Circuit damage prediction method for battery simulator based on multi-time series feature convolution

Abstract

The present invention discloses a method for predicting damage of a battery simulator circuit based on multi-time series feature convolution, the method comprising: using an exponentially weighted moving average filter method to denoise high-frequency current data to obtain a denoised fault data set; using wavelet transform to extract fault spectrum features from the denoised fault data set; using a variational autoencoder optimized by 3-fold cross-validation to extract fault coding features from the denoised fault data set; using a sparse deep belief network to extract fault depth features from the denoised fault data set; and concatenating the three fault features head to tail and inputting them into a Beluga optimized convolutional neural network to predict damage to the battery simulator circuit. The present invention can predict the circuit damage of a battery simulator and avoid the losses caused by sudden failures, and has significant advantages in predicting damage to the battery simulator circuit.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence deep learning fault diagnosis, and specifically relates to a battery simulator circuit damage prediction method based on multi-time series feature convolution. Background Art

[0002] As energy crisis and environmental pollution become increasingly serious, the new energy industry has developed rapidly. Batteries are key devices for new energy storage and conversion, and battery simulators are devices used to simulate battery behavior. They play an important role in the research and development of electric vehicles, renewable energy storage, portable electronic devices and other fields. Traditional battery testing methods usually require a lot of experiments and field tests, which are not only time-consuming and labor-intensive, but also costly. Battery simulators can test and evaluate batteries in a virtual environment, thereby significantly reducing R&D costs and time. By predicting damage to the battery simulator circuit, it helps to identify potential safety risks. Summary of the invention

[0003] The purpose of the present invention is to provide a battery simulator circuit damage prediction method based on multi-time series feature convolution, through the battery simulator circuit damage prediction, thereby reducing safety accidents caused by battery failure.

[0004] In order to solve the above technical problems, the present invention provides a battery simulator circuit damage prediction method based on multi-time series feature convolution, comprising:

[0005] The exponentially weighted moving average filtering method is used to denoise the high-frequency current data to obtain the denoised fault data set;

[0006] Wavelet transform is used to extract fault spectrum features from denoised fault data sets;

[0007] A variational autoencoder optimized by 3-fold cross-validation is used to extract fault coding features from the denoised fault dataset;

[0008] A sparse deep belief network is used to extract fault depth features from the denoised fault dataset;

[0009] The Beluga optimized convolutional neural network is used to predict damage in battery simulator circuits.

[0010] Optionally, the exponentially weighted moving average filtering method is used to preprocess the high-frequency current data, including:

[0011] The high-frequency current data of the drain of the MOS tube of the battery simulator is collected, the smoothing coefficient of the high-frequency current data is determined, an initial value is selected as the observation value, the exponentially weighted average of the current data at each time point is calculated, and the denoised fault data set is obtained after data smoothing.

[0012] Optionally, extracting fault spectrum features from the denoised fault data set using wavelet transform includes:

[0013] The denoised fault data is decomposed into wavelet coefficients of different scales through wavelet transform. By selecting different decomposition levels and wavelet basis functions, the characteristic information of the denoised fault data is detected at different detail levels and the fault spectrum characteristics are extracted.

[0014] Optionally, the extracting fault coding features from the denoised fault dataset using a variational autoencoder optimized by 3-fold cross validation includes:

[0015] The denoised fault dataset is input into the variational autoencoder, and 3-fold cross validation is used to optimize the hyperparameters of the variational autoencoder. The variational autoencoder maps the fault data to a low-dimensional latent space by introducing variational inference to extract fault coding features.

[0016] Optionally, the extracting fault depth features from the denoised fault dataset using a sparse deep belief network includes:

[0017] Multiple layers of restricted Boltzmann machines are stacked, and sparsity constraints are introduced into the hidden layers of the restricted Boltzmann machines to construct a sparse deep belief network. The sparse deep belief network is trained using an unsupervised learning method, and the weights of the sparse deep belief network are adjusted using the back-propagation algorithm. The denoised fault dataset is input into the trained sparse deep belief network to extract the fault depth features.

[0018] Optionally, the use of the Beluga optimized convolutional neural network to predict damage to the battery simulator circuit includes:

[0019] The three fault features are concatenated end to end to form a circuit damage prediction feature, which is then input into the Beluga optimized convolutional neural network. The Beluga algorithm optimizes the parameters of the convolutional neural network by simulating the group behavior of beluga whales. After training, damage prediction is performed on the battery simulator circuit.

