Artificial intelligence driven battery energy storage health state and residual life prediction method

Abstract

The invention provides a battery energy storage health state and residual life prediction method driven by artificial intelligence. The method is applied to the technical field of artificial intelligence, and comprises the following steps: obtaining time sequence data of various characteristics and state information of a historical time point of a battery, and preprocessing the time sequence data; inputting the preprocessed time sequence data into a feature extraction model based on a multi-scale patch hybrid network for feature extraction; in the self-supervised comparative representation learning stage, a nearest neighbor and a farthest neighbor which are meaningful semantically are found for each sample by utilizing learned feature representation, and in the self-supervised classification stage, classification is carried out based on information of the nearest neighbor and the farthest neighbor of all feature samples, and a classification model is trained through a self-defined loss function; and performing classification prediction on the new time sequence data of the battery by using the trained model, and outputting the health state of the battery. According to the invention, the accuracy, stability and generalization ability of battery energy storage health state monitoring and residual life prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-driven battery energy storage health status and remaining life prediction method. Background Art

[0002] Traditional methods face many challenges in monitoring the health status of battery energy storage and predicting the remaining life. In real-world scenarios, labeled data is extremely scarce, which greatly limits the application of supervised learning methods. Most existing methods use an unsupervised approach to learn the normal behavior of unlabeled time series. However, the normal boundary definitions of these methods are often too narrow, resulting in slight deviations being mistakenly judged as abnormalities, resulting in a high false alarm rate. In addition, the generalization ability of the model is also greatly limited, making it difficult to adapt to complex and changing practical application scenarios.

[0003] When traditional contrastive learning methods are applied to battery energy storage health status monitoring and remaining life prediction, their assumptions are seriously insufficient. For example, it is usually believed that the enhanced time series window is a positive sample, and the window with a long distance in time is a negative sample. However, this assumption is often not true in practical applications, because time series enhancement may produce negative samples, and windows with a long distance may also be positive samples. In addition, directly applying image analysis methods to battery energy storage health status monitoring and remaining life prediction is not ideal, because battery data has unique time series characteristics and complex periodicity, which is essentially different from image data.

[0004] These deficiencies in existing technologies have seriously affected the accuracy, stability and generalization of battery energy storage health status monitoring and remaining life prediction. In order to overcome these challenges, a new technical method is urgently needed to more accurately capture the timing characteristics and periodicity of battery data and improve the generalization ability and prediction accuracy of the model. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention provides an artificial intelligence-driven battery energy storage health status and remaining life prediction method, which aims to improve the accuracy, stability and generalization ability of battery energy storage health status monitoring and remaining life prediction, improve the application defects of traditional comparative learning methods, and realize accurate battery energy storage health status monitoring and remaining life prediction, so as to solve the shortcomings of traditional methods in anomaly detection, feature extraction and model optimization.

[0006] In a first aspect, the present invention provides an artificial intelligence-driven battery energy storage health status and remaining life prediction method, comprising: Obtain time series data of various characteristics and status information of the battery at historical time points and perform preprocessing; the various characteristics and status information include voltage data, current data, temperature data, capacity data, battery internal resistance, service life and historical cycle count; Inputting the preprocessed time series data into a feature extraction model based on a multi-scale patch mixing network for feature extraction, wherein the multi-scale patch mixing network is composed of a time mixer and a space mixer; In the self-supervised contrastive representation learning stage, the feature extraction model is trained using a triplet loss function to optimize the time series features extracted by the network, and the learned feature representation is used to find semantically meaningful nearest neighbors and farthest neighbors for each sample; In the self-supervised classification stage, classification is performed based on the nearest neighbor and farthest neighbor information of all feature samples, and the classification model is trained through a custom loss function to achieve the best classification effect; Use the trained model to classify and predict the new time series data of the battery and output the battery health status.

[0007] Furthermore, preprocessing the time series data of various characteristics and status information of the battery at historical time points includes: Use instance normalization to normalize the time series window; In the standardized time series data, based on point anomalies and subsequence anomalies, dimensions and start times are randomly selected and anomalies are injected to construct abnormal samples.

