Method and system for evaluating the working state of a measurement object based on a pressure sensor

Abstract

The present invention discloses a method and system for evaluating the working state of a measurement object based on a pressure sensor, belonging to the technical field of state evaluation. It solves the problem that the existing methods evaluate the on-the-job state of employees to be audited based on a single or a few similarity indicators, which is difficult to meet the evaluation requirements based on pressure sensors. The method includes: preprocessing the time-series pressure electrical signal based on the characteristic frequency combined with the adaptive filtering algorithm to obtain a preprocessing set, and the state evaluation model identifies and analyzes the preprocessing set to output the working state evaluation result of the measurement object; based on whether the comprehensive evaluation value in the working state evaluation result exceeds the preset evaluation threshold; in the present invention, the working state evaluation result generated by the state evaluation model combined with the adaptive filtering algorithm has higher accuracy, and the state evaluation model can realize the up-order reconstruction function of the time-series pressure electrical signal, so as to comprehensively reflect the working state of the measurement object.

Description

Technical Field

[0001] The present invention belongs to the technical field of state evaluation, and particularly relates to a method and system for evaluating the working state of a measurement object based on a pressure sensor. Background Art

[0002] At present, in many office environments, it is necessary to know whether the staff are at their work posts, which can facilitate the transfer of calls or the handover of work. For the on-the-job status of office staff, the attendance is usually counted through an attendance card punching device, which can usually only count the working hours of going to work and getting off work, and lacks real-time statistics and positioning of the on-the-job status of personnel.

[0003] Chinese Patent CN116798087A discloses a method and system for detecting the on-the-job status of employees. The method includes: inputting an image of an employee to be audited into a target detection model to obtain first feature data of the image of the employee to be audited; calculating the similarity between the first feature data and the second feature data corresponding to each employee image pre-stored in a database based on a target similarity function to determine the employee identity information corresponding to the employee image with the highest similarity; judging the on-the-job status of the employee to be audited according to the employee identity information and the work station information of the employee to be audited. However, when the existing method uses the target similarity function to judge the on-the-job status of the employee to be audited, it is evaluated based on a single or a few similarity indexes, and it is difficult to meet the evaluation requirements based on a pressure sensor. To solve the above problems, we propose a method and system for evaluating the working state of a measurement object based on a pressure sensor. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for evaluating the working state of a measurement object based on a pressure sensor in view of the deficiencies of the prior art, and to solve the problem that when the existing method uses the target similarity function to judge the on-the-job status of an employee to be audited, it is evaluated based on a single or a few similarity indexes and it is difficult to meet the evaluation requirements based on a pressure sensor.

[0005] The present invention is implemented as follows. A method for evaluating the working state of a measurement object based on a pressure sensor, the method for evaluating the working state of a measurement object based on a pressure sensor includes:

[0006] Acquiring time-sequential pressure electrical signals based on a pressure sensor, preprocessing the time-sequential pressure electrical signals based on a characteristic frequency and an adaptive filtering algorithm to obtain a preprocessing set, and uploading the preprocessing set to a state database;

[0007] Pre-constructing a state evaluation model, extracting a modeling sample set from the state database, dividing the modeling sample set into a training set and a test set according to a ratio of 3:1, training the state evaluation model by using the training set and the test set, and outputting a converged state evaluation model;

[0008] Load the preprocessing set in real time. Taking the preprocessing set as the input, execute the state evaluation model. The state evaluation model identifies and analyzes the preprocessing set, and outputs the working state evaluation result of the measurement object;

[0009] Obtain the working state evaluation result. Based on whether the comprehensive evaluation value in the working state evaluation result exceeds the preset evaluation threshold, if it exceeds the preset evaluation threshold, trigger a working state warning instruction.

[0010] Preferably, the method for preprocessing the time-series pressure electrical signal based on the characteristic frequency combined with the adaptive filtering algorithm specifically includes:

[0011] Load the time-series pressure electrical signal. Based on the preset sampling period, divide the time-series pressure electrical signal into at least one set of characteristic frequency sets, and process the outliers and missing values of the characteristic frequency sets;

[0012] Load the characteristic frequency set. Based on the preset judgment period, identify the discrete frequency distribution, continuous frequency distribution, and no-pressure frequency distribution in the characteristic frequency set, and assign labels to the sampling periods corresponding to the discrete frequency distribution, continuous frequency distribution, and no-pressure frequency distribution. Among them, the discrete frequency distribution is marked as the working state, the continuous frequency distribution is marked as the to-be-identified state, and the no-pressure frequency distribution is marked as the state of leaving the work station;

[0013] Load the continuous frequency distribution. Based on the fast Fourier transform, obtain the spectrum sequence corresponding to the continuous frequency distribution, and calculate the amplitude energy and phase of the spectrum sequence;

[0014] Among them, the amplitude energy of the spectrum sequence is calculated by the following formula:

[0015] (1)

[0016] (2)

[0017] Among them, represents the amplitude energy of the spectrum sequence, is the phase of the spectrum sequence, represents the number of continuous signal points, represents the input representation of the continuous frequency distribution based on the fast Fourier transform, is the signal frequency;

[0018] Adopt a clustering algorithm to identify the central amplitude energy within the preset judgment period. Load the amplitude energy and phase of the spectrum sequence, and judge whether there is an approximate signal frequency within the preset judgment period;

[0019] If there is no approximate frequency, do not perform suppression filtering processing on the signal frequency within the preset judgment period;

[0020] If there is an approximate frequency, perform suppression processing on the approximate signal frequency based on a low-pass filter, obtain the amplitude energy corresponding to the approximate signal frequency in combination with the low-pass filter, and output the continuous frequency distribution after suppression filtering processing;

[0021] Load the continuous frequency distribution, discrete frequency distribution, pressure-free frequency distribution after suppression filtering processing and the corresponding marking status, and integrate them to obtain a preprocessing set.

