanding of this embodiment, a remote transport monitoring method for a chemical filter disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0030] 101. The chemical filter is sealed and packaged and the AMC monitoring module is installed to obtain a target monitoring box;

[0031] It is understandable that the execution subject of the present invention may be a remote transport monitoring device for a chemical filter, or may be a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0032] Specifically, the chemical filter is sealed with an antistatic plastic bag and sealed with hot melt technology to ensure its sealing effect, thereby obtaining a sealed chemical filter. The sealed chemical filter is designed with multi-layer protection, and the outer layer is treated with materials that meet the IP67 protection level, so that it has strong waterproof and dustproof capabilities. At the same time, an anti-vibration pad is set in the inner layer to effectively resist the vibration and impact that may be received during transportation, thereby obtaining a multi-layer protection chemical filter. According to the size of the multi-layer protection chemical filter, the monitoring box is customized to ensure that the internal space of the monitoring box can fully match the size and shape of the filter to avoid shaking and damage caused by inappropriate size, thereby obtaining a special monitoring box. The AMC monitoring module is installed on the special monitoring box, and the AMC sensor, temperature and humidity sensor, and acceleration sensor are integrated into the inner side of the monitoring box to form a complete sensor integrated monitoring box. The data acquisition and transmission module is installed on the sensor integrated monitoring box, and a large-capacity storage and a high-speed wireless communication module are integrated to enable it to have powerful data processing and real-time transmission capabilities, thereby obtaining a data processing monitoring box. According to the requirements of AMC concentration monitoring, the AMC sensor is calibrated to ensure that it can accurately detect the AMC concentration in the environment. The monitoring accuracy is set to ±0.1ppm to meet the high-precision monitoring requirements and obtain a high-precision AMC monitoring module. In order to adapt to different monitoring states, the sampling frequency of the high-precision AMC monitoring module is set. Under normal conditions, the sampling frequency is set to once every 5 minutes, and when an abnormal state is detected, the sampling frequency is set to once every 1 minute, forming an adaptive sampling AMC monitoring module that can dynamically adjust the sampling frequency according to actual conditions to improve the flexibility and response speed of monitoring. At the same time, according to the characteristics of the transportation environment, the temperature and humidity sensors and acceleration sensors are configured with parameters, and the sampling range and frequency of temperature, humidity and vibration intensity are set to ensure the comprehensiveness and accuracy of environmental data, and obtain an environmental parameter monitoring module. The data fusion algorithm is designed for the adaptive sampling AMC monitoring module and the environmental parameter monitoring module, and time synchronization and multi-sensor data fusion technology are used to integrate data from different sensors to form a comprehensive data monitoring system. The time synchronization technology is used to ensure that the data of each sensor is compared and analyzed on the same time axis to improve the accuracy and consistency of data analysis. The multi-layer protective chemical filter is placed in the integrated data monitoring system, and the system is sealed to ensure that it will not be contaminated or damaged by the external environment during transportation. At the same time, functional tests are carried out to confirm that its monitoring function is normal and stable, and finally the target monitoring box is obtained.

[0033] 102. Initialize and collect benchmark data of the AMC concentration in the target monitoring box, obtain the AMC concentration benchmark data, and set the warning threshold of the target monitoring box;

[0034] Specifically, the target monitoring box is preheated and the AMC monitoring module is run for 24 hours to ensure that the system reaches a stable state and obtain a stable AMC monitoring system. The stable AMC monitoring system is sampled at high frequency, and AMC concentration data is collected once a minute for 24 hours to obtain a comprehensive original AMC concentration data set. The original AMC concentration data set is detected for outliers, and possible abnormal data points are removed to obtain a cleaned AMC concentration data set, and the cleaned AMC concentration data set is decomposed into a time series to extract the time series characteristics of the AMC concentration. According to the time series characteristics of the AMC concentration, the cleaned AMC concentration data set is subjected to exponential weighted moving average processing to smooth data fluctuations and obtain the AMC concentration benchmark data. The AMC concentration benchmark data is statistically analyzed to determine the distribution characteristics of the AMC concentration. Based on the AMC concentration distribution characteristics, the warning threshold is dynamically set to obtain the initial threshold. In order to ensure the rationality and effectiveness of the warning threshold, a sensitivity analysis is performed, and the threshold-performance curve is drawn by simulating the false alarm rate and false alarm rate under different thresholds. By analyzing the threshold-performance curve, the balance point between the false alarm rate and the missed alarm rate is found, the initial threshold is optimized and adjusted, and the most suitable warning threshold is determined. The AMC concentration baseline data and the optimized warning threshold are stored in the local storage of the target monitoring box and the remote server to complete the initialization of the AMC monitoring system.

[0035] 103. Collect and transmit real-time data of the target monitoring box in a graded manner to obtain real-time data of AMC concentration as well as real-time data of environment and vibration;

[0036] Specifically, the AMC sensor, temperature and humidity sensor and acceleration sensor in the target monitoring box are synchronously triggered to form a multi-sensor synchronous sampling system to ensure that all sensors can collect data at the same time point and obtain time-consistent data. According to the preset sampling frequency of the normal state, the multi-sensor synchronous sampling system is sampled at intervals of 5 minutes to obtain the original multi-dimensional sensor data stream. The collected raw data needs to be checked for data quality in real time, and the data quality index is obtained by calculating the data integrity, validity and consistency. According to the data quality index, the original multi-dimensional sensor data stream is adaptively filtered to remove noise and outliers to obtain the filtered multi-dimensional sensor data. The filtered multi-dimensional sensor data is feature extracted to calculate the mean, standard deviation and change rate of the AMC concentration, the mean of the temperature and humidity and the peak acceleration of the vibration to form a multi-dimensional feature vector. The feature vector is used to quickly evaluate the current state to determine whether there is an abnormal situation during the transportation process and obtain the state evaluation result. According to the state evaluation result, the sampling frequency is increased to once per minute if it is determined to be an abnormal state, and the environmental changes are monitored at a higher frequency to obtain the adjusted sampling strategy. According to the adjusted sampling strategy, the multi-sensor synchronous sampling system is reconfigured and the sampling parameters of the sensors are updated to ensure that the rapid changes in the environment can be captured in time. After the data collection is completed, the collected data is packaged and processed in a hierarchical manner. For the data in the normal state, compression encoding is performed to save storage space and transmission bandwidth, while for the data in the abnormal state, the original accuracy is maintained to ensure that no details are missed, forming a hierarchical data packet. According to the network status and data priority, the hierarchical data packets are adaptively transmitted, and the data is efficiently transmitted to the remote server using a dynamically adjusted sending frequency to ensure that key data can be delivered in time. Data in the normal state can be transmitted at a lower frequency to reduce the network load, while data in the abnormal state needs to be transmitted frequently so that the remote server can respond and process in time, and finally obtain the real-time data of AMC concentration as well as the real-time data of environment and vibration.

[0037] 104. Perform concentration change trend analysis on the AMC concentration real-time data and the AMC concentration benchmark data to obtain an AMC concentration change feature set, and perform feature extraction on the environment and vibration real-time data to obtain an environment and vibration parameter feature set;

[0038] Specifically, the real-time data of AMC concentration is decomposed into time series, and the AMC concentration data is decomposed into multiple components of different scales by wavelet decomposition or empirical mode decomposition, so as to obtain the multi-scale components of AMC concentration. According to the multi-scale components, the trend component is analyzed in detail, and the slope and intercept are calculated by linear regression analysis to obtain the AMC concentration change rate index, which reflects the speed and direction of the change of AMC concentration. The dynamic time warping distance is calculated for the real-time data of AMC concentration and the benchmark data of AMC concentration. Dynamic time warping is an algorithm to measure the similarity of two time series, which can effectively calculate the degree of deviation between real-time data and benchmark data, and obtain the degree of deviation of AMC concentration from the benchmark. The real-time data of AMC concentration is subjected to sliding window statistical analysis, and the statistical characteristics of the data such as mean, variance, skewness, kurtosis, etc. are calculated in the sliding window to form the statistical characteristic vector of AMC concentration. At the same time, the real-time data of AMC concentration is subjected to Fourier transform, and the time domain data is converted into frequency domain data, and the spectrum characteristics of AMC concentration are extracted to reveal the periodic changes and frequency components in the data. Based on the above characteristics, a complete AMC concentration change feature set is constructed according to the AMC concentration change rate index, the AMC concentration deviation from the benchmark index, the AMC concentration statistical feature vector and the AMC concentration spectrum feature. At the same time, the real-time data of environment and vibration are analyzed. The real-time data of environmental temperature and humidity are subjected to correlation analysis, the Pearson correlation coefficient of temperature and humidity is calculated, and the linear correlation between them is evaluated; at the same time, the degree of information coupling between them is calculated using mutual information to obtain the temperature and humidity coupling feature. According to the vibration data in the real-time data of environment and vibration, the time domain features and frequency domain features are calculated respectively. In the time domain features, the mean, variance, peak value and RMS value of the vibration signal are calculated, while in the frequency domain features, the main frequency components and spectrum energy distribution are extracted by fast Fourier transform to obtain the vibration intensity and frequency features, which reflect the intensity and frequency changes of the vibration signal and provide effective information support for monitoring the vibration situation during transportation. The real-time data of environmental temperature and humidity and the real-time data of vibration are subjected to cross-correlation analysis, the correlation coefficients under different time delays are calculated, the relationship between the temperature and humidity changes and the vibration intensity is evaluated, and the environment-vibration interaction features are obtained. Through analysis, the impact of environmental changes on vibration signals is identified. Based on the temperature and humidity coupling characteristics, vibration intensity and frequency characteristics, and environment-vibration interaction characteristics, a comprehensive set of environmental and vibration parameter features is constructed.

