Battery simulator circuit health assessment method based on star computation-A3C

Abstract

The invention discloses a battery simulator circuit health assessment method based on star computation-A3C, and the method comprises the steps: carrying out the normalization processing of a high-frequency current signal, and carrying out the signal reconstruction through a period sparse method, and obtaining period sparse data; a recurrence plot method is adopted to convert the feature map into an evaluation feature map; an improved star operation convolutional network is introduced, a multi-thread Actor-Critic cooperation mechanism of an A3C algorithm is combined, an Actor network adaptively adjusts a feature fusion weight, a Critic network evaluates a feature fusion value, and high-order feature association is dynamically captured; and a support vector machine optimized based on an energy valley algorithm is adopted to realize battery simulator circuit health state evaluation. According to the method, the health state of the battery simulator circuit is evaluated through feature collaborative optimization and parallel computing acceleration of the star computation-A3C, and the method has remarkable technical advantages and application value.

Description

Technical Field

[0001] The present invention belongs to the field of reinforcement learning fault diagnosis, and particularly relates to a method for evaluating the circuit health of a battery simulator based on star operation - A3C. Background Art

[0002] With the transformation of the global energy structure towards renewable energy, the importance of energy storage batteries in fields such as electric vehicles, smart grids, photovoltaic, and wind energy storage systems has become increasingly prominent. As the core energy storage unit, the performance of the battery directly affects the system efficiency and safety. However, real battery tests have problems such as high cost, uncontrollable aging, and high environmental risks. A safe and efficient circuit health assessment is achieved through a battery simulator. The battery simulator is an indispensable tool in the battery research and development process, and the circuit health status of the battery simulator is comprehensively affected by various factors. Electrically, insulation damage caused by voltage stress; environmentally, circuit damage caused by high temperature, high humidity, and the accumulation of dust and pollutants. In terms of manufacturing processes, unreasonable designs can cause problems such as stress concentration, heat dissipation, and electromagnetic compatibility, all of which can affect the circuit health status of the battery simulator. By evaluating the circuit health status of the battery simulator, reasonable maintenance strategies can be formulated to improve the long-term operation reliability of the battery and reduce losses caused by equipment failures. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for evaluating the circuit health of a battery simulator based on star operation - A3C, which can take timely preventive measures through the evaluation of the circuit health status of the battery simulator, thereby increasing the service life of the battery and reducing the occurrence of safety accidents.

[0004] To solve the above technical problems, the present invention provides a method for evaluating the circuit health of a battery simulator based on star operation - A3C, including: Normalize the high-frequency current signal and reconstruct the signal using the periodic sparse method to obtain periodic sparse data; Convert the periodic sparse data into an evaluation feature map using the recurrence graph method; Dynamically capture high-order feature associations through the star operation - A3C module; Use a support vector machine optimized by the energy valley algorithm to achieve the evaluation of the circuit health status of the battery simulator.

[0005] Optionally, the normalization of the high-frequency current signal and the reconstruction of the signal using the periodic sparse method to obtain periodic sparse data includes: Normalize the high-frequency current signal, and use the periodic sparse method to process the inherent periodicity contained in the current data, reducing the information loss during the signal reconstruction process. First, downsample the normalized high-frequency current signal and segment it into cross-period subsequences. Then, predict the trend of the cross-period subsequences through sparse sliding prediction to obtain a prediction matrix. Finally, upsample, transpose, and reconstruct the prediction matrix to obtain periodic sparse data.

[0006] Optionally, converting the periodic sparse data into an evaluation feature map by using the recurrence plot method includes: Use the recurrence plot method to capture the non-linear relationships in the data for non-stationary sparse data that is difficult to handle by traditional methods. First, perform phase space reconstruction on the periodic sparse data, calculate the distances between each pair of points in the reconstructed phase space using the Euclidean distance to form a distance matrix, then binarize the distance matrix by setting a threshold to generate a recurrence plot matrix, and finally map the recurrence plot matrix to a grayscale image to obtain the evaluation feature map.

[0007] Optionally, dynamically capturing high-order feature associations through the star operation - A3C module includes: Input the evaluation feature map into the star operation - A3C module. Mine the high-order relationships between features through the star operation and map the input to a high-dimensional non-linear feature space. Perform two linear transformations on the evaluation feature map to obtain two feature vectors, multiply the two transformed feature vectors element by element to obtain the star operation interaction feature. Use multi-threaded parallel processing, where each thread independently calculates the gradient and asynchronously updates the Actor-Critic network parameters. The Actor network adaptively adjusts the feature fusion weights in different subspaces, and the Critic network evaluates the feature fusion value in real time through the advantage function. Aggregate the feature interaction results of all threads to obtain the global evaluation feature.