[0022] Preferably, the method for performing suppression processing on the approximate signal frequency based on a low-pass filter specifically further includes:

[0023] Load the approximate signal frequency and the amplitude energy corresponding to the approximate signal frequency within a preset judgment period, and calculate the overlap similarity value between the approximate signal frequency and the central signal frequency corresponding to the central amplitude energy;

[0024] Perform harmonic suppression on the approximate signal frequency based on the low-pass filter in combination with the overlap similarity value, and output the approximate signal frequency after filtering;

[0025] Obtain the approximate signal frequency after filtering, and use the approximate signal frequency after filtering to perform an inverse fast Fourier transform to complete the suppression of the approximate signal frequency;

[0026] Among them, the overlap similarity value is calculated by the following formula:

[0027] (3)

[0028] Among them, represents the overlap similarity value, is the number of iterative filtering times of the low-pass filter, respectively represent the amplitude energy corresponding to the approximate signal frequency and the central amplitude energy, are the approximate signal frequency and the central signal frequency respectively;

[0029] When using the approximate signal frequency after filtering to perform an inverse fast Fourier transform, the inverse fast Fourier transform formula is:

[0030] (4)

[0031] Among them, represents the overlap similarity value, is the sampling angular difference of the approximate signal frequency, is the sampling frequency of the approximate signal.

[0032] Preferably, the method for training the state evaluation model using the training set and the test set specifically includes:

[0033] Load the pre-constructed state evaluation model, training set and test set, and set the hyperparameters, number of iterative rounds and training batches of the state evaluation model;

[0034] Obtain a training set, optimize the performance of the state evaluation model using the cross-entropy loss function, and adjust the connection weights between the encoder, decoder, and convolutional layer in the hidden layer in combination with the backpropagation algorithm;

[0035] Train the state classifier in the state evaluation model using the root mean square error loss function. For the embedded feature fusion result generated by the feature fusion layer, calculate the mean square error loss with the true classification result respectively. The root mean square error loss function is expressed as:

[0036] (5)

[0037] where respectively represent the elements in the th row and th column of the feature fusion matrix for the embedded feature fusion result and the true feature fusion result, and respectively represent the elements in the th row and th column of the feature fusion matrix for the embedded feature fusion label and the true feature fusion label;

[0038] Based on the state evaluation model whose training converges within the preset number of iterations, obtain a test set, test the state evaluation model based on the test set, and output the test result;

[0039] Load the test result, verify whether the state evaluation model meets the preset model accuracy based on the test result. If it meets the preset model accuracy, output the converged state evaluation model;

[0040] If it does not meet the preset model accuracy, use the Nadam optimizer to update the learning rate and hyperparameters of the state evaluation model, and continue to iteratively train the state evaluation model through the training set.

[0041]

[0041] Preferably, when using the Nadam optimizer to update the learning rate and hyperparameters of the state evaluation model, the first and second moments of the gradient tracked by the Nadam optimizer are expressed as:

[0042] (6)

[0043] (7)

[0044] (8)

[0045] where represents the hyperparameters of the state evaluation model, is the learning rate, is the loss function with respect to the hyperparameters of the state evaluation model the gradient at are the first - order moment and second - order moment estimates of the gradient respectively, and are the first and second decay rate parameters.

[0046] Preferably, the state evaluation model uses a multi - layer perceptron as the initial model. The initial model consists of an input layer, a hidden layer, and an output layer. A peak - shifting time function is introduced in the input layer to improve the hidden layer. An encoder and a decoder are introduced in the hidden layer, and three convolutional layers are set between the encoder and the decoder. The convolutional layer is a 3×3 convolution, and the activation function of the hidden layer is the ReLU activation function;

[0047] The encoder consists of two groups of max - pooling layers and a fully - connected layer. The fully - connected layer is connected to the max - pooling layer respectively. The decoder consists of two groups of average - pooling layers and a transposed - convolution layer. The size of the transposed - convolution kernel of the transposed - convolution layer is 3×3;

[0048] When pre - constructing the state evaluation model, freeze the output layer of the initial model, and use a feature fusion layer and a state classifier to replace the output layer of the initial model. The state classifier uses a three - layer multi - layer perceptron to implement the output of the state classification result. The state classifier is expressed as:

[0049] (9)

[0050] where, represents the embedded feature fusion result output by the feature fusion layer, represents the state classification result, is the feature of the output of the third - layer Neck network at node and is the number of nodes in the Neck network.

[0051] Preferably, the method for the state evaluation model to identify and analyze the pre - processing set specifically includes:

[0052] Load the pre - processing set. The input layer determines the to - be - identified state corresponding to the continuous frequency distribution in the pre - processing set based on the peak - shifting time function. The peak - shifting time function captures the characteristic frequency of the continuous frequency distribution and determines and marks the working state of the continuous frequency distribution;

[0053] Among them, the peak - shifting time function is expressed as:

[0054] (10)

[0055] where, represents the peak - shifting time function, respectively represent the approximate signal frequency points in the continuous frequency distribution the frequency timestamps between and represents the frequency preset delay, is the function attenuation coefficient;

[0056] Obtain the characteristic frequency of the continuous frequency distribution and determine the marker. The encoder uses the Gauss-Newton method to matrix-elevate the expression of the characteristic frequency of the continuous frequency distribution, perform a convolution operation on the matrix-elevated expression of the characteristic frequency, and output the result of the convolution operation;

[0057] Load the result of the convolution operation. The decoder splices the result of the convolution operation based on skip connections and average pooling, and outputs the frequency reconstruction result;

[0058] Obtain the frequency reconstruction result. The feature fusion layer performs horizontal splicing of dimensions and features on the frequency reconstruction result, and the state classifier outputs the state classification result based on the Neck network;

[0059] Evaluate the state classification result based on the cascading failure combined with the structural similarity algorithm, calculate the comprehensive evaluation value, and output the comprehensive evaluation value;

[0060] The comprehensive evaluation value is calculated by the following formula:

[0061] (11)

[0062] Where, represents the comprehensive evaluation value, is the weight of the cascading failure node of, represents the state classification result, is the structural similarity index, is the bias constant.

[0063] On the other hand, the present invention also provides a system for evaluating the working state of a measurement object based on a pressure sensor. The system for evaluating the working state of a measurement object based on a pressure sensor includes:

[0064] A preprocessing module, which obtains the time-series pressure electrical signal based on the pressure sensor, preprocesses the time-series pressure electrical signal based on the characteristic frequency combined with the adaptive filtering algorithm, obtains the preprocessing set, and uploads the preprocessing set to the state database;

[0065] A model construction module, which is used to pre-construct a state evaluation model, extract a modeling sample set from the state database, divide the modeling sample set into a training set and a test set according to a ratio of 3:1, and train the state evaluation model with the training set and the test set, and output the converged state evaluation model;

[0066] A state evaluation module, which is used to load the preprocessing set in real time, use the preprocessing set as the input, execute the state evaluation model, and the state evaluation model identifies and analyzes the preprocessing set, and outputs the working state evaluation result of the measurement object;

[0067] A status judgment module, configured to obtain the work status evaluation result, and based on whether the comprehensive evaluation value in the work status evaluation result exceeds a preset evaluation threshold, if it exceeds the preset evaluation threshold, trigger a work status warning instruction.

