Electro-hydrogen coupling system equipment life prediction method, equipment and medium

Abstract

The invention belongs to the technical field of equipment life prediction, and particularly relates to an electricity-hydrogen coupling system equipment life prediction method, equipment and a medium. In order to overcome the defects that the prediction precision of an existing equipment life prediction method integrating LSTM and CNN is reduced and historical data information cannot be well utilized, the electricity-hydrogen coupling system equipment life prediction method comprises the steps of obtaining state operation data of equipment in a past period of time, performing preprocessing, and dividing the state operation data into a training set and a test set; constructing a CNN-LSTM fusion neural network model, and training the model through the training set; a multi-step rolling average mode is adopted to predict the service life of the electro-hydrogen coupling system equipment, a continuous service life prediction sequence is generated, and verification is carried out through a test set; and applying the trained model to equipment life prediction. According to the method, the advantages of the CNN and the LSTM are combined; and by adopting multi-step rolling average prediction, the method has the advantages of high prediction precision and good real-time performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of equipment life prediction, and in particular relates to a method, equipment and medium for predicting the equipment life of an electric-hydrogen coupling system. Background Art

[0002] With the continuous development of energy technology, electricity-hydrogen coupling has become an important morphological feature of the development of energy systems. As an efficient and clean energy conversion device, hydrogen-to-electric fuel cells have shown great application potential in the fields of new energy consumption, transportation, distributed power generation, etc. However, the life prediction of hydrogen-to-electric fuel cells has always been one of the key factors restricting their widespread application. Accurate life prediction not only helps to optimize the maintenance strategy of hydrogen-to-electric fuel cell systems, but also significantly reduces operating costs and improves system reliability and safety. The electrolyzer of the hydrogen-to-electric coupling system is also similar to that of hydrogen-to-electric fuel cells.

[0003] Traditional methods for predicting the life of hydrogen-to-electric fuel cells mainly rely on physical models or empirical formulas. These methods require a deep understanding of the internal mechanism and working environment of hydrogen-to-electric fuel cells during the modeling process, and the prediction accuracy is often affected by multiple factors. However, the life of key equipment in the hydrogen-to-electric coupling system has more factors, the degradation mechanism is more complex, and the spatial and temporal correlation within the data is more difficult to mine. In recent years, with the rapid development of artificial intelligence technology, data-driven prediction methods have gradually become a research hotspot. Among them, long short-term memory networks (LSTM) and convolutional neural networks (CNN), as representative algorithms in the field of deep learning, have shown unique advantages in predicting the life of hydrogen-to-electric fuel cells. LSTM networks are good at processing time series data and can capture the dynamic characteristics of hydrogen-to-electric fuel cells that change over time during operation, such as the fluctuation of parameters such as voltage, current, and temperature. CNN networks are good at extracting spatial features of data and can identify key indicators and patterns in the performance degradation process of hydrogen-to-electric fuel cells.

[0004] The prediction methods based on the neural network model include single-step prediction, multi-step prediction and rolling prediction. Single-step prediction is generally used for short-term prediction. The accuracy of multi-step prediction is not easy to guarantee. The disadvantage of rolling prediction is that the prediction accuracy will drop sharply after multiple rolling. It is commonly used to predict data trends by combining multi-step prediction and rolling prediction, that is, predicting the same number of points each time and using the predicted results as input data for subsequent predictions.

[0005] Some scholars have proposed combining LSTM with CNN to construct a LSTM-CNN fusion neural network model for battery life prediction. CN114636932A discloses a method and system for predicting the remaining service life of a battery. The main process includes: obtaining data during the battery charging and discharging process, inputting it into the LSTM-CNN model, CNN is used to extract the spatial correlation features of the data, and LSTM is used to extract the time features of the spatial correlation features to predict the remaining service life of the battery. CN114839553A discloses a method and device for predicting the remaining service life of a battery. The specific steps are: performing charge and discharge experiments on multiple sample batteries to obtain experimental data of multiple charge and discharge cycles of each sample battery; obtaining aging characteristic data of each sample battery according to the experimental data of multiple charge and discharge cycles of each sample battery; performing preprocessing and feature engineering on the aging characteristic data of each sample battery, forming a data set with selected features, and dividing the data set into a training set and a test set; constructing a combined model of CNN and LSTM, wherein the CNN removes the fully connected layer and outputs a one-dimensional feature vector, and the output of the CNN is used as the input of the LSTM; training the combined model of CNN and LSTM through the training set to obtain a battery remaining service life predictor, and testing the battery remaining service life predictor through the test set; and predicting the remaining service life of the battery to be predicted through the finally obtained battery remaining service life predictor.