[0008] Optionally, implementing the battery simulator circuit health state evaluation by using a support vector machine optimized based on the energy valley algorithm includes: Optimize the parameters of the support vector machine based on the energy valley algorithm. The energy valley algorithm simulates the formation mechanism of energy valleys in physical systems, efficiently searches for the global optimal solution in a complex high-dimensional parameter space, adaptively adjusts the search step size and direction, and reaches stable convergence within fewer iterations. Input the global evaluation feature into the optimized support vector machine to implement the battery simulator circuit health state evaluation.

[0009] The present invention provides a method for evaluating the health of a battery simulator circuit based on the star operation - A3C. The method includes normalizing high - frequency current signals, reconstructing signals using the periodic sparse method to obtain periodic sparse data; converting it into an evaluation feature map using the recurrence plot method; introducing an improved star operation convolutional network, combining the multi - thread Actor - Critic cooperation mechanism of the A3C algorithm, where the Actor network adaptively adjusts the feature fusion weights, and the Critic network evaluates the feature fusion value to dynamically capture high - order feature associations; and implementing the health state evaluation of the battery simulator circuit using a support vector machine optimized by the energy valley algorithm. Through the feature collaborative optimization and parallel computing acceleration of the star operation - A3C, the present invention realizes the health state evaluation of the battery simulator circuit, having significant technical advantages and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0011] Figure 1 It is a schematic flowchart of the method for evaluating the health of a battery simulator circuit provided by an embodiment of the present invention; Figure 2 It is a schematic framework diagram of the star operation - A3C module of the method for evaluating the health of a battery simulator circuit provided by an embodiment of the present invention; Figure 3 It is a schematic flowchart of the energy valley algorithm of the method for evaluating the health of a battery simulator circuit provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] To enable those skilled in the art to better understand the solution of the present invention, the following will further elaborate on the present invention in conjunction with the drawings and specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0013] As Figure 1 shown, Figure 1 It is a schematic flowchart of the method for evaluating the health of a battery simulator circuit provided by an embodiment of the present invention, and the method specifically includes 4 parts.

[0014] S1: Normalize the high - frequency current signal, and reconstruct the signal using the periodic sparse method to obtain periodic sparse data.

[0015] It should be noted that the present invention uses a periodic sparse method to process the inherent periodicity contained in current data, reducing information loss in the signal reconstruction process; this method can provide more accurate and reliable periodic sparse data when dealing with complex and non-stationary current signals, laying a good foundation for subsequent feature extraction. This method specifically includes 3 steps.

[0016] S11: For the collected high-frequency current signal, calculate its mean value, subtract the mean value from each data point to obtain the normalized high-frequency current signal.

[0017] S12: Divide the normalized high-frequency current signal into a sub-sequences according to the period a, perform sliding aggregation on each sub-sequence, fuse the information of adjacent time points, smooth the outliers through weighted average, and enhance the continuity of the local trend within the sub-sequence.

[0018] S13: Perform sparse prediction on the trend of each sub-sequence through a linear layer with shared parameters to obtain a prediction matrix, then transpose and reconstruct the prediction matrix through up-sampling to restore it to a complete time series signal, and finally process to obtain periodic sparse data.

[0019] Based on the above discussion, in an optional embodiment of the present invention, for the above-mentioned signal reconstruction using the periodic sparse method to obtain periodic sparse data, it specifically includes: In an optional embodiment of S1, the periodic sparse method consists of two main parts: a one-dimensional convolutional layer for sliding aggregation and a linear layer with shared parameters for sparse sliding prediction. The number of parameters in the one-dimensional convolutional layer is determined by the kernel size, and it is trained for 35 epochs using the Adam optimizer.

[0020] S2: Use the recurrence plot method to convert the periodic sparse data into an evaluation feature map.

[0021] It should be noted that the present invention uses the recurrence plot method to convert it into an evaluation feature map because this method can intuitively display the dynamic characteristics and complex relationships of the signal, can better understand the working state and health trend of the battery simulator circuit, and also provides a rich data basis for subsequent feature extraction.

[0022] Using the recurrence plot method to convert the periodic sparse data into an evaluation feature map specifically includes 3 steps.