[0068] Preferably, the status judgment module includes:

[0069] A result recognition unit, configured to obtain the work status evaluation result;

[0070] A status judgment unit, based on whether the comprehensive evaluation value in the work status evaluation result exceeds a preset evaluation threshold;

[0071] A status warning unit, configured to trigger a work status warning instruction when the work status evaluation result exceeds a preset evaluation threshold.

[0072] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects:

[0073] In the embodiments of the present invention, based on the characteristic frequency combined with the adaptive filtering algorithm, the time-series pressure electrical signal is preprocessed, so as to effectively suppress the interference signal in the time-series pressure electrical signal interference, reduce the misjudgment caused by interference, and thus improve the accuracy of the status evaluation. The work status evaluation result generated by the status evaluation model combined with the adaptive filtering algorithm has higher accuracy. The status evaluation model can realize the up-order reconstruction function of the time-series pressure electrical signal, and then comprehensively reflect the work status of the measurement object. It overcomes the problem that the existing method uses the target similarity function to judge the on-duty status of the employee to be audited based on a single or a few similarity indicators for evaluation, which is difficult to meet the evaluation requirements based on the pressure sensor and cannot comprehensively reflect the work status of the measurement object.

[0074] In the embodiments of the present invention, when preprocessing the time-series pressure electrical signal based on the characteristic frequency combined with the adaptive filtering algorithm, tags are assigned to the sampling periods corresponding to the discrete frequency distribution, the continuous frequency distribution, and the no-pressure frequency distribution, so as to facilitate the status evaluation model to quickly identify the response preprocessing set. At the same time, the amplitude energy corresponding to the approximate signal frequency is obtained by combining with a low-pass filter, and the continuous frequency distribution after the suppression filtering process is output, so as to effectively eliminate the influence of the interference signal on the pressure electrical signal, and ensure the evaluation accuracy and efficiency of the status evaluation model.

[0075] In the embodiments of the present invention, by filtering out high-frequency noise, the low-pass filter helps to retain the low-frequency components in the signal, thereby improving the overall quality of the signal, further reducing noise interference, and the low-pass filter can effectively remove interference noise, making the signal clearer, facilitating subsequent analysis and processing by the state evaluation model. Moreover, when the low-pass filter suppresses the frequency harmonics of the approximate signal, an overlapping similarity value is introduced, which can ensure that the low-frequency characteristics in the signal are not lost, avoid losing important information due to over-filtering, and ensure the integrity of the pressure electrical signal.

[0076] In the embodiments of the present invention, a state evaluation model and a training method are provided. A peak-shifting time function is introduced into the state evaluation model, which facilitates the identification and marking of abnormal states. At the same time, an encoder and a decoder are introduced into the hidden layer to improve the feature extraction ability, and the decoder network hierarchy is designed through a skip connection mechanism to effectively capture and reconstruct the behavioral characteristics of the measurement object, thereby capturing more global feature information. At the same time, a feature fusion layer and a state classifier are used to replace the output layer of the initial model, further effectively neutralizing the impact of noise signals on state evaluation and proposing a feature fusion method to constrain the state classifier to ensure its applicability and accuracy in state evaluation. Brief Description of the Drawings

[0077] Figure 1 It is a schematic flowchart of the implementation of the method for evaluating the working state of a measurement object based on a pressure sensor provided by the present invention.

[0078] Figure 2 It shows a schematic flowchart of the implementation of the method for preprocessing a time-series pressure electrical signal based on a characteristic frequency combined with an adaptive filtering algorithm.

[0079] Figure 3 It shows a schematic flowchart of the implementation of the method for suppressing the frequency of an approximate signal based on a low-pass filter.

[0080] Figure 4 It shows a schematic flowchart of the implementation of the method for training a state evaluation model using a training set and a test set.

[0081] Figure 5 It shows a schematic flowchart of the implementation of the method for the state evaluation model to identify and analyze a preprocessing set.

[0082] Figure 6 It is a schematic structural diagram of the system for evaluating the working state of a measurement object based on a pressure sensor provided by the present invention. Detailed Description of the Embodiments

[0083] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0084] When the existing method uses the target similarity function to judge the on-the-job status of the employee to be audited, it is evaluated based on a single or a few similarity indicators, which is difficult to meet the evaluation requirements based on the pressure sensor. To address the above problems, we propose a method and system for evaluating the working status of a measurement object based on a pressure sensor. When the method for evaluating the working status of a measurement object based on a pressure sensor is implemented, first, a time-series pressure electrical signal is obtained based on the pressure sensor, and the time-series pressure electrical signal is preprocessed based on the characteristic frequency combined with the adaptive filtering algorithm. At the same time, a state evaluation model is pre-constructed, and then the state evaluation model is executed. The state evaluation model identifies and analyzes the preprocessing set and outputs the evaluation result of the working status of the measurement object. Finally, based on whether the comprehensive evaluation value in the evaluation result of the working status exceeds the preset evaluation threshold, if it exceeds the preset evaluation threshold, a working status warning instruction is triggered. In the embodiment of the present invention, the time-series pressure electrical signal is preprocessed based on the characteristic frequency combined with the adaptive filtering algorithm, so that the interference signal in the interference of the time-series pressure electrical signal can be effectively suppressed, and the misjudgment caused by the interference can be reduced, thereby improving the accuracy of the state evaluation. The evaluation result of the working status generated by the state evaluation model combined with the adaptive filtering algorithm has higher accuracy. The state evaluation model can realize the function of up-order reconstruction of the time-series pressure electrical signal, and then comprehensively reflect the working status of the measurement object. It overcomes the problems that the existing method uses the target similarity function to judge the on-the-job status of the employee to be audited, is evaluated based on a single or a few similarity indicators, is difficult to meet the evaluation requirements based on the pressure sensor, and cannot comprehensively reflect the working status of the measurement object.