[0006] However, the multi-step rolling prediction model of LSTM and CNN fusion disclosed in the aforementioned patent application fails to mine historical data information well, and the accumulation of prediction errors caused by accidental factors in training and prediction will cause a decrease in prediction accuracy after multiple rolling. Summary of the invention

[0007] The present invention aims at the disadvantages that the prediction accuracy of the existing battery life prediction method integrating LSTM and CNN is reduced and the historical data information is not well utilized, and a method for predicting the life of an electric hydrogen coupling system device is provided to improve the prediction accuracy. The present invention also provides a computer device and a computer-readable storage medium.

[0008] To achieve the above object, the present invention adopts the following technical solution: a method for predicting the life of equipment in an electric-hydrogen coupling system, the method comprising: Step S1, obtaining the state operation data of the battery of the electric-hydrogen coupling system in the past period of time and preprocessing it, and dividing the data obtained after the preprocessing into a training set and a test set; Step S2, constructing a CNN-LSTM fusion neural network model, and training the model using a training set; Step S3, using a multi-step rolling average mode to predict the battery life of the electric-hydrogen coupling system, generating a continuous life prediction sequence, and verifying the model through a test set; Step S4: Apply the trained model to the actual battery life prediction of the electric-hydrogen coupling system.

[0009] As an improvement, in step S1, the state operation data includes at least one of output voltage, output current, output power, air flow rate, hydrogen flow rate, reaction gas temperature, reaction gas humidity, pressure, cooling gas temperature and cooling gas flow rate; and / or, In step S1, the electricity-hydrogen coupling system equipment is a fuel cell or an electrolyzer.

[0010] As an improvement, in step S1, the specific process of preprocessing includes: Step S11, using the absolute median difference to remove outliers; Step S12, using SG filtering method to remove noise; Step S13: normalize the data using the min-max normalization method. The calculation formula of the min-max normalization is as follows:

[0011] in, is the normalized data, is the minimum value in the data, is the maximum value in the data.

[0012] As an improvement, the training process of step S2 includes: Step S21, using a sliding window module to convert the standardized data sample set into a feature matrix as the input of the input layer; Step S22, performing a convolution operation on the input layer through a convolution layer: using multiple convolution kernels to calculate the corresponding feature sequence, stacking to form a multi-dimensional feature sequence array, and the feature sequence array is processed by a ReLU activation function to introduce nonlinear factors; Step S23, pooling the feature sequence array through a pooling layer: compressing the feature sequence array with a maximum pooling function, removing redundant feature information, extracting the most important feature values, and reducing the feature data dimension; Step S24, compressing the multi-dimensional information after pooling into one-dimensional data through the Flatten layer and inputting it into the LSTM layer; Step S25, feature learning is performed on the extracted feature data through the LSTM layer, long-term time series features are extracted, and predicted values ​​are output; wherein, through the Dropout layer, the units of the LSTM layer are temporarily discarded from the network according to a certain probability, so that the activation values ​​are discarded in a certain proportion during the network training process, thereby improving the generalization performance of the neural network and solving the problem of network overfitting.

[0013] As an improvement, in step S3, the multi-step rolling average mode is: applying different forecast input samples including long-term time series to make multiple forecasts on the data at a certain point in the future to obtain multiple forecast values, and averaging these forecast values ​​as the final forecast value.