[0039] 105. Input the AMC concentration change feature set and the environment and vibration parameter feature set into the preset dual-branch time series and anomaly detection fusion model to evaluate the transportation state, and obtain the target transportation state evaluation index;

[0040] Specifically, the AMC concentration change feature set is subjected to feature normalization. The feature data is scaled to the same range by the normalization method to eliminate the influence of the dimension, and the normalized AMC feature vector is obtained. The feature set of the environment and vibration parameters is subjected to feature standardization. The feature data is converted into a standard normal distribution with a mean of zero and a variance of one by the normalization method, and the standardized environmental vibration feature vector is obtained. A long short-term memory network (LSTM) is constructed as the time series branch. The network contains an input layer, two LSTM layers and a fully connected layer. The LSTM layer can capture the long-term dependencies in the time series data. The tanh activation function is used to ensure the nonlinear transformation of the data, and the AMC time series feature extractor is obtained. At the same time, a convolutional neural network (CNN) is constructed as the spatial feature branch. The network contains three two-dimensional convolutional layers and a global average pooling layer. The convolutional layer can extract the spatial features of the data. The ReLU activation function is used to improve the nonlinear expression ability of the model, and the environmental vibration spatial feature extractor is obtained. The normalized AMC feature vector is input into the AMC time series feature extractor. Through the forward propagation process, the LSTM layer processes the input data layer by layer and extracts a 128-dimensional AMC time series feature vector. The standardized environmental vibration feature vector is input into the environmental vibration spatial feature extractor. Through the forward propagation process, the convolution layer extracts features layer by layer to obtain a 64-dimensional environmental vibration spatial feature vector. The AMC time series feature vector and the environmental vibration spatial feature vector are concatenated to obtain a 192-dimensional multimodal feature vector. A three-layer fully connected neural network is constructed as a fusion module. The number of hidden layer units of the network is 128 and 64 respectively. The ReLU activation function is used to increase the nonlinear ability of the model. The multimodal feature vector is input into the fusion module. Through the forward propagation process, the data is processed layer by layer to finally obtain a 64-dimensional fusion feature representation. In order to achieve a comprehensive evaluation of the transport status, a dual-output layer structure is designed, in which the anomaly detection output layer uses the Sigmoid activation function, which can map the output value to the [0, 1] interval to represent the probability of anomaly occurrence; and the AMC concentration prediction output layer uses a linear activation function to directly predict the value of the AMC concentration. The fused feature representation is input into the dual-output layer structure, and the anomaly probability and predicted AMC concentration value are obtained through forward propagation. The weighted summation method is used to combine the anomaly probability and the predicted AMC concentration value to obtain the target transport status evaluation index. Given a certain weight, the anomaly probability and the predicted AMC concentration value are combined, and a comprehensive evaluation index is generated through weighted summation calculation, which can comprehensively reflect the status of the chemical filter during transportation.

[0041] The normalized AMC feature vector is time-step expanded to convert it into an input sequence of T time steps, where each time step contains N-dimensional features. Based on the input sequence of T time steps, the forget gate of the first LSTM layer is calculated. The forget gate determines how much information will be retained in the current time step, and the forget gate output is calculated as a function of the current input and the hidden state of the previous time step. After the forget gate output is calculated, the input gate of the first LSTM layer is calculated based on the input sequence of T time steps and the forget gate output. The input gate controls how much new information is written to the cell state, and its output also depends on the current input and the hidden state of the previous time step. Combined with the input sequence of T time steps, the forget gate output, and the input gate output, the candidate cell state of the first LSTM layer is calculated. Through nonlinear transformation, the candidate new cell state value is generated. After obtaining the candidate cell state, the cell state of the first LSTM layer is updated according to the forget gate output, the input gate output, and the candidate cell state. The new cell state is obtained by multiplying the cell state of the previous time step by the forget gate output, plus the input gate output multiplied by the candidate cell state. The output gate of the first LSTM layer is calculated based on the input sequence of T time steps and the updated cell state. The output gate controls how much of the cell state will be passed to the hidden state, and its calculation depends on the current input and the updated cell state. According to the output of the output gate and the updated cell state, the hidden state of the first LSTM layer is updated to obtain the updated hidden state. The hidden state is the output of the LSTM layer and also part of the input of the next time step. After the calculation of the first LSTM layer is completed, its 256-dimensional output is input to the second LSTM layer for processing. The second LSTM layer further extracts time series features from the input through a calculation process similar to the first LSTM layer, and finally obtains a 192-dimensional second LSTM layer output. This process captures higher-level temporal dependencies and features by stacking multiple layers of LSTM. According to the 192-dimensional second LSTM layer output, it is input to the fully connected layer for linear transformation. The fully connected layer generates intermediate feature representations through weighted summation and bias. On this basis, the activation function is applied to the intermediate features for nonlinear transformation to obtain a 128-dimensional AMC time series feature vector. The feature vector integrates the temporal information of the input sequence at multiple time steps and different levels, and can effectively characterize the dynamic changes of AMC concentration.

[0042] The standardized environmental vibration feature vector is reshaped to be converted into a four-dimensional tensor input suitable for convolutional neural network processing, ensuring that the shape of the input data meets the requirements of the convolution layer. Usually, the one-dimensional feature vector is reshaped into a two-dimensional or three-dimensional feature map of appropriate size for subsequent convolution operations. The reshaped four-dimensional tensor is input to the first convolution layer of the environmental vibration spatial feature extractor for convolution operation. The convolution operation extracts local spatial features by applying multiple filters to the input tensor to obtain the output feature map of the first convolution layer. The feature map contains low-level features of the input data, such as edges and textures. The output feature map of the first convolution layer is batch normalized to stabilize and accelerate the network training process to obtain the normalized first-layer feature map. Batch normalization improves the robustness of the model by standardizing the output of each small batch to have zero mean and unit variance. The normalized first-layer feature map is subjected to maximum pooling to obtain the output feature map of the first pooling layer. Max pooling reduces the size of the feature map by taking the maximum value in the local area, while retaining important spatial features, reducing the amount of calculation and the risk of overfitting. The output feature map of the first pooling layer is input to the second convolutional layer, and a second convolution operation is performed to obtain the output feature map of the second convolutional layer. Through deeper convolution operations, higher-level features of the input data, such as shape and pattern, are extracted. The output feature map of the second convolutional layer is subjected to a second batch normalization process to obtain a normalized second-layer feature map. Normalization again helps stabilize the network training process and ensure that the numerical range of the features is consistent. The second max pooling operation is performed on the normalized second-layer feature map to obtain the output feature map of the second pooling layer. Through further pooling operations, the size of the feature map is reduced while retaining key information. The output feature map of the second pooling layer is input to the third convolutional layer, and a third convolution operation is performed to obtain the output feature map of the third convolutional layer. As the number of network layers increases, the features extracted by the convolution operation become more and more abstract and advanced. The output feature map of the third convolutional layer is subjected to global average pooling to obtain a 128-dimensional global feature vector. Global average pooling obtains a global feature representation by averaging all values ​​of each feature map, significantly reducing the number of parameters and the risk of overfitting. The 128-dimensional global feature vector is input to the fully connected layer for linear transformation and activation function processing to obtain a 64-dimensional environmental vibration spatial feature vector. The fully connected layer integrates and refines features through linear transformation and nonlinear activation functions to form the final feature representation.

[0043] 106. Compare the target transport status assessment indicators and warning thresholds, trigger multi-level alarms and generate exception handling plans and transport quality assessment reports.

[0044] Specifically, the target transportation status evaluation index is normalized and standardized to a unified scale to obtain a normalized evaluation index. According to the dynamic changes of the normalized evaluation index, the warning threshold is adjusted in real time to obtain an adaptive warning threshold. The dynamic adjustment method can flexibly set the warning threshold according to the changes in the actual monitoring data, thereby improving the sensitivity and accuracy of the monitoring system. The difference between the normalized evaluation index and the adaptive warning threshold is calculated to obtain the degree of deviation. The fuzzy output is defuzzified to generate a fuzzy deviation index. Converting specific values ​​into fuzzy sets can effectively process uncertainty and fuzzy data and improve the robustness of the analysis. According to the fuzzy deviation index, the preset multi-level alarm rules are inferred to obtain the fuzzy output of the alarm level. The multi-level alarm rules set different alarm levels according to different degrees of deviation to effectively distinguish different abnormal situations in the transportation process. The fuzzy output is defuzzified to obtain a specific multi-level alarm result. Defuzzification converts fuzzy sets into specific values, making the alarm level more intuitive and easy to understand. According to the multi-level alarm results, case matching is performed from the preset exception handling strategy library, and the preliminary exception handling suggestions corresponding to the current alarm level are selected. The handling strategy library stores response plans for different abnormal situations, which can quickly provide effective handling suggestions. The preliminary exception handling suggestions are context-adapted, and the preliminary suggestions are adjusted and optimized in combination with the current transportation environment and specific conditions to obtain the final exception handling plan. At the same time, according to the time series data of the target transportation status evaluation indicators, the transportation quality trend is analyzed to predict the change trend of future transportation quality. Through the time series analysis method, potential problems and risks in the transportation process are identified, and preventive measures are taken in advance. The multi-level alarm results, exception handling plans and transportation quality prediction results are integrated to generate a transportation quality evaluation report. The evaluation report records the current transportation status, exception handling plan and future quality prediction.