[0008] Furthermore, the temporal mixer utilizes convolution kernels of different sizes to capture short-term and long-term dependencies and extract variable time series information; the spatial mixer performs feature aggregation based on the maximum variance framework to extract multi-scale information on the variable dimension.

[0009] Further, the time mixer uses convolution kernels of different sizes to capture short-term and long-term dependencies and extract variable time series information, including: The time mixer extracts variable time series information by using linear layers and multi-layer perceptrons along the time dimension, and adds the input and output through short connections or residual connections. The specific formula is: , in, is the input time series feature; It is a linear layer, which is used to perform linear layer normalization on the input time series features; It is a multi-layer perceptron used for feature transformation and extraction, and performs nonlinear mapping on the normalized data to learn deeper representations; is the final output feature, that is, the extracted variable time series information.

[0010] Furthermore, the spatial mixer performs feature aggregation based on the maximum variance framework, extracting multi-scale information on the variable dimension including: Data cycles were analyzed using a maximum variance framework to identify periodic patterns in the data; Based on the results of data period analysis, multi-scale patch segmentation is performed, including: determining the patch sizes of different scales according to the data period; dividing the original pre-processed time series data into patches of multiple scales, each patch containing data within a specific time range; For each scale of patch, a parameter-sharing multilayer perceptron is used to extract features along the variable dimension; The multi-scale feature aggregation weight is determined based on the amplitude calculated based on the maximum variance framework, and the multi-scale features are fused according to the calculated multi-scale feature aggregation weight, and added to the output of the time mixer through a short connection. The specific formula is: , in, is the i-th scale patch determined based on the data periodicity analysis; n represents the number of scales; represents the features extracted by applying a multi-layer perceptron to the i-th scale patch; is the aggregation weight of the i-th scale feature determined by the magnitude calculated by the maximum variance framework; is the output of the time mixer; S is the final output of the space mixer.

[0011] Further, wherein the analyzing the data periodicity using the maximum variance framework to identify periodic patterns in the data includes: The maximum variance framework matches the modified circulant matrix, that is, finds a modified circulant matrix that maximizes the variance of the data CX obtained after the modified circulant matrix C transforms the original data X. By maximizing the variance, the most important feature in the time series data, namely the data periodic pattern, is extracted.

[0012] Furthermore, in the self-supervised contrastive representation learning stage, the feature extraction model is trained using a triplet loss function to optimize the time series features extracted by the network, including: The feature extraction model is trained using a triplet loss function to minimize the distance between the anchor sample and the positive sample, while maximizing the distance between the anchor sample and the negative sample, and the difference is at least a preset interval. ; If the distance between the negative sample and the anchor point does not exceed the distance between the positive sample and the anchor point plus the preset interval , a loss will occur, otherwise the loss is zero, thereby optimizing the time series features extracted by the feature extraction model to better distinguish normal and abnormal samples.

[0013] Furthermore, in the self-supervised classification stage, classification is performed based on the nearest neighbor and farthest neighbor information of all feature samples, and the classification model is trained through a custom loss function to achieve the best classification effect, including: The classification model is trained by a custom loss function so that the classification model can accurately classify the nearest neighbor and farthest neighbor information of all feature samples; the loss function maximizes the similarity between the window representation and its nearest neighbor and minimizes the similarity between the window representation and its farthest neighbor. The loss function formula is: , in, Indicates windows; represents the nearest neighbor; represents the farthest neighbor; represents a similarity measure; Control the weight of nearest neighbor similarity optimization; Control the weight of the farthest neighbor similarity optimization; Indicates Windows Its nearest neighbors The similarity between Indicates Windows Its furthest neighbor ; α is a hyperparameter used to control the weight of the regularization term; H(x) represents entropy.

[0014] Furthermore, the trained model is used to classify and predict the new time series data of the battery, and the battery health status is output, including: After model training is complete, the class assignment and majority class of the dataset are determined through the inference process, including: First, use the trained classification model to classify the battery data, identify normal and abnormal battery data, and then count the proportion of the majority class, that is, normal battery data; Calculate the probability that the new window belongs to the majority class, determine whether the new window is abnormal based on the probability, and output the abnormal score of the battery health status indicator.