[0014] As an improvement, in step S3, a multi-step rolling average mode is used to predict the life of the equipment in the electric-hydrogen coupling system, and the specific process of generating a continuous life prediction sequence includes: Step S31: Assume that the input feature vector size is m, the output feature vector size is n, and the historical data is , using the last m+n-1 data in the historical data as the initial data for rolling forecast, the model input data is: ; Step S32: input the model input data into the CNN-LSTM fusion neural network model in step S2, and the obtained prediction data is: ; Step S33: n indivual Find the average value as the t +1 predicted value ; Step S34: i =2; Step S35: construct input vector Input_Predict for: ; Step S36: Input_Predict i is input into the prediction model to obtain the output vector Output_Predict i is: ; Step S37: Output_Predict i and Output_Predict i-1 Merge to form a new prediction output matrix Output_Predict : ; Step S38: n indivual Find the average value as the t +i The predicted value of the data ; Step S39: Verification Is it established? Pred_Num is the predicted length. If it does not hold, then i = i+ 1 and return to step S35 to continue calculation; if established, then from the 1st to the Pred_Num Total Pred_Num The predicted values ​​are taken as the prediction results, namely: .

[0015] As an improvement, step S3 also includes: using standard root mean square error and standard mean absolute error for evaluation to optimize parameters in the CNN-LSTM fusion neural network model.

[0016] As an improvement, in step S4, the data acquisition system is used to collect real-time operating status data of the battery of the electric-hydrogen coupling system, extract the output voltage data, construct an input feature matrix, and input it into the trained CNN-LSTM fusion neural network model to predict the output voltage change trend in real time, thereby realizing the remaining life prediction of the battery of the electric-hydrogen coupling system.

[0017] The computer device includes a processor and a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by the processor, the life prediction method of the electric-hydrogen coupling system equipment is implemented.

[0018] A computer-readable storage medium stores a computer program, which, when executed, implements the aforementioned method for predicting the life of equipment in an electric-hydrogen coupling system.

[0019] The device life prediction method for an electric-hydrogen coupling system based on a CNN-LSTM fusion neural network of the present invention takes the output voltage of a device such as an electric-hydrogen fuel cell or an electrolyzer as a performance degradation indicator, utilizes CNN to mine the internal spatial characteristics of output voltage data, utilizes LSTM to process the time correlation of voltage data, and adopts a multi-step rolling average prediction mode to fully mine historical data information, and has the advantages of high prediction accuracy and good real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a method for predicting the life of equipment in an electric-hydrogen coupling system according to an embodiment of the present invention.

[0021] Figure 2 It is a schematic diagram of the structure of the CNN-LSTM fusion neural network adopted in an embodiment of the present invention.

[0022] Figure 3It is a schematic diagram of the LSTM of the CNN-LSTM fusion neural network adopted in an embodiment of the present invention.

[0023] Figure 4 It is an explanatory diagram of a multi-step rolling average prediction mode of a method for predicting equipment life of an electric-hydrogen coupling system according to an embodiment of the present invention.

[0024] Figure 5 It is a comparison diagram of output voltage prediction results obtained by using the method of the embodiment of the present invention and the existing method.

[0025] Figure 6 yes Figure 5 Comparison chart of the test set stage in . DETAILED DESCRIPTION

[0026] The technical solutions of the embodiments of the present invention are explained and described below, but the following embodiments are only preferred embodiments of the present invention, not all. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work are all within the protection scope of the present invention.

[0027] See also Figure 1 The method for predicting the life of an electric-hydrogen coupling system device according to an embodiment of the present invention includes: Step S1, obtaining the state operation data of the battery of the electric-hydrogen coupling system in the past period of time and preprocessing it, and dividing the data obtained after the preprocessing into a training set and a test set; Step S2, constructing a CNN-LSTM fusion neural network model, and training the model using a training set; Step S3, using a multi-step rolling average mode to predict the battery life of the electric-hydrogen coupling system, generating a continuous life prediction sequence, and verifying the model through a test set; Step S4: Apply the trained model to the actual battery life prediction of the electric-hydrogen coupling system.