[0045] In the embodiment of the present invention, by integrating AMC sensors, temperature and humidity sensors and acceleration sensors, multi-dimensional monitoring of the chemical filter during transportation is realized, and the filter status and environmental changes are fully grasped. Adaptive sampling and filtering technology is adopted to dynamically adjust the sampling frequency according to the status evaluation results, and the original data is checked and processed in real time to ensure the accuracy and reliability of the monitoring data. A dual-branch fusion model is constructed by combining LSTM and CNN networks, and the temporal characteristics of AMC concentration and the spatial characteristics of environmental vibration are fully utilized to improve the recognition accuracy of abnormal states. Through a multi-level alarm mechanism and fuzzy reasoning, different levels of alarms are intelligently triggered according to the degree of deviation, and targeted abnormal handling solutions are generated to effectively reduce transportation risks. A hierarchical transmission strategy is adopted to perform adaptive transmission according to data importance and network conditions, reasonably allocate bandwidth resources, and improve data transmission efficiency. Through remote data transmission and cloud analysis, real-time monitoring and remote management of the transportation status of chemical filters are realized, and supervision efficiency is improved. By generating a transportation quality assessment report, including trend analysis and prediction results.

[0046] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0047] The chemical filter is sealed with an antistatic plastic bag and sealed with hot-melt technology to obtain a sealed chemical filter. The sealed chemical filter is designed with multiple layers of protection, with IP67 protection grade materials used on the outer layer and shockproof pads set on the inner layer to obtain a multi-layer protection chemical filter.

[0048] According to the size of the multi-layer protective chemical filter, the monitoring box is customized to obtain a special monitoring box, and the AMC monitoring module is installed in the special monitoring box. The AMC sensor, temperature and humidity sensor and acceleration sensor are integrated into the inner side of the monitoring box to obtain a sensor integrated monitoring box;

[0049] Install data acquisition and transmission modules on the sensor integrated monitoring box, integrate large-capacity storage and high-speed wireless communication modules to obtain a data processing monitoring box;

[0050] According to the AMC concentration monitoring requirements, the AMC sensor is calibrated and the monitoring accuracy is set to ±0.1ppm to obtain a high-precision AMC monitoring module;

[0051] The sampling frequency of the high-precision AMC monitoring module is set to once every 5 minutes under normal conditions and once every 1 minute under abnormal conditions, thereby obtaining an adaptive sampling AMC monitoring module;

[0052] According to the characteristics of the transportation environment, the temperature and humidity sensors and acceleration sensors are configured with parameters, and the sampling range and frequency of temperature, humidity and vibration intensity are set to obtain the environmental parameter monitoring module;

[0053] The data fusion algorithm is designed for the adaptive sampling AMC monitoring module and the environmental parameter monitoring module, and the time synchronization and multi-sensor data fusion technology are used to obtain a comprehensive data monitoring system;

[0054] Place the multi-layer protective chemical filter into the integrated data monitoring system, and perform sealing and functional tests on the integrated data monitoring system to obtain the target monitoring box.

[0055] Specifically, the chemical filter is sealed with an antistatic plastic bag, which can effectively prevent static electricity from damaging the chemical filter. The hot melt technology is used for sealing, and the edge of the plastic bag is melted and bonded by heating to ensure the sealing effect. After the sealing process is completed, a sealed chemical filter is obtained. The sealed chemical filter is designed with multiple layers of protection. In the outer layer, it is treated with materials that meet the IP67 protection level. This material not only has the functions of dustproof and waterproof, but also can withstand the influence of harsh environments. The inner layer is provided with a shockproof pad, which can absorb and buffer the vibration during transportation, ensure the stability and safety of the chemical filter, and obtain a multi-layer protection chemical filter. According to the size of the multi-layer protection chemical filter, the monitoring box is customized to ensure that the monitoring box can perfectly adapt to the size and shape of the chemical filter, and avoid shaking and damage caused by inappropriate space during transportation. After completing the customized design, a special monitoring box is obtained. On the inner side of the special monitoring box, an AMC monitoring module is installed, including an AMC sensor, a temperature and humidity sensor, and an acceleration sensor, to monitor the air quality, ambient temperature and humidity, and vibration around the chemical filter in real time, and obtain a sensor integrated monitoring box. The data acquisition and transmission modules are installed on the sensor integrated monitoring box. In order to ensure the reliable storage and transmission of data, a large-capacity memory and a high-speed wireless communication module are integrated. The large-capacity memory can store a large amount of sensor data, while the high-speed wireless communication module ensures that the data can be transmitted to the remote server in real time, and a data processing monitoring box is obtained. According to the AMC concentration monitoring requirements, the AMC sensor is calibrated for accuracy. By setting the monitoring accuracy to ±0.1ppm, it is ensured that the sensor can accurately detect the AMC concentration in the air, and a high-precision AMC monitoring module is obtained. In order to improve the flexibility and response speed of the monitoring system, the sampling frequency of the high-precision AMC monitoring module is set. Under normal conditions, the sampling frequency is set to once every 5 minutes to ensure regular data updates. When an abnormal state is detected, the sampling frequency is set to once every 1 minute to capture environmental changes in time, and an adaptive sampling AMC monitoring module is obtained. According to the characteristics of the transportation environment, the temperature and humidity sensors and acceleration sensors are configured with parameters, and the sampling range and frequency of temperature, humidity and vibration intensity are set to obtain an environmental parameter monitoring module. Data fusion algorithms are designed for the adaptive sampling AMC monitoring module and the environmental parameter monitoring module. By adopting time synchronization and multi-sensor data fusion technology, comprehensive analysis of multi-source data is achieved. Multi-layer protective chemical filters are placed in the comprehensive data monitoring system, and the comprehensive data monitoring system is sealed to ensure that it will not be contaminated or damaged by the external environment during transportation. In order to verify the function and performance of the system, a series of functional tests are carried out. These tests include simulating various transportation environments to ensure that the system can operate normally under different conditions and obtain the target monitoring box.

[0056] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0057] Preheat the target monitoring box and run the AMC monitoring module for 24 hours to obtain a stable AMC monitoring system;

[0058] The AMC monitoring system in a stable state is sampled at high frequency, and AMC concentration data is collected once a minute for 24 hours to obtain the original AMC concentration data set;

[0059] Perform outlier detection on the original AMC concentration data set to obtain a cleaned AMC concentration data set, and perform time series decomposition on the cleaned AMC concentration data set to obtain the AMC concentration time series features;

[0060] According to the time series characteristics of AMC concentration, the cleaned AMC concentration data set is processed by exponential weighted moving average to obtain the AMC concentration benchmark data, and the AMC concentration benchmark data is statistically analyzed to obtain the AMC concentration distribution characteristics;

[0061] According to the AMC concentration distribution characteristics, the warning threshold is dynamically set to obtain the initial threshold, and the sensitivity analysis of the initial threshold is performed to simulate the false alarm rate and missed alarm rate under different thresholds to obtain the threshold-performance curve;

[0062] According to the threshold-performance curve, the initial threshold is optimized and adjusted, and the balance point between the false alarm rate and the missed alarm rate is selected to obtain the warning threshold;

[0063] The AMC concentration baseline data and warning threshold are stored in the local storage of the target monitoring box and the remote server to obtain an initialized AMC monitoring system.

[0064] Specifically, the target monitoring box is initialized to ensure that all sensors and data acquisition modules work properly. In this stage, the AMC monitoring module is run for 24 hours to gradually stabilize the system and obtain a stable AMC monitoring system. During this period, the internal sensors are self-checked and calibrated to ensure that they can accurately and stably monitor the AMC concentration. The AMC monitoring system in a stable state is sampled at high frequency, and AMC concentration data is collected once a minute for 24 hours to obtain a detailed original AMC concentration data set. This process will generate a large amount of data, which will help analyze the long-term stability of the system and the changes in AMC concentration in the environment. The original AMC concentration data set is detected for outliers by statistical methods or machine learning algorithms. For example, the standard deviation method is used to detect outliers: for each data point, if its value deviates from the mean by more than a certain multiple of the standard deviation, it is considered an outlier. Assuming that the mean of the AMC concentration data set is μ and the standard deviation is σ, for a data point x, if |x-μ|>kσ (where k is a constant), x is an outlier. In this way, the noise and outliers in the data are cleaned up to obtain a cleaned AMC concentration data set. The cleaned AMC concentration data set is subjected to time series decomposition to extract the time series characteristics of AMC concentration. Time series decomposition can use methods such as empirical mode decomposition or wavelet transform to decompose the time series into different frequency components to better understand the periodicity and trend changes of the data. Through decomposition, characteristics of different time scales are obtained to provide a basis for subsequent analysis. According to the time series characteristics of AMC concentration, the cleaned AMC concentration data set is subjected to exponentially weighted moving average processing to obtain the AMC concentration benchmark data. The exponentially weighted moving average formula is as follows:

[0065] ;

[0066] in, represents the smoothed value at the current moment, Indicates the actual value at the current moment. is the smoothing coefficient, which is usually between 0.1 and 0.3. Through exponentially weighted moving average processing, random fluctuations in the data are smoothed and long-term trends are extracted. The AMC concentration benchmark data is statistically analyzed to obtain the distribution characteristics of the AMC concentration. By calculating statistics such as mean, variance, skewness and kurtosis, the distribution of AMC concentration is fully described. According to the statistical characteristics, the warning threshold is dynamically set to obtain the initial threshold. The initial threshold can be set according to historical data and distribution characteristics to ensure that it is within the normal range. Sensitivity analysis is performed to simulate the false alarm rate and false negative rate under different thresholds, and the threshold-performance curve is drawn. The false alarm rate refers to the probability of triggering an alarm under normal circumstances, and the false negative rate refers to the probability of not triggering an alarm under abnormal circumstances. By adjusting the threshold, different combinations of false alarm rate and false negative rate are obtained to form a threshold-performance curve. According to this curve, a suitable balance point is selected to achieve the optimal balance between the false alarm rate and the false negative rate, and the initial threshold is optimized. According to the threshold-performance curve, the initial threshold is optimized and adjusted, the balance point of the false alarm rate and the false negative rate is selected, and the warning threshold is finally determined. The optimization process ensures that the system can effectively detect abnormal situations and reduce unnecessary false alarms in actual applications, thereby improving the reliability and accuracy of the monitoring system. The AMC concentration baseline data and the optimized warning threshold are stored in the local storage of the target monitoring box and the remote server to complete the initialization of the AMC monitoring system. This ensures the security and traceability of the data, and can also be centrally managed and analyzed on the remote server.