[0085] The embodiment of the present invention provides a method for evaluating the working status of a measurement object based on a pressure sensor, Figure 1 which shows a schematic flowchart of the implementation of the method for evaluating the working status of a measurement object based on a pressure sensor. The method for evaluating the working status of a measurement object based on a pressure sensor specifically includes:

[0086] Step S10, obtaining a time-series pressure electrical signal based on the pressure sensor, preprocessing the time-series pressure electrical signal based on the characteristic frequency combined with the adaptive filtering algorithm to obtain a preprocessing set, and uploading the preprocessing set to the state database;

[0087] It should be noted that the measurement objects referred to in this embodiment can be workers, drivers, athletes, medical staff, office workers. For example, pressure sensors can be integrated into office chairs or keyboards to monitor the pressure distribution of office workers when using computers, helping them adjust their sitting postures and reduce physical discomfort caused by long-term sitting. At the same time, the measurement objects can also be mechanical equipment, medical equipment, electronic equipment, building structures, etc. In this embodiment, the pressure sensor can be a CS-iTWO-05 type pressure sensor, and the pressure sensor supports OTAA and ABP network access methods. The network access parameters can adopt the factory default parameters. Users can provide the network access parameters to the supplier to write the parameters when the product leaves the factory, or users can also choose to configure the network access parameters by themselves. The installation method of the pressure sensor can be flange, bolt, buckle or embedded installation.

[0088] In this embodiment, the status database can be a SQL Server 2019 database, which facilitates data exchange and service docking between various system platforms, ensuring the compatibility and consistency of the system.

[0089] Step S20: Pre-build a status evaluation model, extract a modeling sample set from the status database, divide the modeling sample set into a training set and a test set according to a ratio of 3:1, and use the training set and the test set to train the status evaluation model to output a converged status evaluation model;

[0090] Step S30: Real-time load the preprocessing set, use the preprocessing set as the input, execute the status evaluation model, and the status evaluation model identifies and analyzes the preprocessing set to output the working status evaluation result of the measurement object;

[0091] Step S40: Obtain the working status evaluation result, and determine whether the comprehensive evaluation value in the working status evaluation result exceeds the preset evaluation threshold; in this embodiment, the preset evaluation threshold can be 0.6 - 0.8.

[0092] Step S50: If it exceeds the preset evaluation threshold, trigger a working status warning instruction.

[0093] If it does not exceed the preset evaluation threshold, upload the working status evaluation result to the status database.

[0094] In the embodiments of the present invention, the time-series pressure electrical signal is preprocessed based on the characteristic frequency in combination with the adaptive filtering algorithm, so as to effectively suppress the interference signal in the time-series pressure electrical signal interference, reduce the misjudgment caused by the interference, and thus improve the accuracy of the state evaluation. The working state evaluation result generated by the state evaluation model in combination with the adaptive filtering algorithm has a higher accuracy. The state evaluation model can realize the up-order reconstruction function of the time-series pressure electrical signal, and further comprehensively reflect the working state of the measurement object. It overcomes the problem that the existing method uses the target similarity function to judge the on-the-job state of the employee to be reviewed based on a single or a few similarity indicators, which is difficult to meet the evaluation requirements based on the pressure sensor and cannot comprehensively reflect the working state of the measurement object.

[0095] The embodiments of the present invention provide a method for preprocessing the time-series pressure electrical signal based on the characteristic frequency in combination with the adaptive filtering algorithm. Figure 2 The schematic diagram of the implementation process of the method for preprocessing the time-series pressure electrical signal based on the characteristic frequency in combination with the adaptive filtering algorithm is shown. The method for preprocessing the time-series pressure electrical signal based on the characteristic frequency in combination with the adaptive filtering algorithm specifically includes:

[0096] Step S101, load the time-series pressure electrical signal, divide the time-series pressure electrical signal into at least one set of characteristic frequency sets based on the preset sampling period, and process the outliers and missing values of the characteristic frequency sets.

[0097] In this embodiment, the method for processing the outliers of the characteristic frequency set can be direct deletion (for the outliers that are significantly deviated from the normal range, these records can be directly deleted), average value correction (using the average value of the previous and subsequent observations to correct the outlier), and the method for processing the missing values of the characteristic frequency set can be interpolation method and replacement method.

[0098] Step S102, load the characteristic frequency set, identify the discrete frequency distribution, continuous frequency distribution, and no-pressure frequency distribution in the characteristic frequency set based on the preset judgment period, and assign labels to the sampling periods corresponding to the discrete frequency distribution, continuous frequency distribution, and no-pressure frequency distribution. Among them, the discrete frequency distribution is marked as the working state, the continuous frequency distribution is marked as the to-be-identified state, and the no-pressure frequency distribution is marked as the state of leaving the work station.

[0099] It should be noted that the sampling period can be 0.01 - 0.5 seconds. In this embodiment, the discrete frequency distribution in the characteristic frequency set is regarded as the continuous activity of the measurement object, and then the current sampling period is assigned as the working state. The continuous frequency distribution is regarded as the continuous activity / dormancy of the measurement object. Therefore, the continuous frequency distribution is marked as the to-be-identified state.

[0100] Step S103: Load the continuous frequency distribution, obtain the spectrum sequence corresponding to the continuous frequency distribution based on the fast Fourier transform, and calculate the amplitude energy and phase of the spectrum sequence.

[0101] Among them, the amplitude energy of the spectrum sequence is calculated by the following formula:

[0102] (1)

[0103] (2)

[0104] Among them, represents the amplitude energy of the spectrum sequence, is the phase of the spectrum sequence, represents the number of continuous signal points, represents the input representation of the continuous frequency distribution based on the fast Fourier transform, is the signal frequency;

[0105] Step S104: Use a clustering algorithm to identify the central amplitude energy within the preset judgment period, and load the amplitude energy and phase of the spectrum sequence.

[0106] In this embodiment, the hierarchical clustering algorithm is used to cluster the signals. After clustering, the characteristics of each cluster are analyzed, especially the characteristics of the center point or representative point of the cluster. These center points or representative points often represent the typical characteristics of the signals within the cluster, including the central amplitude energy. By comparing the central amplitude energies of different clusters, the signal clusters with specific central amplitude energies within the preset judgment period can be identified.

[0107] Step S105: Judge whether there is an approximate signal frequency within the preset judgment period.

[0108] It should be noted that when judging whether there is an approximate signal frequency within the preset judgment period, factors such as the periodicity, stability, frequency components of the signal, as well as sampling and aliasing need to be considered, and the LabVIEW software is used for analysis and verification.