[0028] In this embodiment, the electric-hydrogen coupling system device is an electric-to-hydrogen fuel cell. In other embodiments, the electric-hydrogen coupling system device may also be an electrolyzer.

[0029] In this embodiment, in step S1, the state operation data includes output voltage, output current, output power, air flow rate, hydrogen flow rate, reaction gas temperature, reaction gas humidity, pressure, cooling gas temperature and cooling gas flow rate. Among them, the output voltage data is the key data. The output voltage of the electric hydrogen fuel cell is a key indicator to characterize the performance of the electric hydrogen fuel cell, which can be directly measured by the sensor. By monitoring the degradation law of the output voltage of the electric hydrogen fuel cell, the life prediction of the electric hydrogen fuel cell can be achieved. The voltage signal collected during the operation of the electric hydrogen fuel cell may be subject to noise interference, communication obstruction, etc., resulting in null values ​​and interference data in the original electric hydrogen fuel cell output voltage data. If these abnormal data are not processed, the accuracy of the subsequent model prediction will be affected. Therefore, it is necessary to process the original electric hydrogen fuel cell output voltage data to remove the null values ​​and abnormal values ​​in the original data.

[0030] In this embodiment, in step S1, the original output voltage data is preprocessed, and the specific process includes: Step S11, use the absolute median difference to remove outliers. There are some outliers in the output voltage data of the hydrogen production fuel cell that are too different from the majority of the data. Outliers may have a negative impact on the accuracy and stability of the model. Calculate the median of the data set, calculate the absolute difference between each data point and the median based on the calculated median, and calculate the median of these differences as MAD (mean absolute deviation). Then, by calculating the absolute deviation score of the data point, the data points that exceed the preset threshold are regarded as outliers and removed; Step S12, use SG filtering (Savitzky-Golay filter) to remove noise. The output voltage data of hydrogen production fuel cells usually contains noise, which may come from measurement errors, sensor failures or other external interference. Noise will increase the amount of calculation of the model, affecting the convergence speed of the data and the accuracy of the prediction. Data smoothing can remove noise and irregularities, highlighting the regularity and changing trends of the data. SG filtering performs local polynomial fitting on the original data and uses the polynomial to replace the original data points; Step S13: normalize the data using the min-max normalization method. The calculation formula for min-max normalization is as follows:

[0031] in, is the normalized data, is the minimum value in the data, is the maximum value in the data.

[0032] In this embodiment, a method for predicting the life of a hydrogen-producing fuel cell based on CNN-LSTM is constructed to improve the accuracy of predicting the life of a hydrogen-producing fuel cell. The advantage of CNN lies in feature extraction, and LSTM can effectively process time series data. In order to give full play to the advantages of the two models, the two are combined to form a CNN-LSTM fusion neural network model. The combination of convolutional layer and LSTM layer enables the prediction model to extract significant features from the original data and mine the time series features hidden in the historical trend.

[0033] In this embodiment, the training process of step S2 includes: Step S21, using a sliding window module to convert the standardized data sample set into a feature matrix as the input of the input layer; Step S22, performing a convolution operation on the input layer through a convolution layer: using multiple convolution kernels to calculate the corresponding feature sequence, stacking to form a multi-dimensional feature sequence array, and the feature sequence array is processed by a ReLU activation function to introduce nonlinear factors; Step S23, pooling the feature sequence array through a pooling layer: compressing the feature sequence array with a maximum pooling function, removing redundant feature information, extracting the most important feature values, and reducing the feature data dimension; Step S24, compressing the multi-dimensional information after pooling into one-dimensional data through the Flatten layer and inputting it into the LSTM layer; Step S25: Through the Dropout layer, the units of the LSTM layer are temporarily discarded from the network according to a certain probability, so that the activation values ​​are discarded according to a certain ratio during the network training process, thereby improving the generalization performance of the neural network and solving the problem of network overfitting; Step S26: Perform feature learning on the extracted feature data through the LSTM layer, extract long-term time series features, and output predicted values.