[0067] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0068] The AMC sensor, temperature and humidity sensor and acceleration sensor in the target monitoring box are synchronously triggered to obtain a multi-sensor synchronous sampling system;

[0069] According to the preset sampling frequency of the normal state, the multi-sensor synchronous sampling system is sampled at a 5-minute interval to obtain the original multi-dimensional sensor data stream;

[0070] Performing real-time data quality check on the original multi-dimensional sensor data stream to obtain data quality indicators, and adaptively filtering the original multi-dimensional sensor data stream according to the data quality indicators to obtain filtered multi-dimensional sensor data;

[0071] Perform feature extraction on the filtered multidimensional sensor data, calculate the mean, standard deviation, and change rate of AMC concentration, as well as the mean of temperature and humidity and the peak acceleration of vibration, and obtain a multidimensional feature vector;

[0072] According to the multi-dimensional feature vector, the current state is quickly evaluated to obtain the state evaluation result, and the state evaluation result is graded. If it is determined to be an abnormal state, the sampling frequency is increased to once per minute to obtain the adjusted sampling strategy;

[0073] According to the adjusted sampling strategy, the multi-sensor synchronous sampling system is reconfigured to obtain updated sensor sampling parameters;

[0074] The collected data is packaged in different levels, the normal state data is compressed and encoded, and the abnormal state data is kept with original accuracy to obtain a graded data packet;

[0075] According to the network status and data priority, the graded data packets are adaptively transmitted, and the data is transmitted to the remote server using a dynamically adjusted sending frequency to obtain real-time AMC concentration data as well as environmental and vibration data.

[0076] Specifically, a multi-sensor synchronous sampling system is constructed. The system triggers each sensor through a unified clock signal, so that it can collect data at the same time point to ensure the time consistency of the data. According to the preset sampling frequency of the normal state, the multi-sensor synchronous sampling system is sampled at intervals of 5 minutes to obtain the original multi-dimensional sensor data stream. In each 5-minute interval, AMC concentration, temperature, humidity and vibration data are collected simultaneously, and these data are packaged into a multi-dimensional data point to form a continuous original data stream. In order to ensure the reliability of the data, the original multi-dimensional sensor data stream is checked for real-time data quality. The data quality check identifies possible problematic data points by detecting indicators such as data integrity, validity and consistency. For example, by calculating the mean and standard deviation of each data point, the abnormal deviation value is detected. Assuming that the normal data range of a sensor is μ±3σ, if a data point x exceeds this range, it is considered to be abnormal data. According to the data quality indicators, the original multi-dimensional sensor data stream is adaptively filtered to remove noise and outliers to obtain the filtered multi-dimensional sensor data. Feature extraction is performed on the filtered multi-dimensional sensor data. Calculate the mean, standard deviation and rate of change of AMC concentration. These features can describe the basic statistical characteristics and change trends of AMC concentration. At the same time, calculate the mean of temperature and humidity to reflect the basic state of the environment. For vibration data, calculate its peak acceleration to reflect the maximum impact force during transportation. Through feature extraction, a multidimensional feature vector containing AMC concentration, temperature and humidity, and vibration characteristics is obtained. According to the multidimensional feature vector, a quick assessment of the current state is performed. During the assessment process, the current features are compared with the features of historical normal data to determine whether the current state is abnormal. Assume that the multidimensional feature vector of the current state is x, and the mean vector of historical normal data is , the covariance matrix is , the state deviation degree can be calculated by Mahalanobis distance:

[0077] ;

[0078] If the Mahalanobis distance exceeds the set threshold, the current state is determined to be abnormal. The state assessment results are graded. If it is determined to be an abnormal state, the sampling frequency is increased to once a minute, and the environmental changes are monitored at a higher frequency to obtain the adjusted sampling strategy. The multi-sensor synchronous sampling system needs to be reconfigured according to the new sampling strategy, and the sampling parameters of the sensor are adjusted to ensure that more environmental data can be captured in time. During the data collection process, the collected data is graded and packaged. For the data in the normal state, compression encoding is performed to reduce the data volume and transmission bandwidth. For the data in the abnormal state, the original accuracy is maintained to ensure that no detail information is lost, and a graded data packet is obtained. The graded data packet is adaptively transmitted according to the network status and data priority. The data is transmitted to the remote server through the dynamically adjusted sending frequency. The data in the normal state can be transmitted at a lower frequency to save network resources; while the data in the abnormal state needs to be transmitted frequently so that the remote server can respond and process in time. In this way, the real-time data of AMC concentration and the real-time data of environment and vibration are finally obtained.

[0079] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0080] The real-time data of AMC concentration is decomposed into time series to obtain multi-scale components of AMC concentration. Based on the multi-scale components of AMC concentration, linear regression analysis is performed on the trend component to calculate the slope and intercept to obtain the AMC concentration change rate index.

[0081] Perform dynamic time warping distance calculation on the AMC concentration real-time data and the AMC concentration benchmark data to obtain the AMC concentration deviation from the benchmark index;

[0082] According to the real-time data of AMC concentration, the statistical characteristics in the sliding window are calculated to obtain the statistical characteristic vector of AMC concentration, and the real-time data of AMC concentration is subjected to Fourier transform to obtain the spectrum characteristics of AMC concentration;

[0083] According to the AMC concentration change rate index, the AMC concentration deviation from the benchmark index, the AMC concentration statistical feature vector and the AMC concentration spectrum feature, an AMC concentration change feature set is constructed;

[0084] Perform correlation analysis on the ambient temperature and humidity real-time data in the environment and vibration real-time data, calculate the Pearson correlation coefficient and mutual information of temperature and humidity, obtain the temperature and humidity coupling characteristics, and calculate the time domain characteristics and frequency domain characteristics based on the vibration real-time data in the environment and vibration real-time data to obtain the vibration intensity and frequency characteristics;

[0085] Perform cross-correlation analysis on the real-time data of ambient temperature and humidity and the real-time data of vibration, calculate the correlation coefficients under different time delays, and obtain the environment-vibration interaction characteristics;

[0086] Based on the temperature and humidity coupling characteristics, vibration intensity and frequency characteristics, and environment-vibration interaction characteristics, an environment and vibration parameter feature set is constructed.

[0087] Specifically, the real-time data of AMC concentration is decomposed into a time series. Time series decomposition can be accomplished through methods such as wavelet transform or empirical mode decomposition. For example, using wavelet transform, the time series is decomposed into multi-scale components of different frequency components, which represent different characteristics in the original signal. By analyzing the multi-scale components of AMC concentration, a linear regression analysis is performed on the trend component, and the slope and intercept of the trend component are calculated to obtain the AMC concentration change rate index. Assuming that the time series of the trend component is y(t), use the linear regression model:

[0088] ;

[0089] Among them, a is the slope, which represents the rate of change of AMC concentration over time, and b is the intercept, which represents the initial AMC concentration. The slope a and intercept b are solved by the least squares method to determine the change trend of AMC concentration. Dynamic time warping distance calculation is performed on the AMC concentration real-time data and AMC concentration benchmark data. Dynamic time warping is an algorithm to measure the similarity of two time series. Through the dynamic programming method, the two time series are effectively aligned, the minimum distance between them is calculated, and the degree of deviation of AMC concentration from the benchmark is obtained. Assume that the real-time data series is And the benchmark data series is , the DTW distance can be expressed as:

[0090] ;

[0091] in, Represents a sequence element and The distance between them. Calculate the statistical features within the sliding window. By setting a sliding window, the statistical features such as the mean, standard deviation, maximum value, and minimum value of the data in the window are calculated to form the statistical feature vector of the AMC concentration. At the same time, the real-time data of AMC concentration is Fourier transformed, and the time domain data is converted into frequency domain data to obtain the spectral characteristics of the AMC concentration, revealing the periodic changes and frequency components in the data. Based on these analysis results, including the AMC concentration change rate index, the AMC concentration deviation from the benchmark index, the AMC concentration statistical feature vector, and the AMC concentration spectrum characteristics, a comprehensive AMC concentration change feature set is constructed to comprehensively reflect the changes in AMC concentration during transportation. Perform correlation analysis on the real-time data of ambient temperature and humidity, and calculate the Pearson correlation coefficient and mutual information of temperature and humidity. The Pearson correlation coefficient measures the linear relationship between two variables, and the mutual information measures the degree of information sharing between them, thereby obtaining the temperature and humidity coupling characteristics. Assume that the temperature series is , the humidity series is , the calculation formula of Pearson correlation coefficient r is:

[0092] ;

[0093] in, and are the mean values ​​of temperature and humidity, respectively. According to the vibration data in the environmental and vibration real-time data, the time domain features and frequency domain features are calculated. In the time domain features, the mean, variance, peak value and RMS value of the vibration signal are calculated, while in the frequency domain features, the main frequency components and spectrum energy distribution are extracted by fast Fourier transform to obtain the vibration intensity and frequency characteristics. Cross-correlation analysis is performed on the environmental temperature and humidity real-time data and the vibration real-time data, and the correlation coefficients under different time delays are calculated to evaluate the relationship between temperature and humidity changes and vibration intensity, obtain the environment-vibration interaction characteristics, and construct a comprehensive set of environmental and vibration parameter features.