[0023] S21: Perform phase space reconstruction on the periodic sparse data, select the embedding dimension and time delay, construct a series of reconstruction vectors, and calculate the Euclidean distance between each pair of points Y i and Y j in the reconstructed phase space.

[0024] S22: Perform threshold binarization processing based on the distance between points to generate a recurrence graph matrix, and its formula is as follows:

[0025] where θ is the Heaviside step function. If the distance between Y i and Y j is less than the set threshold ε, the corresponding position R ij on the recurrence graph is marked as 1; otherwise, it is marked as 0.

[0026] S23: Map the recurrence graph matrix to a grayscale image to obtain an evaluation feature map.

[0027] Based on the above discussion, in an alternative embodiment of the present invention, for the above method of converting periodic sparse data into an evaluation feature map using the recurrence graph method, it specifically includes: In an alternative embodiment of S2, when using the recurrence graph method for phase space reconstruction, the condition m≥2D + 1 needs to be satisfied, where m is the embedding dimension and D is the true dimension of the data; calculate the autocorrelation function of the periodic sparse data, and select the first zero crossing as the time delay.

[0028] S3: Dynamically capture high-order feature associations through the star operation - A3C module.

[0029] It should be noted that the star operation - A3C module is introduced in the present invention because through the high-dimensional mapping and feature interaction of the star operation, complex relationships between features can be mined; through the multi-threaded parallel processing of the A3C algorithm, each thread independently calculates the gradient and updates the network parameters asynchronously, avoiding the waiting time in traditional synchronous training, and greatly improving the convergence speed. The star operation - A3C module combines the feature interaction ability of the star operation and the reinforcement learning optimization ability of the A3C algorithm to handle complex feature relationships and optimize the feature fusion strategy.

[0030] As Figure 2 shown, dynamically capturing high-order feature associations through the star operation - A3C module specifically includes 4 steps.

[0031] S31: Input the evaluation feature map into the star operation - A3C module, and perform two linear transformations on the evaluation feature map to obtain two feature vectors, and its formula is as follows:

[0032] where G is the evaluation feature map, W e and W f are learnable weight parameters, b e and b fis the bias term, and E and F are the transformed eigenvectors.

[0033] S32: Multiply the two transformed eigenvectors element by element to capture the non-linear correlation between channels, and output the star operation interaction features. The formula is as follows:

[0034] where Z is the star operation interaction feature, and ⊙ represents element-by-element multiplication.

[0035] S33: Adopt multi-thread parallel processing, and the Actor network adaptively adjusts the feature fusion weights of different subspaces. The formula is as follows:

[0036] where α k is the feature fusion weight coefficient of the Actor network, exp is the exponential function, T is the function parameter of the Actor network, and k is the number of subspaces.

[0037] The Critic network evaluates the feature fusion value in real time through the advantage function. The objective function of the advantage function A(G,α) is as follows:

[0038] where Q(G,α) represents the cumulative reward for completing actions in the Critic network, V critic (Z) is the expected value reward of the Critic network, γ is the discount factor, r t is the immediate reward obtained at time t, Z t+n is the star operation interaction feature at time t+n, which is used as the input of the Critic network; n is the number of time steps for temporal difference estimation.

[0039] S34: The agents of each thread independently calculate the gradients, asynchronously update the Actor-Critic network parameters, and aggregate the feature interaction results of all threads to obtain the global evaluation features.

[0040] Based on the above discussion, in an optional embodiment of the present invention, for the dynamic capture of high-order feature correlations through the star operation - A3C module, it specifically includes: In an optional embodiment of S3, the star operation - A3C module sets the number of training steps to 512, the number of iterations to 500, and the learning rate to 0.005 in the experiment.

[0041] S4: Use a support vector machine optimized by the energy valley algorithm to realize the evaluation of the health state of the battery simulator circuit.

[0042] It should be noted that the health state of the battery simulator circuit is divided into four levels. Healthy means the device has no faults and the parameters are normal; sub-healthy means the device parameters deviate and the device needs to be monitored in real time; minor fault means the device function degrades and immediate maintenance is required; severe fault means the device function fails and shutdown for repair and emergency treatment are needed. The energy valley algorithm adopted by the present invention can efficiently search for the global optimal solution in a complex high-dimensional parameter space, avoiding the problem that traditional optimization algorithms are prone to fall into local optima. Compared with traditional optimization algorithms, the energy valley algorithm has a faster convergence speed when optimizing the support vector machine, which is crucial for the evaluation of the health state of the battery simulator circuit with high real-time requirements.