[0109] Step S106: If there is no approximate frequency, no suppression filtering process is performed on the signal frequency within the preset judgment period.

[0110] Step S107: If there is an approximate frequency, perform suppression processing on the approximate signal frequency based on a low-pass filter, combine the low-pass filter to obtain the amplitude energy corresponding to the approximate signal frequency, and output the continuous frequency distribution after the suppression filtering process.

[0111] Step S108: Load the continuous frequency distribution, discrete frequency distribution, non-pressure frequency distribution after the suppression filtering process and their corresponding marked states, and integrate them to obtain the preprocessing set.

[0112] In the embodiment of the present invention, when preprocessing the time-series piezoresistive signal based on the characteristic frequency in combination with the adaptive filtering algorithm, tags are assigned to the sampling periods corresponding to the discrete frequency distribution, the continuous frequency distribution, and the no-pressure frequency distribution, which facilitates the state evaluation model to quickly identify the response preprocessing set. At the same time, the amplitude energy corresponding to the approximate signal frequency is obtained by combining with a low-pass filter, and the continuous frequency distribution after suppression filtering is output, so as to effectively eliminate the influence of the interference signal on the piezoresistive signal, and ensure the evaluation accuracy and efficiency of the state evaluation model.

[0113] The embodiment of the present invention provides a method for suppressing the approximate signal frequency based on a low-pass filter. Figure 3 The schematic implementation flow diagram of the method for suppressing the approximate signal frequency based on a low-pass filter is shown. The method for suppressing the approximate signal frequency based on a low-pass filter specifically includes:

[0114] Step S1071: Load the approximate signal frequency and the amplitude energy corresponding to the approximate signal frequency within the preset judgment period, and calculate the overlapping similarity value between the approximate signal frequency and the central signal frequency corresponding to the central amplitude energy;

[0115] Step S1072: Based on the low-pass filter and the overlapping similarity value, suppress the harmonics of the approximate signal frequency, and output the filtered approximate signal frequency;

[0116] Step S1073: Obtain the filtered approximate signal frequency, and use the filtered approximate signal frequency to perform the inverse fast Fourier transform to complete the suppression of the approximate signal frequency;

[0117] Among them, the overlapping similarity value is calculated by the following formula:

[0118] (3)

[0119] Among them, represents the overlapping similarity value, is the iterative filtering times of the low-pass filter, which can be 5 - 15, respectively represent the amplitude energy corresponding to the approximate signal frequency and the central amplitude energy, are the approximate signal frequency and the central signal frequency respectively;

[0120] When performing the inverse fast Fourier transform using the filtered approximate signal frequency, the inverse fast Fourier transform formula is:

[0121] (4)

[0122] Among them, represents the overlapping similarity value, is the sampling angular difference of the approximate signal frequency, is the sampling frequency of the approximate signal.

[0123] In the embodiments of the present invention, by filtering out high-frequency noise, the low-pass filter helps to retain the low-frequency components in the signal, thereby improving the overall quality of the signal, further reducing noise interference, and the low-pass filter can effectively remove interference noise, making the signal clearer, facilitating subsequent analysis and processing by the state evaluation model. Moreover, when the low-pass filter suppresses the frequency harmonics of the approximate signal, an overlapping similarity value is introduced, which can ensure that the low-frequency characteristics in the signal are not lost, avoid losing important information due to over-filtering, and ensure the integrity of the pressure electrical signal.

[0124] The embodiments of the present invention provide a method for training a state evaluation model using a training set and a test set. Figure 4 FIG. shows a schematic implementation flow diagram of a method for training a state evaluation model using a training set and a test set. The method for training a state evaluation model using a training set and a test set specifically includes:

[0125] Step S201, load the pre-constructed state evaluation model, training set, and test set, and set the hyperparameters, number of iterations, and training batches of the state evaluation model;

[0126] In this embodiment, a total of 800 pieces of data are collected for the training set and the test set. The number of iterations is set to 120 - 150 times, and the training batch is 8 - 10.

[0127] Step S202, obtain the training set, optimize the performance of the state evaluation model using the cross-entropy loss function, and adjust the connection weights between the encoder, decoder, and convolutional layer in the hidden layer in combination with the backpropagation algorithm;

[0128] Step S203, train the state classifier in the state evaluation model using the root mean square error loss function. For the embedded feature fusion result generated by the feature fusion layer, calculate the mean square error loss with the true classification result respectively. The root mean square error loss function is expressed as:

[0129] (5)

[0130] Where, respectively represent the elements in the row and the column of the feature fusion matrix of the embedded feature fusion result and the true feature fusion result, respectively represent the elements in the row and the column of the feature fusion matrix of the embedded feature fusion label and the true feature fusion label, row and the column;

[0131] Step S204: Based on the state evaluation model whose training converges within the preset number of iteration rounds, obtain a test set, and test the state evaluation model based on the test set to output a test result;

[0132] Step S205: Load the test result, and verify whether the state evaluation model meets the preset model accuracy based on the test result;

[0133] Step S206: If it meets the preset model accuracy, output the converged state evaluation model;

[0134] Step S207: If it does not meet the preset model accuracy, use the Nadam optimizer to update the learning rate and hyperparameters of the state evaluation model, return to Step S202, and continue to iteratively train the state evaluation model through the training set.

[0135] It should be noted that the model accuracy is set to 0.92 - 0.98, and the model accuracy evaluation metrics can be accuracy, precision, and recall. In the embodiments of the present invention, the accuracy, precision, and recall of the state evaluation model are 0.9315, 0.9522, and 0.9345 respectively, indicating that the accuracy and robustness of the state evaluation model trained in this embodiment are relatively good, and it verifies that the state evaluation model has a good recognition and analysis effect on the preprocessing set.

[0136] In this embodiment, when using the Nadam optimizer to update the learning rate and hyperparameters of the state evaluation model, the first and second moments of the gradient tracked by the Nadam optimizer are expressed as:

[0137] (6)

[0138] (7)

[0139] (8)

[0140] Wherein, represents the hyperparameter of the state evaluation model, is the learning rate. In this embodiment, the learning rate is set to 0.02, is the gradient of the loss function at the hyperparameter of the state evaluation model, are the first and second moment estimates of the gradient respectively, are the first and second decay rate parameters. In this embodiment, the first and second decay rate parameters can be 0.2 - 0.5.