[0094] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0095] Normalize the AMC concentration change feature set to obtain a normalized AMC feature vector;

[0096] Perform feature standardization on the environment and vibration parameter feature sets to obtain a standardized environment vibration feature vector;

[0097] Construct an LSTM network as the time series branch, which includes an input layer, two LSTM layers and a fully connected layer. Use the tanh activation function to obtain the AMC time series feature extractor.

[0098] A CNN network is constructed as the spatial feature branch, which includes three 2D convolutional layers and one global average pooling layer. The ReLU activation function is used to obtain the environmental vibration spatial feature extractor.

[0099] The normalized AMC feature vector is input into the AMC time series feature extractor, and a 128-dimensional AMC time series feature vector is obtained through forward propagation;

[0100] The standardized environmental vibration feature vector is input into the environmental vibration spatial feature extractor, and a 64-dimensional environmental vibration spatial feature vector is obtained through forward propagation;

[0101] The AMC time series feature vector and the environmental vibration spatial feature vector are concatenated to obtain a 192-dimensional multimodal feature vector.

[0102] A three-layer fully connected neural network is constructed as the fusion module, with 128 and 64 hidden layer units respectively. The ReLU activation function is used to input the multimodal feature vector to obtain a 64-dimensional fusion feature representation.

[0103] Design a dual output layer structure, where the anomaly detection output layer uses the Sigmoid activation function, and the AMC concentration prediction output layer uses the linear activation function. The fused feature representation is input to obtain the anomaly probability and predicted AMC concentration value.

[0104] The weighted summation method is used to combine the abnormal probability and the predicted AMC concentration value to obtain the target transportation status evaluation index.

[0105] Specifically, the AMC concentration change feature set is feature normalized, and the eigenvalues ​​are scaled to a uniform range, usually [0, 1] or [-1, 1], to obtain a normalized AMC feature vector. The environmental and vibration parameter feature sets are feature normalized. The eigenvalues ​​are converted to a standard normal distribution with a mean of 0 and a variance of 1 to obtain a standardized environmental vibration feature vector, ensuring that all features have the same statistical properties. Construct an LSTM network as a time series branch. The network contains an input layer, two LSTM layers, and a fully connected layer. LSTM (Long Short-Term Memory Network) is a neural network structure that can capture long-term dependencies in time series data. Each LSTM unit contains an input gate, a forget gate, and an output gate. Through these gating mechanisms, information on time steps is effectively retained and updated. Assume that the input sequence is , the state update formula of the LSTM unit is:

[0106] ;

[0107] ;

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] in, , and are the activation values ​​of the forget gate, input gate, and output gate, respectively. is a candidate cell state, is the updated cell state, is the hidden state. Through forward propagation, a 128-dimensional AMC time series feature vector is obtained. At the same time, a CNN network is constructed as a spatial feature branch. The network contains three two-dimensional convolutional layers and a global average pooling layer. The convolution layer extracts spatial features by performing convolution operations on the input feature map. Assuming that the input feature map is X, the convolution kernel is K, and the output feature map of the convolution operation is Y, the convolution operation can be expressed as:

[0113] ;

[0114] Through the superposition of multiple convolutional layers, more abstract features are extracted layer by layer. The global average pooling layer performs a global average pooling operation on the feature map to obtain a global feature representation. Through forward propagation, a 64-dimensional environmental vibration spatial feature vector is obtained. The normalized AMC feature vector is input into the AMC time series feature extractor. Through the forward propagation of the LSTM layer, the input sequence is processed layer by layer to extract the time series features, and a 128-dimensional AMC time series feature vector is obtained. Similarly, the standardized environmental vibration feature vector is input into the environmental vibration spatial feature extractor. Through the forward propagation of the CNN layer, the spatial features are extracted layer by layer to obtain a 64-dimensional environmental vibration spatial feature vector. The AMC time series feature vector and the environmental vibration spatial feature vector are feature concatenated, and the two feature vectors are connected to form a 192-dimensional multimodal feature vector. A three-layer fully connected neural network is constructed as a fusion module. The number of hidden layer units of the network is 128 and 64 respectively, and the ReLU activation function is used. ReLU (rectified linear unit) is a commonly used activation function that increases the network's expressive power by performing nonlinear transformation on the input. Assuming the input is Z and the output of the fully connected layer is Y, the ReLU activation function can be expressed as:

[0115] ;

[0116] Through forward propagation, the 192-dimensional multimodal feature vector is input into the fusion module, and the features are processed layer by layer to obtain a 64-dimensional fusion feature representation. In order to achieve a comprehensive evaluation of the transportation status, a double output layer structure is designed. Among them, the anomaly detection output layer uses the Sigmoid activation function, which can map the output value to the [0, 1] interval, indicating the probability of anomaly occurrence. Assuming the input is u, the output of the Sigmoid function is , it can be expressed as:

[0117] ;

[0118] The AMC concentration prediction output layer uses a linear activation function to directly predict the AMC concentration value. Through forward propagation, the 64-dimensional fusion feature representation is input into the dual output layer structure to obtain the abnormal probability and predicted AMC concentration value respectively. The weighted summation method is used to combine the abnormal probability and the predicted AMC concentration value to obtain the target transportation status evaluation index. Assuming that the abnormal probability is p and the predicted AMC concentration value is , then the target transportation status evaluation index It can be expressed as:

[0119] ;

[0120] in, and They are the weights of abnormal probability and predicted AMC concentration value. Through appropriate weight setting, the two can be effectively combined to obtain a comprehensive evaluation index.

[0121] In a specific embodiment, the execution step of inputting the normalized AMC feature vector into the AMC time series feature extractor, and the process of obtaining the 128-dimensional AMC time series feature vector by forward propagation may specifically include the following steps:

[0122] The normalized AMC feature vector is expanded in time steps to obtain an input sequence of T time steps, each of which contains N-dimensional features;

[0123] According to the input sequence of T time steps, the forget gate of the first LSTM layer of the AMC time series feature extractor is calculated to obtain the forget gate output;

[0124] According to the input sequence of T time steps and the output of the forget gate, the input gate of the first LSTM layer is calculated to obtain the input gate output;

[0125] According to the input sequence of T time steps, the output of the forget gate and the output of the input gate, the candidate cell states of the first LSTM layer are calculated to obtain the candidate cell states;

[0126] According to the forget gate output, the input gate output and the candidate cell state, the cell state of the first LSTM layer is updated to obtain the updated cell state;

[0127] According to the input sequence of T time steps and the updated cell state, the output gate of the first LSTM layer is calculated to obtain the output gate output;

[0128] According to the output of the output gate and the updated cell state, the hidden state of the first LSTM layer is updated to obtain the updated hidden state;

[0129] According to the 256-dimensional output of the first LSTM layer, the second LSTM layer is calculated to obtain the 192-dimensional output of the second LSTM layer;

[0130] According to the 192-dimensional output of the second LSTM layer, the fully connected layer is linearly transformed to obtain the intermediate features, and the activation function of the fully connected layer is calculated based on the intermediate features to obtain a 128-dimensional AMC time series feature vector.

[0131] Specifically, the normalized feature vector is reshaped into an input sequence of T time steps. Each time step contains N-dimensional features, thereby converting the original single-dimensional data into a time series data structure suitable for LSTM processing. Assuming that the original normalized AMC feature vector is X, after reshaping, a three-dimensional tensor is obtained , whose shape is (T, N), that is, each time step contains N features, and there are T time steps in total. According to the input sequence of T time steps, the forget gate of the first LSTM layer of the AMC time series feature extractor is calculated. The function of the forget gate is to determine how much information in the current cell state needs to be forgotten. The calculation formula of the forget gate is:

[0132] ;

[0133] in, is the output of the forget gate, is the Sigmoid activation function, is the weight matrix of the forget gate, is the hidden state at the previous time step, is the input of the current time step, is the bias of the forget gate. According to the input sequence of T time steps and the output of the forget gate, the input gate of the first LSTM layer is calculated. The function of the input gate is to determine how much information of the current input needs to be written into the cell state. The calculation formula of the input gate is:

[0134] ;

[0135] in, is the output of the input gate, is the weight matrix of the input gate, is the bias of the input gate. Based on the input sequence of T time steps, the output of the forget gate and the output of the input gate, the candidate cell state of the first LSTM layer is calculated. The candidate cell state represents the update suggestion for the cell state of the current input. Its calculation formula is:

[0136]

[0137] in, is the candidate cell state, tanh is the hyperbolic tangent activation function, is the weight matrix of candidate cell states, is the bias of the candidate cell state. According to the forget gate output, input gate output and candidate cell state, the cell state of the first LSTM layer is updated. The update formula is:

[0138] ;