[0015] Furthermore, the method further comprises: The trained model is used to extract features from the battery health status data of the battery, and the battery health status data after feature extraction is input into the remaining life prediction model for prediction, and the remaining life prediction value of the battery is output. If the remaining life prediction value is less than a certain threshold, an alarm is triggered or a suggestion to replace the battery is made.

[0016] Compared with the prior art, the present invention achieves the following beneficial effects: (1) Improving prediction accuracy: The present invention introduces a multi-scale patch hybrid network to fully utilize the multi-scale characteristics of time series data, thereby improving the model's prediction accuracy for battery health status and remaining life; the design of a custom loss function enables the model to better distinguish between normal and abnormal data, further improving the prediction accuracy.

[0017] (2) Enhanced model stability: The present invention improves the stability of the model by applying the modified circulant matrix and the maximum variance framework, which enables the model to model non-uniform periodic changes more flexibly. The similarity between different feature layers is measured by the centralized kernel alignment method, which improves the stability and generalization ability of data features.

[0018] (3) Improving generalization ability: The method of the present invention does not rely on a large amount of labeled data. It improves the generalization ability of the model through self-supervised learning and feature extraction. The application of anomaly injection technology enables the model to learn more comprehensive anomaly features, further improving the generalization ability of the model.

[0019] (4) It has important practical value: The present invention proposes a new multi-scale patch hybrid network structure, which provides new ideas and methods for feature extraction of time series data. By combining self-supervised learning and feature extraction, it can achieve accurate prediction of battery health status and remaining life, which has important practical value.

[0020] In summary, the present invention overcomes the shortcomings of traditional methods in battery energy storage health status monitoring and remaining life prediction, significantly improves the accuracy, stability and generalization ability of prediction, and has important practical value.

[0021] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein: Figure 1 It is a flow chart of an artificial intelligence driven battery energy storage health status and remaining life prediction method according to an embodiment of the present invention; Figure 2 It is a schematic diagram of the specific steps of an artificial intelligence-driven battery energy storage health status and remaining life prediction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0025] The embodiment of the present invention provides an artificial intelligence driven battery energy storage health status and remaining life prediction method, and the present invention includes the following innovations: 1. The present invention proposes a multi-scale patch hybrid network to make full use of the multi-scale characteristics of time series data. The network consists of a time mixer and a space mixer. The time mixer uses convolution kernels of different sizes to capture short-term and long-term dependencies, and combines residual connections to enhance the time series modeling capability. The spatial mixer performs feature aggregation based on the maximum variance framework, extracts multi-scale information on the variable dimension, and improves the model's ability to characterize the battery health state. 2. In view of the complex periodicity and non-stationary characteristics of battery energy storage data, the present invention constructs a modified circulant matrix and a maximum variance framework, which can more flexibly model non-uniform periodic changes and improve the time series pattern recognition capability. 3. The present invention designs a custom loss function suitable for self-supervised contrastive learning to optimize the effect of time series feature extraction. Through the triple loss function, the distance between the anchor sample and the positive sample is minimized, while the distance between the negative sample is increased, thereby improving the model's ability to distinguish between normal and abnormal data; based on the nearest neighbor and farthest neighbor loss functions, the similarity constraints between windows are emphasized to ensure that the time series representation has stronger discriminability and robustness. 4. In order to further optimize the characterization ability of time series, the present invention uses a centralized kernel alignment method to measure the similarity between different feature layers, improve the stability and generalization ability of data features, and enhance the credibility and interpretability of the model.