[0141] It should be noted that the Nadam optimizer is an optimizer that combines the advantages of the Adam algorithm and the Nesterov Accelerated Gradient (NAG) algorithm. It performs excellently in the training of deep learning models. By introducing the concept of Nesterov momentum, Nadam takes into account the velocity of the previous step, thus "seeing" the direction of the next step in advance when calculating the update. This acceleration mechanism helps the model converge to the optimal solution faster, especially when dealing with high curvature, small but consistent gradients, or noisy gradients.

[0142] In this embodiment, the state evaluation model takes a multi-layer perceptron as the initial model. The initial model consists of an input layer, a hidden layer, and an output layer. A peak-shifting time function is introduced in the input layer to improve the hidden layer. An encoder and a decoder are introduced in the hidden layer, and three convolutional layers are set between the encoder and the decoder. The convolutional layer is a 3×3 convolution, and the activation function of the hidden layer is the ReLU activation function.

[0143] The encoder consists of two groups of max pooling layers and a fully connected layer. The fully connected layer is connected to the max pooling layers respectively. The decoder consists of two groups of average pooling layers and a deconvolution layer. The deconvolution kernel size of the deconvolution layer is 3×3.

[0144] When pre-building the state evaluation model, freeze the output layer of the initial model, and replace the output layer of the initial model with a feature fusion layer and a state classifier. The state classifier uses a three-layer multi-layer perceptron to output the state classification result, and the state classifier is expressed as:

[0145] (9)

[0146] Among them, represents the embedded feature fusion result output by the feature fusion layer, represents the state classification result, is the feature of the output of the third-layer Neck network node, is the number of nodes in the Neck network.

[0147] In the embodiment of the present invention, a state evaluation model and a training method are provided. A peak-shifting time function is introduced in the state evaluation model, which facilitates the identification and marking of abnormal states. At the same time, by introducing an encoder and a decoder in the hidden layer, the feature extraction ability is improved, and the decoder network hierarchy is designed through a skip connection mechanism to effectively capture and reconstruct the behavioral characteristics of the measurement object, thereby capturing more global feature information. At the same time, the output layer of the initial model is replaced with a feature fusion layer and a state classifier, which further effectively neutralizes the influence of noise signals on state evaluation and proposes a feature fusion method to constrain the state classifier to ensure its applicability and accuracy in state evaluation.

[0148] An embodiment of the present invention provides a method for a state evaluation model to identify and analyze a preprocessing set. Figure 5 The schematic diagram of the implementation process of the method for the state evaluation model to identify and analyze the preprocessing set is shown. The method for the state evaluation model to identify and analyze the preprocessing set specifically includes:

[0149] Step S301, load the preprocessing set. The input layer determines the to-be-identified state corresponding to the continuous frequency distribution in the preprocessing set based on the off-peak time function. The off-peak time function captures the characteristic frequency of the continuous frequency distribution and determines and marks the working state of the continuous frequency distribution.

[0150] Among them, the off-peak time function is expressed as:

[0151] (10)

[0152] Among them, represents the off-peak time function, respectively represent the approximate signal frequency points in the continuous frequency distribution the frequency timestamps between, represents the frequency preset delay, is the function attenuation coefficient. In this embodiment, the function attenuation coefficient can be 0.02 - 0.08;

[0153] Step S302, obtain the characteristic frequency of the continuous frequency distribution and the determined mark. The encoder uses the Gauss-Newton method to matrix and up-grade the expression of the characteristic frequency of the continuous frequency distribution, and performs a convolution operation on the matrixed and up-graded characteristic frequency, and outputs the result of the convolution operation.

[0154] Step S303, load the result of the convolution operation. The decoder splices the result of the convolution operation based on skip connection and average pooling, and outputs the frequency reconstruction result.

[0155] Step S304, obtain the frequency reconstruction result. The feature fusion layer performs horizontal splicing of dimensions and features on the frequency reconstruction result, and the state classifier outputs the state classification result based on the Neck network.

[0156] Step S305, evaluate the state classification result based on the cascading failure combined with the structural similarity algorithm, calculate the comprehensive evaluation value, and output the comprehensive evaluation value.

[0157] In this embodiment, the comprehensive evaluation value is calculated by the following formula:

[0158] (11)

[0159] Among them, represents the comprehensive evaluation value, is the cascading failure node The weight represents the state classification result. is the structural similarity index. In this embodiment, the structural similarity index is 0.75 - 0.9. is the bias constant, which can be 0.02 - 0.05.

[0160] On the other hand, the embodiment of the present invention also provides a system for evaluating the working state of a measurement object based on a pressure sensor. Figure 6 FIG. shows a schematic structural diagram of a system for evaluating the working state of a measurement object based on a pressure sensor. The system for evaluating the working state of a measurement object based on a pressure sensor specifically includes:

[0161] A preprocessing module 100, which acquires a time-series pressure electrical signal based on a pressure sensor, preprocesses the time-series pressure electrical signal based on a characteristic frequency in combination with an adaptive filtering algorithm to obtain a preprocessing set, and uploads the preprocessing set to a state database.

[0162] A model construction module 200, which is used to pre-construct a state evaluation model, extracts a modeling sample set from the state database, divides the modeling sample set into a training set and a test set according to a ratio of 3:1, and trains the state evaluation model using the training set and the test set to output a converged state evaluation model.

[0163] A state evaluation module 300, which is used to load the preprocessing set in real time, takes the preprocessing set as an input, executes the state evaluation model, and the state evaluation model identifies and analyzes the preprocessing set to output a working state evaluation result of the measurement object.

[0164] A state judgment module 400, which is used to obtain the working state evaluation result, and based on whether the comprehensive evaluation value in the working state evaluation result exceeds a preset evaluation threshold. If it exceeds the preset evaluation threshold, a working state warning instruction is triggered.

[0165] In this embodiment, the state judgment module 400 includes:

[0166] A result identification unit 410, which is used to obtain the working state evaluation result.

[0167] A state judgment unit 420, which is based on judging whether the comprehensive evaluation value in the working state evaluation result exceeds a preset evaluation threshold.

[0168] A state warning unit 430, which is used to trigger a working state warning instruction when the working state evaluation result exceeds a preset evaluation threshold.

[0169] It should be noted that the system for evaluating the working state of a measurement object based on a pressure sensor provided in the embodiments of the present invention corresponds to the method for evaluating the working state of a measurement object based on a pressure sensor described above. For the explanations, examples, beneficial effects, and other parts of the relevant content, reference can be made to the corresponding content in the method for evaluating the working state of a measurement object based on a pressure sensor, and details will not be repeated here.