[0020] The present invention provides a method for predicting damage of a battery simulator circuit based on multi-time series feature convolution, the method comprising: using an exponentially weighted moving average filter method to denoise high-frequency current data to obtain a denoised fault data set; using wavelet transform to extract fault spectrum features from the denoised fault data set; using a variational autoencoder optimized by 3-fold cross-validation to extract fault coding features from the denoised fault data set; using a sparse deep belief network to extract fault depth features from the denoised fault data set; and concatenating the three fault features head to tail and inputting them into a Beluga optimized convolutional neural network to predict damage to the battery simulator circuit. The present invention can predict the circuit damage of the battery simulator and avoid the losses caused by sudden failures, and has significant advantages in predicting damage to the battery simulator circuit. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A schematic diagram of a flow chart of a method for predicting damage of a battery simulator circuit provided by an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of a wavelet decomposition process of a battery simulator circuit damage prediction method provided by an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of a sparse deep belief network framework for a battery simulator circuit damage prediction method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] like Figure 1 As shown, Figure 1 A flowchart of a method for predicting damage to a battery simulator circuit provided by an embodiment of the present invention is provided. The method for predicting damage to a battery simulator circuit specifically includes five contents.

[0027] S11: Use the exponentially weighted moving average filtering method to denoise the high-frequency current data to obtain a denoised fault data set.

[0028] It should be noted that the present invention adopts the exponentially weighted moving average filtering method to denoise the high-frequency current data because the calculation method of the exponentially weighted moving average filtering method is relatively simple, easy to understand and program, does not require a large amount of computing resources, can effectively smooth the high-frequency current data, reduce the impact of noise, and make the useful current signal components more prominent. The method specifically includes two steps.

[0029] S111: Connect the battery simulator to the data acquisition system, synchronously collect the high-frequency current data of the MOS tube drain, determine a smoothing coefficient, select the average value of the first few points as the first current data point after filtering, and start from the second current data point, iteratively calculate according to formula (1):

[0030]

[0031] In the formula, S t is the filtered value at the current moment, a is the smoothing coefficient, X t is the current data value at the current moment, S t-1 is the filtered value at the previous moment.

[0032] S112: Repeat the above calculation process to obtain all current data values ​​after exponentially weighted moving average filtering, and organize them to form a denoised fault data set.

[0033] Based on the above discussion, in an optional embodiment of the present invention, the high-frequency current data is denoised by using the exponentially weighted moving average filtering method to obtain a denoised fault data set, which specifically includes:

[0034] In an optional embodiment of S11, the smoothing coefficient a of the exponentially weighted moving average filtering method is set to 0.4.

[0035] S12: Wavelet transform is used to extract fault spectrum features from the denoised fault dataset.

[0036] It should be noted that the present invention introduces wavelet transform because it can focus on different details of the current signal, observe the overall outline, and deeply analyze local subtle changes. In addition, the algorithm of wavelet transform is relatively simple and the calculation speed is fast, which has obvious advantages in processing large-scale data.

[0037] like Figure 2 As shown in the figure, wavelet transform is used to extract fault spectrum features from the denoised fault data set, which specifically includes three steps.

[0038] S121: Decompose the denoised fault data by wavelet, and perform multi-scale analysis on the current signal by changing the scale parameter and translation parameter to obtain the wavelet coefficients A of different frequencies, as shown in formula (2):

[0039]

[0040] Where b is the scale parameter, c is the translation parameter, and A(t) is the wavelet basis function.

[0041] S122: Draw a spectrum graph of the wavelet coefficients, calculate the amplitude value of each frequency point in the spectrum graph, filter all frequency points in the spectrum graph, find points with larger amplitude values, set an amplitude threshold, and mark points with amplitudes higher than the threshold as potential peak points.

[0042] S123: Further screen the potential peak points, remove the peak points that are not so prominent compared with the surrounding points, and obtain the fault spectrum characteristics through the frequency and amplitude information of the peak points.