[0026] Figure 1 A flow chart of an artificial intelligence-driven method for predicting the health status and remaining life of battery energy storage is shown; Figure 2 The figure shows a schematic diagram of the specific steps of an artificial intelligence driven method for predicting the health status and remaining life of battery energy storage. Figure 1 and Figure 2 As shown, an artificial intelligence driven battery energy storage health state and remaining life prediction method 100 is a battery energy storage health state and remaining life prediction method based on self-supervised contrast representation learning, which is divided into two stages, namely, a self-supervised contrast representation learning stage and a self-supervised classification stage. The method 100 includes: 1. Self-supervised contrastive representation learning stage S110: Obtaining time series data of various characteristics and status information of the battery at historical time points and performing preprocessing; the various characteristics and status information include voltage data, current data, temperature data, capacity data, battery internal resistance, service life, and historical cycle count; In battery energy storage health status monitoring and remaining life prediction, time series data is a series of observation value sequences arranged in chronological order, which reflects various characteristics and status information of the battery at different time points, including voltage data, current data, temperature data, capacity data, battery internal resistance, service life and historical cycle times, etc. In this step S110, the time series data is preprocessed, specifically including: S111: Time Series Data Standardization Instance normalization is used to normalize the time series window to eliminate the drift of input data distribution. Since the distribution of time series data is easily affected by natural or human factors and shifted, and the training set is difficult to cover all distributions. This module connects the prediction timestamp with the historical sequence for synchronous normalization (using instance normalization), removes the denormalization term, and only retains instance normalization for input stability to help the model cope with changes in data distribution. The instance normalization formula is: , in, Indicates The sample Features, Indicates The mean of the features, Indicates The variance of the feature, To prevent small constants with zero denominators, is the normalized result.

[0027] By standardizing time series data, the dimensional differences between different time series can be eliminated, making subsequent analysis more accurate.

[0028] S112: Time series data anomaly injection Anomaly injection is usually performed on standardized time series data, using point anomalies (single-point context deviation) and subsequence anomalies (periodicity, trend anomalies, etc.) to construct anomaly samples. Randomly select dimensions and start times in multivariate time series data for anomaly injection: , : The original value of the dimension at time in the multivariate time series; : The value after the exception is injected; : Abnormal disturbance term (point anomaly, or subsequence anomaly such as periodic anomaly or trend anomaly can be used).

[0029] In order to enrich the training samples by generating time series data similar to real anomalies when data is insufficient or the annotation cost is high, time series data anomaly injection can improve the model's ability to identify anomalies. By injecting different types of anomalies, a variety of anomalies can be simulated, allowing the model to learn more comprehensive anomaly features. By training the model on data containing anomalies, the model's robustness to anomalies can be improved, reducing false positives and false negatives.

[0030] S120: Input the preprocessed time series data into a feature extraction model based on a multi-scale patch mixing network for feature extraction, wherein the multi-scale patch mixing network consists of a time mixer and a space mixer; wherein the time mixer uses convolution kernels of different sizes to capture short-term and long-term dependencies and extract variable time series information; and the space mixer performs feature aggregation based on a maximum variance framework to extract multi-scale information on the variable dimension.

[0031] In this step S120, a multi-scale patch hybrid network is used for feature extraction, and different channels and kernel sizes are used to extract multi-scale window features of the time series. Different channels correspond to different dimensions, and different kernel sizes are used to capture multi-time scale features. Finally, a multi-layer perceptron layer is added to generate a low-dimensional feature vector. The multi-scale patch hybrid network structure consists of a time mixer and a space mixer. Specifically: S121: The temporal mixer uses convolution kernels of different sizes to capture short-term and long-term dependencies and extract variable time series information, including: The temporal mixer is a key component of the multi-scale patch mixing network, which is responsible for extracting variable time series information along the time dimension. The input of the temporal mixer is usually a three-dimensional tensor whose dimensions can be expressed as (number of samples, time steps, number of features). This tensor contains the time series data of multiple samples, and each sample records the values ​​of multiple features at multiple time steps. The temporal mixer first uses a linear layer (also called a fully connected layer or dense layer) to process the input data. The linear layer maps the original feature space to a new feature space by applying a weight matrix and a bias term to the input data to capture the linear relationship in the input data. After the linear layer, the temporal mixer uses a multi-layer perceptron (MLP) to further extract complex nonlinear features. The MLP consists of multiple hidden layers, each of which contains a linear layer. The number of neurons in each hidden layer can be the same or different. Usually, as the number of layers increases, the number of neurons decreases to capture higher-level features. Activation functions (such as ReLU, Tanh, etc.) are usually used after each hidden layer to increase nonlinearity. To improve the training efficiency and performance of the network, the temporal mixer uses short connections or residual connections to add the input to the output. This step helps alleviate the gradient vanishing problem in deep networks and facilitates the learning of the identity mapping.