[0034] In this embodiment, see Figure 2 ,The CNN-LSTM fusion neural network model in step S2 includes input layer, convolution layer, pooling layer, Flatten layer, Dropout layer, LSTM layer, fully connected layer and output layer in sequence. Figure 2In the figure, ① is the input layer, ② is the convolution layer, ③ is the pooling layer, ④ is the Flatten layer, ⑤ is the Dropout layer, ⑥ is the LSTM layer, ⑦ is the fully connected layer, and ⑧ is the output layer. The Dropout layer is actually a rule restriction on the LSTM layer, that is, the processing units in the LSTM layer are selectively discarded, so that some processing units in the LSTM layer are not activated. The output voltage data of the hydrogen fuel cell preprocessed in step S1 is divided into a training set and a test set according to a certain ratio (for example, 7:3). Determine the sliding window width, determine the input and output feature matrices of the CNN-LSTM fusion neural network model, train the model, and determine the model parameters.

[0035] In this embodiment, see Figure 3 , the LSTM layer introduces a gate mechanism, adds input gates, output gates, and forget gates inside the hidden layer, and adds units for storing memory. Each LSTM neuron is composed of a cell state, that is, a long-term state and real-time status , Input Gate , Forget Gate , output gate Consists of.

[0036] The forget gate determines which information should be deleted and which information should be remembered; the input gate controls which information can be input; and the output gate controls which information can be output to the next neuron.

[0037] Input Gate control What information should be kept if the input gate is closed, i.e. When, real-time status No new status information will be written. The saved status information is only the same as the status at the last moment. The input gate uses the sigmoid function to select which information to keep and the tanh function to construct a new vector to form a new cell state. The calculation formula is as follows: ; ; Where: (.) is the sigmoid function, is the weight matrix of the input gate; Indicates the output at the previous moment; Represents the input at the current moment; is the bias term of the input gate; tanh(.) is the linear rectification activation function; The weight matrix calculated for the state value; Bias term for state value calculation; Forget Gate Decide How much should be kept and discarded to the current state? If the forget gate is closed, that is, , the historical information stored in the memory cell will be deleted, and the cell state at the previous moment will not affect the cell state at the current moment. The calculation formula of the forget gate is as follows: ; Where: is the weight matrix of the forget gate; is the bias term of the forget gate; Update according to the following formula The value of Determine the output results : ; In the formula, Represents the output at the current moment, is the weight matrix of the output gate; is the bias term of the output gate; for t The output result obtained by the output gate at each moment.

[0038] In this embodiment, in step S3, the prediction performance of the constructed prediction model in the test set is evaluated using the standard root mean square error (RMSE) and the standard mean absolute error (MAE). The formulas of RMSE and MAE are as follows:

[0039]

[0040] in, is the predicted value, is the actual value.

[0041] Step S3 also includes: using standard root mean square error and standard mean absolute error for evaluation to optimize parameters in the CNN-LSTM fusion neural network model.

[0042] In this embodiment, see Figure 4 In step S3, the battery life of the electric-hydrogen coupling system is predicted using a multi-step rolling average mode, and the specific process of generating a continuous life prediction sequence includes: Step S31: Assume that the input feature vector size is m=6, the output feature vector size is n=3, and the historical data is x 1, x 2, x 3, x 4, x 5, x 6, x7, x 8, x 9, x 10 , x 11 , x 12 , take the last 8 data in the historical data as the initial data of rolling forecast, and construct the model input data as: ; Step S32: input it into the CNN-LSTM fusion neural network model in step S2, and the prediction data can be obtained as follows: ; Step S33: 3 of them x 13 Find the average value as the predicted value of the 13th data x_pred 13 ; Step S34: i =2; Step S35: construct input vector Input_Predict 2 is: ; Step S36: Input_Predict 2 Input to the prediction model and get the output vector Output_Predict 2 is: ; Step S37: Output_Predict 2 and Output_Predict 1 merge to form a new prediction output matrix Output_Predict : ; Step S38: 3 of them x 14 Find the average value as the predicted value of the 14th data x_pred 14 ; Step S39: Verification i ≥ Pred_Num Is it established? Pred_Num is the predicted length. If not, then i = i+ 1 and return to step (5) to continue the calculation; if it is established, then Pred_Num Total Pred_Num The predicted values ​​are taken as the prediction results, namely: .