[0043] Based on the above discussion, in an optional embodiment of the present invention, the above-mentioned extraction of fault spectrum features from the denoised fault data set using wavelet transform specifically includes:

[0044] In an optional embodiment of S12, the denoised fault data is subjected to a 4-layer wavelet decomposition to obtain wavelet coefficients A1, A2, A3, A4, and A5 from the high frequency band to the low frequency band.

[0045] S13: A variational autoencoder optimized with 3-fold cross validation is used to extract fault encoding features from the denoised fault dataset.

[0046] It should be noted that the present invention introduces a variational autoencoder optimized by 3-fold cross-validation because the use of 3-fold cross-validation for training can reduce the risk of overfitting. The variational autoencoder maps high-dimensional data to a low-dimensional latent space by learning the latent space representation of denoised fault data, and the essential characteristics of the denoised fault data can be captured in this latent space.

[0047] A variational autoencoder optimized by 3-fold cross validation is used to extract fault coding features from the denoised fault dataset, which specifically includes three steps.

[0048] S131: The denoised fault dataset is randomly divided into three subsets of equal size and input into the variational autoencoder for 3-fold cross validation.

[0049] S132: In each round of cross-validation, two subsets are used as training sets to train the parameters of the variational autoencoder so that the reconstruction error of the variational autoencoder is minimized, and the remaining subset is used as the validation set. This process is repeated three times, and a different subset is selected as the validation set in each round to verify and optimize the variational autoencoder.

[0050] S133: The variational autoencoder maps the fault data to a low-dimensional latent space by introducing variational inference, and uses the encoder part of the variational autoencoder to encode the fault data to obtain fault encoding features.

[0051] Based on the above discussion, in an optional embodiment of the present invention, the above variational autoencoder optimized by 3-fold cross validation extracts fault coding features from the denoised fault data set, specifically including:

[0052] In an optional embodiment of S13, the variational autoencoder uses Adam as the optimizer, the number of iterations is 300, the batch size is 64, and the learning rate is set to 0.001.

[0053] S14: Extracting fault depth features from denoised fault datasets using sparse deep belief networks.

[0054] It should be noted that the present invention introduces a sparse deep belief network because the sparsity constraint of the sparse deep belief network makes the features learned by the sparse deep belief network more sparse and representative, and the sparse deep belief network is based on an energy-based probabilistic model and a layer-by-layer training method, which can better learn the feature representation of fault data.

[0055] like Figure 3 As shown in the figure, a sparse deep belief network is used to extract fault depth features from the denoised fault dataset, which specifically includes three steps.

[0056] S141: Multiple layers of restricted Boltzmann machines are stacked, and sparsity constraints are introduced into the hidden layers of the restricted Boltzmann machines to construct a sparse deep belief network. The energy function of a single restricted Boltzmann machine is:

[0057]

[0058] Where v is the visible layer vector, h is the hidden layer vector, and n v and n h are the number of nodes in the visible layer and the hidden layer, d and e are the biases of the visible layer and the hidden layer, respectively. ij is the weight matrix connecting the visible layer and the hidden layer.

[0059] S142: Input the denoised fault data set into the sparse deep belief network, train the first layer of restricted Boltzmann machine RBM, use the output of the first layer of restricted Boltzmann machine RBM as the input of the second layer of restricted Boltzmann machine RBM to train the second layer of restricted Boltzmann machine RBM, and use the back propagation algorithm BP to adjust the weights of the sparse deep belief network, and repeat the above process until the optimal parameters of the sparse deep belief network are obtained.

[0060] S143: Extracting fault depth features from denoised fault datasets using trained sparse deep belief networks.

[0061] Based on the above discussion, in an optional embodiment of the present invention, the method of extracting fault depth features from a denoised fault data set using a sparse deep belief network specifically includes:

[0062] In an optional embodiment of S14, denoised fault data is first received from the visible layer of the bottom-level restricted Boltzmann machine, and its hidden layer becomes the next layer in the subsequent hierarchical structure. Each new restricted Boltzmann machine is trained using the hidden layer of the previous restricted Boltzmann machine as its visible layer.

[0063] S15: The three fault features are concatenated end to end and input into the Beluga optimized convolutional neural network to predict damage to the battery simulator circuit.