[0032] The time mixer extracts variable time series information by using linear layers and multi-layer perceptrons along the time dimension, and adds the input and output through short connections or residual connections. The specific formula is: , in, To input time series features, it can represent multidimensional time series data within a time window and is input into the model after being standardized; It is a linear layer, which is used to perform linear layer normalization on the input time series features to improve the stability and convergence speed of model training; It is a multi-layer perceptron used for feature transformation and extraction. It performs nonlinear mapping on the normalized data to learn deeper representations. is the final output feature, that is, the extracted variable time series information, which is the input feature On the basis of, the incremental adjustment after the multi-layer perceptron transformation enables the network to better learn the feature information in the time series.

[0033] S122: The spatial mixer performs feature aggregation based on the maximum variance framework and extracts multi-scale information on the variable dimension including: The spatial mixer is responsible for extracting variable dimension features and performing multi-scale feature aggregation based on the maximum variance framework. This process includes data periodicity analysis, multi-scale patch segmentation, feature extraction, weight calculation, and feature fusion. Data periodicity analysis: The spatial mixer first analyzes the data using the maximum variance framework to identify periodic patterns in the data. The maximum variance framework is a statistical method that can reveal the most significant components or periods in the data. By analyzing the data periodicity, the spatial mixer can determine the different scales (such as short-term fluctuations, long-term trends, etc.) present in the data. Multi-scale patch segmentation: Based on the results of the data periodicity analysis, the spatial mixer performs multi-scale patch segmentation. This step aims to divide the data into segments or "patches" of different scales so that features of different scales can be extracted and analyzed independently later. Feature extraction: For each scale patch, the spatial mixer uses a parameter-sharing multi-layer perceptron (MLP) to extract features along the variable dimension. Parameter sharing means that patches of different scales use the same MLP structure for feature extraction, which helps reduce the number of model parameters and improve generalization. Among them, the multi-layer perceptron includes an input layer, multiple hidden layers and an output layer; MLP is applied to patches of each scale to extract features on the variable dimension; the number of hidden layers and the number of neurons of the MLP can be adjusted according to the specific task and data characteristics. Weight calculation: The spatial mixer determines the weight of multi-scale feature aggregation based on the amplitude calculated by the maximum variance framework. The amplitude reflects the importance or significance of features at different scales, so it can be used as a weight to fuse multi-scale features. Feature fusion and short connection: The spatial mixer fuses multi-scale features according to the calculated weights and adds them to the output of the temporal mixer through short connections. This step realizes multi-scale data fusion analysis and feature representation enhancement.

[0034] The spatial mixer can realize multi-scale data fusion analysis and feature representation enhancement through the above process. Specifically, it includes: S1221: Analyze data cycles using the maximum variance framework to identify periodic patterns in the data; Further, step S1221: using the maximum variance framework to analyze the data periodicity to identify the periodic pattern in the data, specifically including: The maximum variance framework matches the modified circulant matrix, that is, finds a modified circulant matrix that maximizes the variance of the data CX obtained after the modified circulant matrix C transforms the original data X. By maximizing the variance, the most important feature of the time series data, namely the data periodic pattern, is extracted. The maximum variance framework extracts the data periodic pattern and enhances the time series modeling capability by matching the modified circulant matrix. The specific formula is: , in, It is a modified circulant matrix used to capture periodic patterns in time series data. The modified circulant matrix is ​​based on the standard circulant matrix and introduces additional adjustments (such as weighting, non-uniform shifting, etc.) to adapt to non-stationary time series, making it more flexible to real-world data. By modifying the circulant matrix, complex time dependencies can be better modeled, especially capturing key features in non-strictly periodic data (such as battery health data). :The goal of this optimization problem is to find a modified circulant matrix , so that the transformed data Has the largest variance, which emphasizes the changing pattern of periodic signals.

[0035] S1222: Based on the result of the data period analysis, multi-scale patch segmentation is performed, including: determining patch sizes of different scales according to the data period; dividing the original pre-processed time series data into patches of multiple scales, each patch containing data within a specific time range; S1223: For each scale patch, extract features along the variable dimension using a parameter-sharing multi-layer perceptron (MLP); S1224: Determine the multi-scale feature aggregation weight based on the amplitude calculated by the maximum variance framework, fuse the multi-scale features according to the calculated multi-scale feature aggregation weight, and add them to the output of the time mixer through a short connection to achieve multi-scale data fusion analysis and feature representation enhancement. The specific formula is: , in, is the ii-th scale patch determined based on the data periodicity analysis; I represents the number of scales; represents the features extracted by applying a multi-layer perceptron to the patch at the ii-th scale; is the aggregation weight of the ii-th scale feature determined by the magnitude calculated by the maximum variance framework; is the output of the time mixer; S is the final output of the space mixer.