[0170] On the other hand, the embodiments of the present invention also provide a computer-readable storage medium storing computer program instructions executable by a processor. When the computer program instructions are executed, the methods of any of the above embodiments are implemented.

[0171] In yet another aspect of the embodiments of the present invention, a computer device is further provided. The computer device includes a memory and a processor. A computer program is stored in the memory, and when the computer program is executed by the processor, the methods of any of the above embodiments are implemented.

[0172] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as program instructions / modules corresponding to the method for evaluating the working state of a measurement object based on a pressure sensor in the embodiments of the present application. The memory may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created during the use of the method for evaluating the working state of a measurement object based on a pressure sensor, etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely provided with respect to the processor, and these remote memories can be connected to the local module through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0173] In summary, the present invention provides a method and system for evaluating the working state of a measurement object based on a pressure sensor. In the embodiments of the present invention, the time-series pressure electrical signal is preprocessed based on the characteristic frequency and the adaptive filtering algorithm, so as to effectively suppress the interference signal in the time-series pressure electrical signal interference, reduce the misjudgment caused by interference, and thus improve the accuracy of state evaluation. The working state evaluation result generated by combining the state evaluation model with the adaptive filtering algorithm has higher accuracy. The state evaluation model can realize the function of up-order reconstruction of the time-series pressure electrical signal, and further comprehensively reflect the working state of the measurement object. It overcomes the problem that the existing method judges the on-the-job state of the employee to be audited by using the target similarity function, which is evaluated based on a single or a few similarity indexes, is difficult to meet the evaluation requirements based on the pressure sensor, and cannot comprehensively reflect the working state of the measurement object.

[0174] It should be noted that for the foregoing embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps may be in other sequences or performed simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0175] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the above division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection between devices or units can be in the form of telecommunications or other forms.

[0176] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the protection scope of the invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on these embodiments, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art can still, without conflict, make combinations, additions, deletions, or other adjustments to the features in the embodiments of the present invention according to the situation without creative efforts, so as to obtain different technical solutions that are essentially not divorced from the concept of the present invention, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A method for evaluating the working state of a measurement object based on a pressure sensor, characterized in that, The method for evaluating the working state of a measurement object based on a pressure sensor includes: Acquire time-series pressure electrical signals based on a pressure sensor, preprocess the time-series pressure electrical signals based on a characteristic frequency combined with an adaptive filtering algorithm to obtain a preprocessing set, and upload the preprocessing set to a state database; Pre-construct a state evaluation model, extract a modeling sample set from the state database, divide the modeling sample set into a training set and a test set according to a ratio of 3:1, and train the state evaluation model using the training set and the test set to output a converged state evaluation model; Load the preprocessing set in real time, use the preprocessing set as an input, execute the state evaluation model, and the state evaluation model identifies and analyzes the preprocessing set to output an evaluation result of the working state of the measurement object; Obtain the evaluation result of the working state, and based on whether the comprehensive evaluation value in the evaluation result of the working state exceeds a preset evaluation threshold, if it exceeds the preset evaluation threshold, trigger a working state warning instruction; The method for preprocessing the time-series pressure electrical signals based on a characteristic frequency combined with an adaptive filtering algorithm specifically includes: Load the time-series pressure electrical signals, divide the time-series pressure electrical signals into at least one group of characteristic frequency sets based on a preset sampling period, and process outliers and missing values in the characteristic frequency sets; Load the characteristic frequency sets, identify discrete frequency distributions, continuous frequency distributions, and no-pressure frequency distributions in the characteristic frequency sets based on a preset judgment period, and assign labels to the sampling periods corresponding to the discrete frequency distributions, continuous frequency distributions, and no-pressure frequency distributions, where the discrete frequency distribution is marked as the working state, the continuous frequency distribution is marked as the to-be-identified state, and the no-pressure frequency distribution is marked as the state of leaving the work station; Load the continuous frequency distribution, obtain the spectrum sequence corresponding to the continuous frequency distribution based on the fast Fourier transform, and calculate the amplitude energy and phase of the spectrum sequence; Among them, the amplitude energy of the spectrum sequence is calculated by the following formula: （1） （2） Among them, represents the amplitude energy of the spectrum sequence, is the phase of the spectrum sequence, represents the number of continuous signal points, represents the input representation of the continuous frequency distribution based on the fast Fourier transform, is the signal frequency; Use a clustering algorithm to identify the central amplitude energy within a preset judgment period, load the amplitude energy and phase of the spectrum sequence, and determine whether there is an approximate signal frequency within the preset judgment period; If there is no approximate frequency, no suppression filtering process is performed on the signal frequency within the preset judgment period; If there is an approximate frequency, perform suppression processing on the approximate signal frequency based on a low-pass filter, combine the low-pass filter to obtain the amplitude energy corresponding to the approximate signal frequency, and output the continuous frequency distribution after the suppression filtering process; Load the continuous frequency distribution, discrete frequency distribution, no-pressure frequency distribution after the suppression filtering process and their corresponding marked states, and integrate them to obtain a preprocessing set; The method for training the state evaluation model using the training set and the test set specifically includes: Load the pre-constructed state evaluation model, training set, and test set, and set the hyperparameters, number of iteration rounds, and training batches of the state evaluation model; Obtain the training set, optimize the performance of the state evaluation model using a cross-entropy loss function, and adjust the connection weights between the encoder, decoder, and convolutional layer in the hidden layer in combination with the backpropagation algorithm; Train the state classifier in the state evaluation model using a root mean square error loss function. For the embedded feature fusion result generated by the feature fusion layer, calculate the mean square error loss with the true classification result respectively. The root mean square error loss function is expressed as: （5） Among them, respectively represent the elements in the th row and th column of the feature fusion matrix for the embedded feature fusion result and the true feature fusion result, respectively represent the elements in the th row and th column of the feature fusion matrix for the embedded feature fusion label and the true feature fusion label; Based on the state evaluation model whose training converges within the preset number of iteration rounds, obtain a test set, test the state evaluation model based on the test set, and output the test results; Load the test results, and verify whether the state evaluation model meets the preset model accuracy based on the test results. If it meets the preset model accuracy, output the converged state evaluation model; If it does not meet the preset model accuracy, use the Nadam optimizer to update the learning rate and hyperparameters of the state evaluation model, and continue to iteratively train the state evaluation model through the training set.