[0064] It should be noted that the present invention introduces the Beluga optimized convolutional neural network because compared with the traditional optimization algorithm, the Beluga optimization algorithm shows better performance when dealing with complex problems, can converge to the global optimal solution more quickly, and the convolution operation of the convolutional neural network can be executed in parallel, and can process multiple convolution operations at the same time, which greatly improves the calculation speed, enabling the convolutional neural network to process large amounts of data in a shorter time.

[0065] The Beluga optimized convolutional neural network is used to predict damage to the battery simulator circuit, which includes three steps.

[0066] S151: The three fault features are concatenated end to end to form a circuit damage prediction feature, and the circuit damage prediction feature is input into a convolutional neural network.

[0067] S152: The convolutional neural network is optimized using the Beluga algorithm. First, the Beluga population is initialized, the fitness value of each individual Beluga is calculated, and the position of the individual Beluga is updated according to the search behavior of the Beluga. The process of calculating the fitness value and updating the position of the individual Beluga is repeated until the stopping condition is met and the maximum number of iterations is reached.

[0068] S153: Use the trained Beluga optimized convolutional neural network to predict damage to the battery simulator circuit.

[0069] Based on the above discussion, in an optional embodiment of the present invention, the damage prediction of the battery simulator circuit using the Beluga optimized convolutional neural network specifically includes:

[0070] In an optional embodiment of S15, the number of neurons in the fully connected layer of the convolutional neural network is 100, the batch size is set to 64, and the learning rate is set to 0.003.

[0071] It should be noted that, in this document, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0072] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A battery simulator circuit damage prediction method based on multi-time series feature convolution, characterized in that: include: The exponentially weighted moving average filtering method is used to denoise the high-frequency current data to obtain the denoised fault data set; Wavelet transform is used to extract fault spectrum features from denoised fault data sets; A variational autoencoder optimized by 3-fold cross-validation is used to extract fault coding features from the denoised fault dataset; A sparse deep belief network is used to extract fault depth features from the denoised fault dataset; The Beluga optimized convolutional neural network is used to predict damage in battery simulator circuits.

2. The battery simulator circuit damage prediction method as claimed in claim 1, characterized in that: The exponentially weighted moving average filtering method is used to denoise the high-frequency current data to obtain a denoised fault data set, including: The high-frequency current data of the drain of the MOS tube of the battery simulator is collected, the smoothing coefficient of the high-frequency current data is determined, an initial value is selected as the observation value, the exponentially weighted average of the current data at each time point is calculated, and the denoised fault data set is obtained after data smoothing.

3. The battery simulator circuit damage prediction method as claimed in claim 1, characterized in that: Wavelet transform is used to extract fault spectrum features from the denoised fault data set, including: The denoised fault data is decomposed into wavelet coefficients of different scales through wavelet transform. By selecting different decomposition levels and wavelet basis functions, the characteristic information of the denoised fault data is detected at different detail levels and the fault spectrum characteristics are extracted.

4. The battery simulator circuit damage prediction method as claimed in claim 1, characterized in that: A variational autoencoder optimized with 3-fold cross validation is used to extract fault coding features from the denoised fault dataset, including: The denoised fault dataset is input into the variational autoencoder, and 3-fold cross validation is used to optimize the hyperparameters of the variational autoencoder. The variational autoencoder maps the fault data to a low-dimensional latent space by introducing variational inference to extract fault coding features.

5. The battery simulator circuit damage prediction method as claimed in claim 1, characterized in that: A sparse deep belief network is used to extract fault depth features from the denoised fault dataset, including: Multiple layers of restricted Boltzmann machines are stacked, and sparsity constraints are introduced into the hidden layers of the restricted Boltzmann machines to construct a sparse deep belief network. The sparse deep belief network is trained using an unsupervised learning method, and the weights of the sparse deep belief network are adjusted using the back-propagation algorithm. The denoised fault dataset is input into the trained sparse deep belief network to extract the fault depth features.

6. The battery simulator circuit damage prediction method as claimed in claim 1, characterized in that: The Beluga optimized convolutional neural network is used to predict damage to the battery simulator circuit, including: The three fault features are concatenated end to end to form a circuit damage prediction feature, which is then input into the Beluga optimized convolutional neural network. The Beluga algorithm optimizes the parameters of the convolutional neural network by simulating the group behavior of beluga whales. After training, damage prediction is performed on the battery simulator circuit.

Images