[0036] Multi-scale features are extracted from time series data through a multi-scale patch hybrid network feature extraction model, providing effective feature representation for subsequent contrastive representation learning and classification tasks.

[0037] S130: In the self-supervised contrastive representation learning phase, the feature extraction model is trained using a triplet loss function to optimize the time series features extracted by the network, and the learned feature representation is used to find a semantically meaningful nearest neighbor and farthest neighbor for each sample; S131: In the self-supervised contrastive representation learning stage, the feature extraction model is trained using a triplet loss function to optimize the time series features extracted by the network, including: The feature extraction model is trained using a triplet loss function to minimize the distance between the anchor sample and the positive sample, while maximizing the distance between the anchor sample and the negative sample, and the difference is at least a preset interval. ; If the distance between the negative sample and the anchor point does not exceed the distance between the positive sample and the anchor point plus the preset interval , then there will be a loss, otherwise the loss is zero, thereby optimizing the time series features extracted by the feature extraction model to better distinguish normal and abnormal samples. The specific formula is: , in, is the anchor sample, that is, the current sample data points; is a positive sample, that is, a sample that belongs to the same category or is similar to the anchor sample; Negative samples are samples that are different or dissimilar to the anchor sample categories; It is a feature extraction function used to map input data to the embedding space; is the square of the Euclidean distance between the anchor point and the positive sample, indicating the distance between similar samples; is the square of the Euclidean distance between the anchor point and the negative sample, indicating the distance between dissimilar samples; The interval is used to ensure that the distance between the negative sample and the anchor point is greater than the distance between the positive sample and the anchor point, which is usually a hyperparameter; Right now , ensuring that the loss value is non-negative.

[0038] At this stage, a triplet loss function is set for contrastive learning, with the goal of making similar samples closer in the feature space and different samples farther away in the feature space. By setting the triplet loss function to train the model, the model can learn an effective feature representation for distinguishing similar and dissimilar samples.

[0039] S132: Use the learned feature representation to find semantically meaningful nearest and farthest neighbors for each sample, generate prior information, capture fine-grained fluctuations and overall trends of time series, and enhance the richness of feature representation to improve performance in the self-supervised classification stage.

[0040] Specifically, step S132 includes: S1321: Process the time series data through a multi-scale patch hybrid network. The network will output one or more feature vectors, which contain the feature information of the time series data at different scales. Each sample will correspond to one or more feature vectors, which can be regarded as the representation of the sample in the feature space; S1322: In order to compare the similarity between feature vectors, select a suitable measurement method, and commonly used measurement methods include Euclidean distance, cosine similarity, etc. For each pair of samples, calculate the distance or similarity between their feature vectors to provide an indicator for measuring the semantic similarity between samples; S1323: Find the nearest neighbor and the farthest neighbor: (1) Searching for the nearest neighbor: For each sample, calculate its distance or similarity with all other samples in the data set. Then, sort the samples according to these distances or similarities. Among the sorted samples, the sample with the smallest distance (or largest similarity) to the current sample is the nearest neighbor.

[0041] (2) For each sample, we calculate its distance or similarity with all other samples in the data set, and then sort the samples according to these distances or similarities. Among the sorted samples, the sample with the largest distance (or smallest similarity) to the current sample is the farthest neighbor.

[0042] Through the above process, the learned feature representation can be used to find the semantically meaningful nearest neighbors and farthest neighbors for each sample.