[0139] in, is the updated cell state, is the cell state of the previous time step. Based on the input sequence of T time steps and the updated cell state, the output gate of the first LSTM layer is calculated. The function of the output gate is to determine how much information of the cell state needs to be output to the hidden state. The calculation formula of the output gate is:

[0140] ;

[0141] in, is the output of the output gate, is the weight matrix of the output gate, is the bias of the output gate. The hidden state of the first LSTM layer is updated according to the output of the output gate and the updated cell state. The update formula is:

[0142] ;

[0143] in, is the updated hidden state. The output of the 256-dimensional first LSTM layer is used as input to calculate the second LSTM layer. The structure and calculation process of the second LSTM layer are similar to those of the first layer, but the weights and bias parameters are different. After calculation, the 192-dimensional output of the second LSTM layer is obtained. The 192-dimensional output of the second LSTM layer is input to the fully connected layer. The function of the fully connected layer is to map the high-dimensional feature vector to a low-dimensional one to obtain a more compact feature representation. The linear transformation formula of the fully connected layer is:

[0144] ;

[0145] Among them, Z is the output of the fully connected layer, W is the weight matrix of the fully connected layer, h is the input feature vector, and b is the bias of the fully connected layer. The output of the fully connected layer is activated using the ReLU activation function. The formula of the ReLU activation function is:

[0146] ;

[0147] Through the nonlinear transformation of the activation function, a 128-dimensional AMC time series feature vector is obtained. These feature vectors can capture the time series characteristics of AMC concentration data.

[0148] In a specific embodiment, the process of executing the step of inputting the standardized environmental vibration feature vector into the environmental vibration spatial feature extractor and obtaining the 64-dimensional environmental vibration spatial feature vector by forward propagation may specifically include the following steps:

[0149] Reshape the standardized environmental vibration feature vector to obtain a four-dimensional tensor input;

[0150] According to the four-dimensional tensor input, a convolution operation is performed on the first convolution layer of the environmental vibration spatial feature extractor to obtain an output feature map of the first convolution layer;

[0151] According to the output feature map of the first convolutional layer, the first batch of normalized layers are processed to obtain the normalized first layer feature map;

[0152] According to the normalized first layer feature map, a pooling operation is performed on the first maximum pooling layer to obtain an output feature map of the first pooling layer;

[0153] According to the output feature map of the first pooling layer, a convolution operation is performed on the second convolution layer to obtain the output feature map of the second convolution layer;

[0154] According to the output feature map of the second convolutional layer, the second batch of normalization layers are processed to obtain the normalized second layer feature map;

[0155] According to the normalized second layer feature map, a pooling operation is performed on the second maximum pooling layer to obtain an output feature map of the second pooling layer;

[0156] According to the output feature map of the second pooling layer, a convolution operation is performed on the third convolution layer to obtain the output feature map of the third convolution layer;

[0157] According to the feature map output by the third convolutional layer, the global average pooling layer is operated to obtain a 128-dimensional global feature vector;

[0158] According to the 128-dimensional global feature vector, the fully connected layer is linearly transformed and activated to obtain a 64-dimensional environmental vibration spatial feature vector.

[0159] Specifically, the standardized environmental vibration feature vector is converted into a four-dimensional tensor input suitable for convolutional neural network processing. Assuming that the original standardized environmental vibration feature vector is X, its shape is (batch_size, features), the reshaped four-dimensional tensor The shape of can be (batch_size, height, width, channels), where height, width, and channels represent the height, width, and number of channels of the input feature map, respectively. The first convolutional layer of the environmental vibration spatial feature extractor is convolved according to the 4D tensor input. The convolutional layer extracts local spatial features by applying multiple filters to the input feature map. Assume that the input feature map is , the convolution kernel is K, and the output feature map of the convolution operation is Y, then the convolution operation can be expressed as:

[0160] ;

[0161] in, Represents the value of position (i, j) in the output feature map, M and N represent the height and width of the convolution kernel respectively. According to the output feature map of the first convolutional layer, the first batch of normalization layers are processed to obtain the normalized first layer feature map. Batch normalization improves the training stability and convergence speed of the model by standardizing the output of each small batch to have zero mean and unit variance. Assuming the input is Z, the normalized output is , then the batch normalization formula is:

[0162] ;

[0163] in, and are the mean and variance of the small batches, is a small constant, and is a trainable parameter. According to the normalized first layer feature map, the first maximum pooling layer is pooled to obtain the first pooling layer output feature map. Maximum pooling reduces the size of the feature map by taking the maximum value in the local area while retaining important spatial features. Assume that the input feature map is , the pooling window size is , the pooling operation can be expressed as:

[0164] ;

[0165] in, Represents the value of position (i, j) in the feature map after pooling. Based on the output feature map of the first pooling layer, the second convolutional layer is convolved to obtain the output feature map of the second convolutional layer. Similar to the first convolutional layer, the second convolutional layer further extracts higher-level spatial features and convolves the input feature map through multiple convolution kernels. Based on the output feature map of the second convolutional layer, the second batch normalization layer is processed to obtain the normalized second-layer feature map. Performing batch normalization again helps stabilize the training process of the network and ensure that the numerical range of the features is consistent. Based on the normalized second-layer feature map, the second maximum pooling layer is pooled to obtain the output feature map of the second pooling layer. Through further pooling operations, the size of the feature map is further reduced while retaining key information. Based on the output feature map of the second pooling layer, the third convolutional layer is convolved to obtain the output feature map of the third convolutional layer. The third convolutional layer extracts higher-level features of the input data through deeper convolution operations to form the final feature representation. According to the feature map output by the third convolutional layer, the global average pooling layer is operated to obtain a 128-dimensional global feature vector. Global average pooling obtains a global feature representation by averaging all values ​​of each feature map, thereby significantly reducing the number of parameters and reducing the risk of overfitting. According to the 128-dimensional global feature vector, the fully connected layer is linearly transformed and activated to obtain a 64-dimensional environmental vibration space feature vector. The fully connected layer maps the high-dimensional feature vector to a low-dimensional one through weighted summation and bias. The linear transformation formula is:

[0166] ;

[0167] Among them, h is the output of the fully connected layer, W is the weight matrix of the fully connected layer, z is the input feature vector, and b is the bias of the fully connected layer.

[0168] In a specific embodiment, the process of executing step 106 may specifically include the following steps:

[0169] The target transport state evaluation index is normalized to obtain a normalized evaluation index, and the warning threshold is dynamically adjusted according to the normalized evaluation index to obtain an adaptive warning threshold;

[0170] The normalized evaluation index and the adaptive warning threshold are calculated to obtain the degree of deviation, and the degree of deviation is fuzzified to obtain the fuzzy deviation index;

[0171] According to the fuzzification deviation index, the preset multi-level alarm rules are inferred to obtain the fuzzy output of the alarm level, and the fuzzy output is defuzzified to obtain the multi-level alarm result;

[0172] According to the multi-level alarm results, case matching is performed from the preset exception handling strategy library to obtain preliminary exception handling suggestions, and the preliminary exception handling suggestions are context-adapted to obtain exception handling solutions;

[0173] According to the time series of the target transport status evaluation indicators, the transport quality trend analysis is carried out to obtain the transport quality prediction results, and the multi-level alarm results, abnormal handling solutions and transport quality prediction results are integrated to obtain the transport quality evaluation report.

[0174] Specifically, the target transport status evaluation indicators are standardized to ensure that indicators of different dimensions can be compared and analyzed on the same scale to obtain normalized evaluation indicators. According to the normalized evaluation indicators, the warning threshold is dynamically adjusted to obtain an adaptive warning threshold. The dynamic adjustment process is based on the comparison of real-time monitoring data and historical data to adjust the warning threshold to reflect the actual situation of the current transport status. Assume that the initial warning threshold is , the mean and standard deviation of the normalized evaluation index are and , then the adaptive warning threshold T can be expressed as:

[0175] ;

[0176] Among them, K is an adjustment coefficient. By adjusting this coefficient, the sensitivity of the warning threshold can be flexibly adjusted. The difference between the normalized evaluation index and the adaptive warning threshold is calculated to obtain the degree of deviation. The degree of deviation indicates the degree of deviation between the current evaluation index and the warning threshold, which is used to judge the abnormal situation of the current state. Assume that the normalized evaluation index is , the adaptive warning threshold is T, then the degree of deviation It can be expressed as:

[0177] ;

[0178] The calculation of the degree of deviation can quantify the abnormality of the current state and provide a basis for subsequent fuzzy processing. The degree of deviation is fuzzified to obtain a fuzzy deviation index. Fuzzy processing is to convert specific values ​​into fuzzy sets to handle uncertainty and fuzzy data. Assume that the degree of deviation is , the fuzzy deviation index after fuzzy processing is , can be transformed into Mapped to fuzzy sets. Commonly used membership functions include triangular membership functions and trapezoidal membership functions. According to the fuzzy deviation index, the preset multi-level alarm rules are inferred to obtain the fuzzy output of the alarm level. The multi-level alarm rules set different alarm levels according to different fuzzy deviation indicators to effectively distinguish different abnormal situations in the transportation process. Assume that the fuzzy deviation index is The preset multi-level alarm rules can be inferred through the fuzzy inference system to obtain the fuzzy output of the alarm level. . Defuzzify the fuzzy output to obtain a specific multi-level alarm result. Defuzzification is to convert the fuzzy set into a specific value, making the alarm level more intuitive and easy to understand. Common defuzzification methods include the centroid method and the maximum membership method. Assume that the fuzzy output is , the defuzzified multi-level alarm result is L. According to the multi-level alarm results, case matching is performed from the preset exception handling strategy library to obtain preliminary exception handling suggestions. The exception handling strategy library stores response plans for different alarm levels. Through case matching, the preliminary exception handling suggestions corresponding to the current alarm level are quickly selected. After obtaining the preliminary exception handling suggestions, they are contextually adapted, and the preliminary suggestions are adjusted and optimized in combination with the current transportation environment and specific circumstances to obtain the final exception handling plan. According to the time series of the target transportation status evaluation indicators, the transportation quality trend is analyzed to predict the changing trend of future transportation quality. Through time series analysis methods, potential problems and risks in the transportation process are identified, and preventive measures are taken in advance. Common time series analysis methods include ARIMA model and moving average model. Assume that the time series is , the ARIMA model can be expressed as:

[0179] ;

[0180] in, and are the parameters of the AR and MA parts respectively, is a white noise error term. The multi-level alarm results, abnormal handling solutions and transportation quality prediction results are integrated to generate a transportation quality assessment report.