[0043] The support vector machine optimized by the energy valley algorithm is adopted to realize the evaluation of the health state of the battery simulator circuit, which specifically includes 3 steps.

[0044] S41: Optimize the hyperparameters of the support vector machine by the energy valley algorithm, including the regularization parameter c and the kernel function parameter d.

[0045] S42: As Figure 3 shown, in the implementation process of the energy valley algorithm, first, initialization is performed, and a group of candidate solutions are randomly generated in the parameter space. Each solution represents a set of hyperparameter combinations (c, d); an energy field in the physical system is constructed, and the "energy value" of each solution is evaluated through a fitness function. The low-energy region corresponds to a better solution; a global search strategy is carried out to simulate the movement of particles in the energy field, and the candidate solutions are updated by combining gradient descent and random perturbation exploration to avoid falling into local optima; stop when the maximum number of iterations is reached, and output the optimal hyperparameter combination of the support vector machine.

[0046] S43: Input the global evaluation features into the optimized support vector machine to realize the evaluation of the health state of the battery simulator circuit.

[0047] Based on the above discussion, in an optional embodiment of the present invention, for the support vector machine optimized by the energy valley algorithm to realize the evaluation of the health state of the battery simulator circuit, it specifically includes: In an optional embodiment of S4, the support vector machine is adopted to realize the evaluation of the health state of the battery simulator circuit, and four-level health state dynamic thresholds are set. Healthy means all parameters are within the normal range, and the probability range is greater than 95%; sub-healthy means 1-2 parameters deviate slightly, and the probability range is 85%-95%; minor fault means key parameters are continuously abnormal, and the probability range is 60%-85%; severe fault means multiple parameters exceed the limit severely, and the probability range is less than 60%.

[0048] It should be noted that in this text, relational terms such as first and second are only 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.

[0049] In this article, specific examples are used to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A battery simulator circuit health assessment method based on Star Computing-A3C, characterized in that: include: The high-frequency current signal is normalized, and the signal is reconstructed using the periodic sparse method to obtain periodic sparse data; A recursive graph method is used to convert periodic sparse data into evaluation feature graphs; The evaluation feature map is input into the Star Operation-A3C module. The high-order relationship between features is mined through star operation to map the input to a high-dimensional nonlinear feature space. Two linear transformations are performed on the evaluation feature map to obtain two feature vectors. The two transformed feature vectors are multiplied element by element to obtain the star operation interaction feature. Multi-threaded parallel processing is adopted. Each thread independently calculates the gradient and asynchronously updates the Actor-Critic network parameters. The Actor network adaptively adjusts the feature fusion weights of different subspaces. The Critic network evaluates the feature fusion value in real time through the advantage function. The feature interaction results of all threads are aggregated to obtain the global evaluation feature. The support vector machine optimized based on the energy valley algorithm is used to realize the circuit health status assessment of the battery simulator.

2. The battery simulator circuit health assessment method as claimed in claim 1, characterized in that: The high-frequency current signal is normalized and the periodic sparse method is used to reconstruct the signal to obtain periodic sparse data, including: The high-frequency current signal is normalized. First, the normalized high-frequency current signal is downsampled and divided into cross-period subsequences. Then, the trend of the cross-period subsequences is predicted through sparse sliding prediction to obtain a prediction matrix. Finally, the prediction matrix is ​​transposed and reconstructed through upsampling to obtain periodic sparse data.

3. The battery simulator circuit health assessment method as claimed in claim 1, characterized in that: The recursive graph method is used to convert periodic sparse data into evaluation feature graphs, including: Firstly, the periodic sparse data is reconstructed in phase space. The distance between each pair of points in the reconstructed phase space is calculated using Euclidean distance to form a distance matrix. Then, the distance matrix is ​​binarized by setting a threshold to generate a recursive graph matrix. Finally, the recursive graph matrix is ​​mapped to a grayscale image to obtain an evaluation feature map.

4. The battery simulator circuit health assessment method as claimed in claim 1, characterized in that: The support vector machine optimized based on the energy valley algorithm is used to evaluate the circuit health status of the battery simulator, including: The parameters of the support vector machine are optimized based on the energy valley algorithm, the global optimal solution is efficiently searched in the complex high-dimensional parameter space, the search step size and direction are adaptively adjusted, and the global evaluation features are input into the optimized support vector machine to realize the circuit health status evaluation of the battery simulator.

Images