[0043] 2. Self-supervised classification stage S140: In the self-supervised classification stage, classification is performed based on the nearest neighbor and farthest neighbor information of all feature samples, and the classification model is trained through a custom loss function to achieve the best classification effect; In step S140, classification is performed based on the nearest neighbor and farthest neighbor information of all feature samples obtained in step S130, and a learnable method is incorporated as a priori to maximize the similarity between the window representation and the nearest neighbor and minimize the similarity between the window representation and the farthest neighbor. The model is trained by a custom loss function, and the loss function formula is: , in, Indicates windows; represents the nearest neighbor; represents the farthest neighbor; represents a similarity measure; Control the weight of nearest neighbor similarity optimization; Control the weight of the farthest neighbor similarity optimization; Indicates Windows Its nearest neighbors The similarity between Indicates Windows Its furthest neighbor ; α is a hyperparameter used to control the weight of the regularization term; H(x) represents entropy.

[0044] This loss function maximizes the similarity between the window representation and its nearest neighbors, minimizes the similarity between the window representation and its farthest neighbors, and applies the abstract loss to prevent overfitting. Finally, a discriminative representation is learned to distinguish normal and abnormal windows, and the classification performance of the model is optimized.

[0045] In the classification stage, a suitable classification model can be selected or designed, such as a fully connected layer, support vector machine, etc., to classify the extracted features.

[0046] S150: Use the trained model to classify and predict the new time series data of the battery and output the battery health status.

[0047] In step S150, the time series data features after feature extraction and self-supervised learning are used to predict the new time series data using the trained model, and output information such as health status (normal, abnormal, etc.). Specifically, after completing the model training, the category allocation and majority class of the data set are determined through the inference process, further including: S151: first use the trained classification model to classify the battery data, identify normal and abnormal battery data, then count the proportion of the majority class (i.e., normal battery data), calculate the probability that the new window belongs to the majority class, determine whether the new window is abnormal based on the probability, and output the abnormal score of the battery health status indicator; In the process of self-supervised reasoning, the probability P(y=c|x) that the new window belongs to the majority class is calculated to determine whether the window is abnormal. The formula is: , Among them, x represents the sample; Indicates category; represents the majority class; Represents a given sample After that, it belongs to the category The probability is used to determine whether a window is abnormal. When it is lower, it indicates that the sample The probability of belonging to the majority class (normal class) is small, so it can be judged as abnormal; For the model to sample Belongs to category The prediction score of is the total number of categories; For indexed categories The prediction score of , so that it is always positive, which is convenient for normalization; It is the exponential sum of the prediction scores of all categories, which plays a normalization role to ensure that the sum of the probabilities of all categories is 1.

[0048] Based on this probability, it is determined whether the rate mouth is abnormal, and an abnormality score can be generated for further analysis. The lower the score, the higher the possibility of abnormality. The reasoning process will eventually give a judgment on whether the new window is abnormal. By calculating the probability that the new window belongs to the majority class, if the probability P(y=c|x) is lower than a preset threshold (set according to the specific application scenario and the tolerance for false positives and false negatives), the window is judged to be abnormal; if the probability P(y=c|x) is higher than the threshold, it is judged to be normal.

[0049] For battery health status monitoring, the input is the current data of the battery (such as voltage, current, temperature, etc.), and the model outputs the abnormal score of the battery health status indicator. Assuming that the threshold of health status monitoring is set to 0.8, if the predicted value of the current health status of the battery is 0.6, it can be judged that the battery is in an abnormal state.

[0050] S152: Use the trained model to extract features from the battery health status data of the battery, which includes voltage, current, temperature, capacity data, battery internal resistance, service life and historical cycle count, etc. The battery health status data after feature extraction is input into the remaining life prediction model for prediction, and the remaining life prediction value of the battery is output. If the remaining life prediction value is less than a certain threshold, an alarm is triggered or a suggestion to replace the battery is made.

[0051] Specifically, the new time series data is inputted for feature extraction through a multi-scale patch hybrid network. Then, through self-supervised learning, that is, training in contrastive representation learning and classification stages, the model can learn the complex relationship between time series data and health status and remaining life, and then input the extracted features into the remaining life prediction model such as a regressor (such as a fully connected layer or support vector regression SVR, etc.), and output the remaining life, expressed as time or number of charge and discharge cycles.

[0052] For the remaining life prediction, the input is the current health status data of the battery (including voltage, current, temperature capacity data, battery internal resistance, service life, historical cycle count, etc.) to the remaining life prediction model. If the model predicts that the remaining life is less than a certain threshold (such as 50 hours), an alarm is triggered or a suggestion to replace the battery is made.