[0181] The above describes the remote transport monitoring method of the chemical filter in the embodiment of the present invention. The following describes the remote transport monitoring device of the chemical filter in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a remote transport monitoring device for a chemical filter includes:

[0182] Installation module 201, used for sealing and packaging the chemical filter and installing the AMC monitoring module to obtain a target monitoring box;

[0183] Initialization module 202, used to initialize and collect benchmark data of the AMC concentration in the target monitoring box, obtain AMC concentration benchmark data, and set the warning threshold of the target monitoring box;

[0184] The acquisition module 203 is used to collect and grade data of the target monitoring box in real time, and obtain the real-time data of AMC concentration and the real-time data of environment and vibration;

[0185] The analysis module 204 is used to perform concentration change trend analysis on the AMC concentration real-time data and the AMC concentration baseline data to obtain an AMC concentration change feature set, and to perform feature extraction on the environment and vibration real-time data to obtain an environment and vibration parameter feature set;

[0186] Evaluation module 205, used for inputting the AMC concentration change feature set and the environment and vibration parameter feature set into a preset dual-branch time series and anomaly detection fusion model to perform transportation state evaluation and obtain a target transportation state evaluation index;

[0187] The generation module 206 is used to compare the target transportation status evaluation index with the warning threshold, trigger a multi-level alarm and generate an abnormality handling plan and a transportation quality evaluation report.

[0188] Through the cooperation of the above components, by integrating AMC sensors, temperature and humidity sensors and acceleration sensors, multi-dimensional monitoring of the chemical filter during transportation is realized, and the filter status and environmental changes are fully grasped. Adaptive sampling and filtering technology is used to dynamically adjust the sampling frequency according to the status evaluation results, and the original data is checked and processed in real time to ensure the accuracy and reliability of the monitoring data. The dual-branch fusion model is constructed by combining LSTM and CNN networks, making full use of the temporal characteristics of AMC concentration and the spatial characteristics of environmental vibration to improve the recognition accuracy of abnormal states. Through multi-level alarm mechanism and fuzzy reasoning, different levels of alarms are intelligently triggered according to the degree of deviation, and targeted abnormal handling solutions are generated to effectively reduce transportation risks. A hierarchical transmission strategy is adopted to perform adaptive transmission according to data importance and network conditions, reasonably allocate bandwidth resources, and improve data transmission efficiency. Through remote data transmission and cloud analysis, real-time monitoring and remote management of the transportation status of chemical filters are realized, improving supervision efficiency. By generating a transportation quality assessment report, including trend analysis and prediction results.

[0189] above Figure 2 The remote transport monitoring device for the chemical filter in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The remote transport monitoring device for the chemical filter in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0190] Figure 3 1 is a schematic diagram of the structure of a remote transport monitoring device for a chemical filter provided by an embodiment of the present invention. The remote transport monitoring device 300 for a chemical filter may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 may be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the remote transport monitoring device 300 for the chemical filter. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the remote transport monitoring device 300 for the chemical filter to implement the steps of the remote transport monitoring method for the chemical filter described above.

[0191] The remote transport monitoring device 300 for chemical filters may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the remote transport monitoring device for the chemical filter shown does not constitute a limitation on the remote transport monitoring device for the chemical filter provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0192] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the remote transport monitoring method for the chemical filter.

[0193] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0194] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0195] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote transport monitoring method for a chemical filter, characterized in that: The method comprises: The chemical filter is sealed and packaged and the AMC monitoring module is installed to obtain the target monitoring box; Initializing and collecting baseline data of the AMC concentration in the target monitoring box to obtain AMC concentration baseline data, and setting a warning threshold value of the target monitoring box; Performing real-time data collection and hierarchical transmission on the target monitoring box to obtain real-time data of AMC concentration as well as real-time data of environment and vibration; Performing concentration change trend analysis on the AMC concentration real-time data and the AMC concentration baseline data to obtain an AMC concentration change feature set, and performing feature extraction on the environment and vibration real-time data to obtain an environment and vibration parameter feature set; The AMC concentration change feature set and the environment and vibration parameter feature set are input into a preset dual-branch time series and anomaly detection fusion model to evaluate the transport state and obtain the target transport state evaluation index; specifically including: normalizing the AMC concentration change feature set to obtain a normalized AMC feature vector; normalizing the environment and vibration parameter feature set to obtain a standardized environment vibration feature vector; constructing an LSTM network as a time series branch, including an input layer, two LSTM layers and a fully connected layer, using a tanh activation function to obtain an AMC time series feature extractor; constructing a CNN network as a spatial feature branch, including three two-dimensional convolutional layers and a global average pooling layer, using a ReLU activation function to obtain an environment vibration spatial feature extractor; inputting the normalized AMC feature vector into the AMC time series feature extractor, and obtaining it through forward propagation. 128-dimensional AMC time series feature vector; input the standardized environmental vibration feature vector into the environmental vibration spatial feature extractor, and obtain a 64-dimensional environmental vibration spatial feature vector through forward propagation; perform feature splicing on the AMC time series feature vector and the environmental vibration spatial feature vector to obtain a 192-dimensional multimodal feature vector; construct a three-layer fully connected neural network as a fusion module, with the number of hidden layer units being 128 and 64 respectively, and use the ReLU activation function to input the multimodal feature vector to obtain a 64-dimensional fusion feature representation; design a dual output layer structure, in which the anomaly detection output layer uses the Sigmoid activation function, and the AMC concentration prediction output layer uses the linear activation function, and the fusion feature representation is input to obtain the anomaly probability and the predicted AMC concentration value; adopt the weighted summation method to combine the anomaly probability and the predicted AMC concentration value to obtain the target transportation state evaluation index; The target transport status assessment index and the warning threshold are compared to trigger a multi-level alarm and generate an abnormality handling plan and a transport quality assessment report.

2. The remote transport monitoring method of a chemical filter according to claim 1, characterized in that: The chemical filter is sealed and packaged and the AMC monitoring module is installed to obtain a target monitoring box, including: The chemical filter is sealed with an antistatic plastic bag and sealed with a hot melt technology to obtain a sealed chemical filter, and a multi-layer protection design is performed on the sealed chemical filter, the outer layer is treated with IP67 protection grade materials, and an anti-vibration pad is set on the inner layer to obtain a multi-layer protection chemical filter; According to the size of the multi-layer protective chemical filter, a monitoring box is customized to obtain a dedicated monitoring box, and the AMC monitoring module is installed on the dedicated monitoring box, and the AMC sensor, temperature and humidity sensor and acceleration sensor are integrated into the inner side of the monitoring box to obtain a sensor integrated monitoring box; Installing data acquisition and transmission modules on the sensor integrated monitoring box, integrating large-capacity storage and high-speed wireless communication modules, to obtain a data processing monitoring box; According to the AMC concentration monitoring requirements, the AMC sensor is calibrated and the monitoring accuracy is set to ±0.1ppm to obtain a high-precision AMC monitoring module; The sampling frequency of the high-precision AMC monitoring module is set to once every 5 minutes under normal conditions and once every 1 minute under abnormal conditions, thereby obtaining an adaptive sampling AMC monitoring module; According to the characteristics of the transportation environment, the temperature and humidity sensor and the acceleration sensor are configured with parameters, and the sampling range and frequency of temperature, humidity and vibration intensity are set to obtain an environmental parameter monitoring module; Designing a data fusion algorithm for the adaptive sampling AMC monitoring module and the environmental parameter monitoring module, and using time synchronization and multi-sensor data fusion technology to obtain a comprehensive data monitoring system; The multi-layer protective chemical filter is placed in the integrated data monitoring system, and the integrated data monitoring system is subjected to sealing and functional tests to obtain a target monitoring box.

3. The remote transport monitoring method of a chemical filter according to claim 2, characterized in that: The initialization and baseline data collection of the AMC concentration in the target monitoring box to obtain the AMC concentration baseline data and the setting of the warning threshold of the target monitoring box include: Preheat the target monitoring box and run the AMC monitoring module for 24 hours to obtain a stable AMC monitoring system; The AMC monitoring system in a stable state is sampled at high frequency, and AMC concentration data is collected once a minute for 24 hours to obtain the original AMC concentration data set; Performing outlier detection on the original AMC concentration data set to obtain a cleaned AMC concentration data set, and performing time series decomposition on the cleaned AMC concentration data set to obtain AMC concentration time series features; According to the AMC concentration time series characteristics, exponentially weighted moving average processing is performed on the cleaned AMC concentration data set to obtain AMC concentration benchmark data, and statistical analysis is performed on the AMC concentration benchmark data to obtain AMC concentration distribution characteristics; According to the AMC concentration distribution characteristics, the warning threshold is dynamically set to obtain an initial threshold, and a sensitivity analysis is performed on the initial threshold to simulate the false alarm rate and missed alarm rate under different thresholds to obtain a threshold-performance curve; According to the threshold-performance curve, the initial threshold is optimized and adjusted, and the balance point between the false alarm rate and the missed alarm rate is selected to obtain the warning threshold; The AMC concentration benchmark data and the warning threshold are stored in the local memory of the target monitoring box and the remote server to obtain an initialized AMC monitoring system.