[0043] In this embodiment, in step S4, the data acquisition system is used to collect real-time operating status data of the hydrogen production fuel cell system, extract output voltage data, construct an input feature matrix, and input it into the trained CNN-LSTM fusion neural network model to predict the output voltage change trend in real time and realize the remaining life prediction of the hydrogen production fuel cell.

[0044] See also Figure 5 , the life prediction method of the electric hydrogen coupling system equipment of this embodiment is applied to predict the life of the battery and compared with the prediction results of the traditional CNN-LSTM fusion neural network. The model is trained using the first 750 hours of operation data. The improved CNN-LSTM fusion neural network multi-step rolling average prediction method proposed in this embodiment and the traditional CNN-LSTM fusion neural network multi-step rolling prediction method both show good performance in the training set. The trained model is tested using data after 750 hours.

[0045] See also Figure 6 At the beginning of the test set prediction, the improved CNN-LSTM fusion neural network multi-step rolling average prediction method proposed in this embodiment is close to the traditional CNN-LSTM fusion neural network multi-step rolling prediction method in performance. Although in the stage of 750 hours to 1000 hours, the performance of the method of the embodiment of the present invention is lower than the traditional method in a small period of time. However, after a sudden change in the system data structure around 1000 hours, the performance of the improved CNN-LSTM fusion neural network multi-step rolling average prediction method proposed in this embodiment is significantly better than the traditional method and can better predict future trends.

[0046] The multi-step rolling average life prediction method of a hydrogen production fuel cell based on a CNN-LSTM fusion neural network in an embodiment of the present invention takes the output voltage of the hydrogen production fuel cell as an indicator of the performance degradation of the hydrogen production fuel cell system, uses CNN to mine the internal spatial characteristics of the output voltage data, and uses LSTM to process the time correlation of the voltage data, thereby taking advantage of both CNN and LSTM; adopts a multi-step rolling average prediction mode, which can fully mine historical data information; has the advantages of high prediction accuracy and good real-time performance; the CNN-LSTM fusion neural network model has a Dropout layer, which can temporarily discard the units of the LSTM layer from the network according to a certain probability, so that the activation values ​​are discarded according to a certain ratio during the network training process, thereby improving the generalization performance of the neural network and solving the problem of network overfitting.

[0047] An embodiment of the present invention further provides a computer device, including a processor and a storage medium, wherein a computer program is stored in the storage medium, and when the computer program is executed by the processor, the aforementioned method for predicting the life of equipment in an electric-hydrogen coupling system is implemented.

[0048] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the aforementioned method for predicting the life of equipment in an electric-hydrogen coupling system is implemented.

[0049] The above is only a specific implementation of the invention, but the protection scope of the invention is not limited thereto. Those skilled in the art should understand that the invention includes but is not limited to the contents described in the above specific implementation. Any modification that does not deviate from the functional and structural principles of the invention will be included in the scope of the claims.

Claims

1. A method for predicting the life of equipment in an electric-hydrogen coupling system, characterized in that: The method for predicting the life of equipment in the electric-hydrogen coupling system comprises: Step S1, obtaining the state operation data of the electric-hydrogen coupling system equipment in the past period of time and preprocessing it, and dividing the data obtained after the preprocessing into a training set and a test set; Step S2, constructing a CNN-LSTM fusion neural network model, and training the model using a training set; Step S3, using a multi-step rolling average mode to predict the life of the electric-hydrogen coupling system equipment, generating a continuous life prediction sequence, and verifying the model through a test set; Step S4: Apply the trained model to the actual life prediction of the equipment in the electric-hydrogen coupling system.

2. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 1, characterized in that: In step S1, the state operation data includes at least one of output voltage, output current, output power, air flow rate, hydrogen flow rate, reaction gas temperature, reaction gas humidity, pressure, cooling gas temperature and cooling gas flow rate; and / or, In step S1, the electricity-hydrogen coupling system equipment is a fuel cell or an electrolyzer.

3. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 1, characterized in that: In step S1, the specific process of preprocessing includes: Step S11, using the absolute median difference to remove outliers; Step S12, using SG filtering method to remove noise; Step S13: normalize the data using the min-max normalization method. The calculation formula for min-max normalization is as follows: in, is the normalized data, is the minimum value in the data, is the maximum value in the data.

4. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 1, characterized in that: The training process of step S2 includes: Step S21, using a sliding window module to convert the standardized data sample set into a feature matrix as the input of the input layer; Step S22, performing a convolution operation on the input layer through a convolution layer: using multiple convolution kernels to calculate the corresponding feature sequence, stacking to form a multi-dimensional feature sequence array, and the feature sequence array is processed by a ReLU activation function to introduce nonlinear factors; Step S23, pooling the feature sequence array through a pooling layer: compressing the feature sequence array with a maximum pooling function, removing redundant feature information, extracting the most important feature values, and reducing the feature data dimension; Step S24, compressing the multi-dimensional information after pooling into one-dimensional data through the Flatten layer and inputting it into the LSTM layer; Step S25, feature learning is performed on the extracted feature data through the LSTM layer, long-term time series features are extracted, and predicted values ​​are output; wherein, through the Dropout layer, the units of the LSTM layer are temporarily discarded from the network according to a certain probability, so that the activation values ​​are discarded in a certain proportion during the network training process, thereby improving the generalization performance of the neural network and solving the problem of network overfitting.

5. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 1, characterized in that: In step S3, the multi-step rolling average mode is: applying different forecast input samples including long-term time series to make multiple forecasts on the data at a certain time point in the future to obtain multiple forecast values, and calculating the average value of these forecast values ​​as the final forecast value.

6. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 5, characterized in that: In step S3, the life of the equipment in the electric-hydrogen coupling system is predicted using a multi-step rolling average mode, and the specific process of generating a continuous life prediction sequence includes: Step S31: Assume that the input feature vector size is m, the output feature vector size is n, and the historical data is , using the last m+n-1 data in the historical data as the initial data for rolling forecast, the model input data is: ； Step S32: input the model input data into the CNN-LSTM fusion neural network model in step S2, and the obtained prediction data is: ； Step S33: n indivual Find the average value as the t +1 predicted value ; Step S34: i =2; Step S35: construct input vector Input_Predict for: ； Step S36: Input_Predict i is input into the prediction model to obtain the output vector Output_Predict i is: ； Step S37: Output_Predict i and Output_Predict i-1 Merge to form a new prediction output matrix Output_Predict : ； Step S38: n indivual Find the average value as the t +i The predicted value of the data ; Step S39: Verification Is it established? Pred_Num is the predicted length. If it does not hold, then i = i+ 1 and return to step S35 to continue calculation; if established, then from the 1st to the Pred_Num Total Pred_Num The predicted values ​​are taken as the prediction results, namely: 。 7. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 1, characterized in that: Step S3 also includes: using standard root mean square error and standard mean absolute error for evaluation to optimize parameters in the CNN-LSTM fusion neural network model.

8. The method for predicting the life of equipment in an electric-hydrogen coupling system according to claim 1, characterized in that: In step S4, the data acquisition system is used to collect real-time operating status data of the electric-hydrogen coupling system equipment, extract output voltage data, construct an input feature matrix, and input it into the trained CNN-LSTM fusion neural network model to predict the output voltage change trend in real time and realize the remaining life prediction of the electric-hydrogen coupling system equipment.

9. A computer device comprising a processor and a storage medium, wherein the storage medium stores a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the life of equipment in an electric-hydrogen coupling system as claimed in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the method for predicting the life of equipment in an electric-hydrogen coupling system as claimed in any one of claims 1 to 8 is implemented.

Images