2. The method for evaluating the working state of a measurement object based on a pressure sensor according to claim 1, wherein: The method for suppressing the approximate signal frequency based on a low-pass filter specifically further includes: Load the approximate signal frequency and the amplitude energy corresponding to the approximate signal frequency within the preset judgment period, and calculate the overlap similarity value between the approximate signal frequency and the central signal frequency corresponding to the central amplitude energy; Suppress the harmonics of the approximate signal frequency based on the low-pass filter combined with the overlap similarity value, and output the filtered approximate signal frequency; Obtain the filtered approximate signal frequency, and perform an inverse fast Fourier transform using the filtered approximate signal frequency to complete the suppression of the approximate signal frequency; Among them, the overlap similarity value is calculated by the following formula: （3） Among them, represents the overlapping similarity value, is the number of iterations of the low-pass filter for filtering, respectively represent the amplitude energy corresponding to the approximate signal frequency and the central amplitude energy, are the approximate signal frequency and the central signal frequency respectively; When performing an inverse fast Fourier transform using the filtered approximate signal frequency, the inverse fast Fourier transform formula is: （4） Among them, represents the overlapping similarity value, is the sampling angular difference of the approximate signal frequency, is the sampling frequency of the approximate signal.

3. The method for evaluating the working state of a measurement object based on a pressure sensor according to claim 1, wherein: When using the Nadam optimizer to update the learning rate and hyperparameters of the state evaluation model, the first moment and the second moment of the gradient tracked by the Nadam optimizer are expressed as: （6） （7） （8） Among them, represents the hyperparameters of the state evaluation model, is the learning rate, is the gradient of the loss function at the hyperparameters of the state evaluation model, are the first-order moment and second-order moment estimates of the gradient respectively, are the first and second decay rate parameters.

4. The method for evaluating the working state of a measurement object based on a pressure sensor according to claim 3, wherein: The state evaluation model uses a multi-layer perceptron as the initial model. The initial model consists of an input layer, a hidden layer, and an output layer. A staggered time function is introduced into the input layer to improve the hidden layer. An encoder and a decoder are introduced into the hidden layer, and three convolutional layers are set between the encoder and the decoder. The convolutional layer is a 3×3 convolution, and the activation function of the hidden layer is the ReLU activation function; The encoder consists of two groups of max-pooling layers and a fully connected layer. The fully connected layer is respectively connected to the max-pooling layer. The decoder consists of two groups of average-pooling layers and a deconvolution layer. The deconvolution kernel size of the deconvolution layer is 3×3; When pre-constructing the state evaluation model, freeze the output layer of the initial model, and use a feature fusion layer and a state classifier to replace the output layer of the initial model. The state classifier uses a three-layer multi-layer perceptron to implement the output of the state classification result. The state classifier is expressed as: （9） Among them, represents the embedded feature fusion result output by the feature fusion layer, represents the state classification result, is the feature of the output of the third-layer Neck network of the node, is the number of nodes of the Neck network.

5. The method for evaluating the working state of a measurement object based on a pressure sensor according to claim 4, wherein: The method for the state evaluation model to identify and analyze the preprocessing set specifically includes: Load the preprocessing set. The input layer determines the to-be-identified state corresponding to the continuous frequency distribution in the preprocessing set based on the staggered time function. The staggered time function captures the characteristic frequency of the continuous frequency distribution and determines and marks the working state of the continuous frequency distribution; Among them, the staggered time function is expressed as: （10） Among them, represents the off-peak time function, respectively represent the frequency timestamps between the approximate signal frequency points in the continuous frequency distribution, represents the frequency preset delay, is the function attenuation coefficient; Obtain the characteristic frequency of the continuous frequency distribution and the determined mark. The encoder uses the Gauss-Newton method to matrix-elevate and express the characteristic frequency of the continuous frequency distribution, perform a convolution operation on the matrix-elevated characteristic frequency, and output the result of the convolution operation; Load the result of the convolution operation. The decoder splices the result of the convolution operation based on skip connections and average pooling, and outputs the frequency reconstruction result; Obtain the frequency reconstruction result, and the feature fusion layer performs horizontal splicing of dimensions and features on the frequency reconstruction result. The state classifier outputs the state classification result based on the Neck network output; Evaluate the state classification result based on the cascading failure combined with the structural similarity algorithm, calculate the comprehensive evaluation value, and output the comprehensive evaluation value.

6. The method for evaluating the working state of a measurement object based on a pressure sensor according to claim 5, characterized in that: The comprehensive evaluation value is calculated by the following formula: （11） Among them, represents the comprehensive evaluation value, is the weight of the cascading failure nodes and represents the state classification result, is the structural similarity index, is the bias constant.

7. A system for evaluating the working state of a measurement object based on a pressure sensor, which is used to implement the method for evaluating the working state of a measurement object based on a pressure sensor according to any one of claims 1-6, characterized in that: The system for evaluating the working state of a measurement object based on a pressure sensor includes: A preprocessing module, which obtains time-series pressure electrical signals based on a pressure sensor, preprocesses the time-series pressure electrical signals based on the characteristic frequency combined with an adaptive filtering algorithm to obtain a preprocessing set, and uploads the preprocessing set to the state database; A model construction module, which is used to pre-construct a state evaluation model, extract a modeling sample set from the state database, divide the modeling sample set into a training set and a test set according to a ratio of 3:1, and use the training set and the test set to train the state evaluation model, and output a converged state evaluation model; A state evaluation module, which is used to load the preprocessing set in real time, use the preprocessing set as input, execute the state evaluation model, and the state evaluation model identifies and analyzes the preprocessing set to output the working state evaluation result of the measurement object; A state judgment module, which is used to obtain the working state evaluation result, and based on whether the comprehensive evaluation value in the working state evaluation result exceeds a preset evaluation threshold, if it exceeds the preset evaluation threshold, trigger a working state warning instruction.

8. The system for evaluating the working state of a measurement object based on a pressure sensor according to claim 7, wherein: The state judgment module includes: A result identification unit, which is used to obtain the working state evaluation result; A state judgment unit, based on whether the comprehensive evaluation value in the working state evaluation result exceeds a preset evaluation threshold; A state warning unit, which is used to trigger a working state warning instruction when the working state evaluation result exceeds a preset evaluation threshold.

Images