4. The remote transport monitoring method of a chemical filter according to claim 3, characterized in that: The real-time data collection and hierarchical transmission of the target monitoring box to obtain the real-time data of AMC concentration and the real-time data of environment and vibration include: The AMC sensor, the temperature and humidity sensor and the acceleration sensor in the target monitoring box are synchronously triggered to obtain a multi-sensor synchronous sampling system; According to the preset sampling frequency of the normal state, the multi-sensor synchronous sampling system is sampled at intervals of 5 minutes to obtain the original multi-dimensional sensor data stream; Performing real-time data quality check on the original multidimensional sensor data stream to obtain a data quality indicator, and adaptively filtering the original multidimensional sensor data stream according to the data quality indicator to obtain filtered multidimensional sensor data; Performing feature extraction on the filtered multidimensional sensor data, calculating the mean, standard deviation and change rate of AMC concentration, the mean of temperature and humidity and the peak acceleration of vibration, and obtaining a multidimensional feature vector; According to the multi-dimensional feature vector, the current state is quickly evaluated to obtain a state evaluation result, and the state evaluation result is graded. If it is determined to be an abnormal state, the sampling frequency is increased to once per minute to obtain an adjusted sampling strategy; Reconfiguring the multi-sensor synchronous sampling system according to the adjusted sampling strategy to obtain updated sensor sampling parameters; The collected data is packaged in different levels, the normal state data is compressed and encoded, and the abnormal state data is kept with original accuracy to obtain a graded data packet; According to the network status and data priority, the hierarchical data packets are adaptively transmitted, and the data are transmitted to the remote server using a dynamically adjusted sending frequency to obtain real-time data on AMC concentration as well as real-time data on environment and vibration.

5. The remote transport monitoring method of a chemical filter according to claim 1, characterized in that: The AMC concentration real-time data and the AMC concentration baseline data are subjected to concentration change trend analysis to obtain an AMC concentration change feature set, and the environment and vibration real-time data are subjected to feature extraction to obtain an environment and vibration parameter feature set, including: Performing time series decomposition on the AMC concentration real-time data to obtain multi-scale components of AMC concentration, and performing linear regression analysis on trend components based on the multi-scale components of AMC concentration to calculate the slope and intercept to obtain an AMC concentration change rate index; Performing dynamic time warping distance calculation on the AMC concentration real-time data and the AMC concentration benchmark data to obtain an index of the degree of deviation of the AMC concentration from the benchmark; According to the AMC concentration real-time data, the statistical characteristics in the sliding window are calculated to obtain the AMC concentration statistical characteristic vector, and the AMC concentration real-time data is subjected to Fourier transformation to obtain the AMC concentration spectrum characteristics; Constructing an AMC concentration change feature set according to the AMC concentration change rate index, the AMC concentration deviation from the benchmark index, the AMC concentration statistical feature vector and the AMC concentration spectrum feature; Performing correlation analysis on the real-time environmental temperature and humidity data in the real-time environmental and vibration data, calculating the Pearson correlation coefficient and mutual information of temperature and humidity, obtaining temperature and humidity coupling characteristics, and calculating time domain characteristics and frequency domain characteristics according to the real-time vibration data in the real-time environmental and vibration data, obtaining vibration intensity and frequency characteristics; Performing cross-correlation analysis on the real-time data of environmental temperature and humidity and the real-time data of vibration, calculating correlation coefficients under different time delays, and obtaining environmental-vibration interaction characteristics; An environment and vibration parameter feature set is constructed according to the temperature and humidity coupling characteristics, the vibration intensity and frequency characteristics, and the environment-vibration interaction characteristics.

6. The remote transport monitoring method of a chemical filter according to claim 1, characterized in that: The normalized AMC feature vector is input into the AMC time series feature extractor, and a 128-dimensional AMC time series feature vector is obtained by forward propagation, including: Expanding the normalized AMC feature vector in time steps to obtain an input sequence of T time steps, each time step containing N-dimensional features; According to the input sequence of the T time steps, calculating the forget gate of the first LSTM layer of the AMC time series feature extractor to obtain a forget gate output; Calculating the input gate of the first LSTM layer according to the input sequence of the T time steps and the forget gate output to obtain the input gate output; Calculating the candidate cell states of the first LSTM layer according to the input sequence of the T time steps, the forget gate output, and the input gate output to obtain the candidate cell states; According to the forget gate output, the input gate output and the candidate cell state, the cell state of the first LSTM layer is updated to obtain an updated cell state; Calculating the output gate of the first LSTM layer according to the input sequence of the T time steps and the updated cell state to obtain the output gate output; According to the output of the output gate and the updated cell state, the hidden state of the first LSTM layer is updated to obtain an updated hidden state; According to the 256-dimensional output of the first LSTM layer, the second LSTM layer is calculated to obtain the 192-dimensional output of the second LSTM layer; According to the 192-dimensional output of the second LSTM layer, the fully connected layer is linearly transformed to obtain intermediate features, and according to the intermediate features, the activation function of the fully connected layer is calculated to obtain a 128-dimensional AMC time series feature vector.

7. The remote transport monitoring method of a chemical filter according to claim 1, characterized in that: The step of inputting the standardized environmental vibration feature vector into an environmental vibration spatial feature extractor and obtaining a 64-dimensional environmental vibration spatial feature vector through forward propagation includes: Reshaping the standardized environmental vibration feature vector to obtain a four-dimensional tensor input; According to the four-dimensional tensor input, a convolution operation is performed on the first convolution layer of the environmental vibration spatial feature extractor to obtain a first convolution layer output feature map; Processing the first batch of normalized layers according to the output feature map of the first convolutional layer to obtain a normalized first layer feature map; According to the normalized first layer feature map, a pooling operation is performed on the first maximum pooling layer to obtain a first pooling layer output feature map; According to the output feature map of the first pooling layer, a convolution operation is performed on the second convolution layer to obtain an output feature map of the second convolution layer; Processing the second batch of normalization layers according to the output feature map of the second convolutional layer to obtain a normalized second layer feature map; According to the normalized second layer feature map, a pooling operation is performed on the second maximum pooling layer to obtain a second pooling layer output feature map; According to the output feature map of the second pooling layer, a convolution operation is performed on the third convolution layer to obtain an output feature map of the third convolution layer; According to the output feature map of the third convolutional layer, the global average pooling layer is operated to obtain a 128-dimensional global feature vector; According to the 128-dimensional global feature vector, the fully connected layer is linearly transformed and activated by a function to obtain a 64-dimensional environmental vibration spatial feature vector.

8. The remote transport monitoring method of a chemical filter according to claim 3, characterized in that: The target transport status evaluation index is compared with the warning threshold, a multi-level alarm is triggered, and an abnormality handling plan and a transport quality evaluation report are generated, including: Normalizing the target transport state evaluation index to obtain a normalized evaluation index, and dynamically adjusting the warning threshold according to the normalized evaluation index to obtain an adaptive warning threshold; Performing difference calculation between the normalized evaluation index and the adaptive warning threshold to obtain a degree of deviation, and performing fuzzy processing on the degree of deviation to obtain a fuzzy deviation index; According to the fuzzy deviation index, the preset multi-level alarm rule is inferred to obtain the fuzzy output of the alarm level, and the fuzzy output is defuzzified to obtain the multi-level alarm result; According to the multi-level alarm results, case matching is performed from a preset exception handling strategy library to obtain preliminary exception handling suggestions, and the preliminary exception handling suggestions are context-adapted to obtain an exception handling solution; According to the time series of the target transport status evaluation index, a trend analysis is performed on the transport quality to obtain a transport quality prediction result, and the multi-level alarm result, the abnormality handling plan and the transport quality prediction result are integrated to obtain a transport quality evaluation report.

9. A remote transport monitoring device for a chemical filter, characterized in that: A method for remote transport monitoring of a chemical filter according to any one of claims 1 to 8, the device comprising: An installation module is used to seal and package the chemical filter and install the AMC monitoring module to obtain a target monitoring box; An initialization module, used to initialize and collect benchmark data of the AMC concentration in the target monitoring box, obtain AMC concentration benchmark data, and set the warning threshold of the target monitoring box; A collection module is used to collect and grade data of the target monitoring box in real time to obtain real-time data of AMC concentration and real-time data of environment and vibration; An analysis module, configured to perform concentration change trend analysis on the AMC concentration real-time data and the AMC concentration baseline data to obtain an AMC concentration change feature set, and to perform feature extraction on the environment and vibration real-time data to obtain an environment and vibration parameter feature set; An evaluation module, used for inputting the AMC concentration change feature set and the environment and vibration parameter feature set into a preset dual-branch time series and anomaly detection fusion model to perform transport state evaluation and obtain a target transport state evaluation index; The generation module is used to compare the target transportation status evaluation index with the warning threshold, trigger multi-level alarms and generate an abnormality handling plan and a transportation quality evaluation report.

Images