[0053] Preferably, in some embodiments, in order to improve the classification performance, the feature extraction model may be fine-tuned. The fine-tuning process is to adjust the parameters of the last few layers of the feature extraction model to adapt to a specific classification task while keeping most of the parameters of the feature extraction model unchanged.

[0054] Preferably, in some embodiments, model evaluation and optimization are also included: the trained model is evaluated using a validation set or a test set to measure its performance. Evaluation indicators may include accuracy, recall, F1 score, etc. The model is optimized according to the evaluation results, such as adjusting hyperparameters, changing the model architecture, increasing training data, etc. The optimized model will have better generalization ability and performance.

[0055] According to the above embodiment of the present invention, by introducing a multi-scale patch hybrid network, the multi-scale characteristics of time series data are fully utilized to improve the model's prediction accuracy of battery health status and remaining life. By designing a custom loss function, the model can better distinguish between normal and abnormal data, further improving the accuracy of the prediction. Through self-supervised learning and feature extraction, the generalization ability of the model is improved.

[0056] Experimental comparison: The quarterly and weekly data sets of batteries are used for comparison to demonstrate the stability of the method of the present invention in long-term and short-term predictions.

[0057] Calculate the symmetric mean percentage error to measure the prediction error. The result range is from 0 to 1, and the smaller the result, the better the prediction effect: , in, : Total number of samples (number of prediction points).

[0058] : True value (true battery health status or service life).

[0059] : Predicted value (health status or service life predicted by the model).

[0060] : The absolute error between the predicted value and the true value.

[0061] : The average of the true value and the predicted value, used to normalize the error.

[0062] Calculate the mean absolute scaled error to measure whether the model's error is lower than the simple average rate of change of the time series. A mean absolute scaled error less than 1 indicates that the model is better than directly using the average rate of change of the time series: , in, : The total number of samples in the test dataset.

[0063] : Reference series length for seasonal or trend data.

[0064] : true value.

[0065] : predicted value.

[0066] : The mean absolute error between the predicted value and the true value.

[0067] That is, the rate of change in a time series (calculating the average absolute change in historical data).

[0068] Centralized kernel alignment is used to evaluate the effect of representation learning at different levels, and the centralized kernel alignment value between the hidden layer representation and the actual battery life distribution of the present invention is calculated. , the value range is between 0 and 1. The higher the value, the more similar the two distributions are, indicating that the model has learned features close to the true distribution. Compared with the representation capabilities of other methods, the advantages of the method of the present invention in feature extraction are verified.

[0069] , G: Gram matrix of the first set of feature representation (neural network output).

[0070] Q: Gram matrix represented by the second set of features (true battery health status or service life distribution).

[0071] : The trace of the matrix, which represents the sum of the eigenvalues ​​of the matrix.

[0072] Table 1 Experimental comparison results

[0073] In this experiment on battery energy storage health status monitoring and remaining life prediction, we conducted a comprehensive comparative evaluation of the method of the present invention with the traditional convolutional neural network. In terms of prediction error, measured by the two key indicators of symmetric mean percentage error and mean absolute scaled error, the method of the present invention showed significant advantages, and its prediction error was effectively reduced compared with the convolutional neural network, which shows that the method of the present invention has significant advantages in prediction accuracy and can more accurately estimate the health status and remaining life of battery energy storage.

[0074] The experimental comparison results are shown in Table 1. From the perspective of characterization capability evaluation, the centralized kernel alignment value of the method of the present invention reaches 0.812, while that of the convolutional neural network is only 0.693. This clearly shows that the method of the present invention has stronger characterization capability and can better extract and express key features in the data, thereby providing better feature representation for subsequent analysis and prediction.

[0075] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0076] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An artificial intelligence driven method for predicting the health status and remaining life of battery energy storage, characterized in that: include: Obtain time series data of various characteristics and status information of the battery at historical time points and perform preprocessing; Various characteristics and status information including voltage data, current data, temperature data, capacity data, battery internal resistance, service life and historical cycle count; Inputting the preprocessed time series data into a feature extraction model based on a multi-scale patch mixing network for feature extraction, wherein the multi-scale patch mixing network is composed of a time mixer and a space mixer; In the self-supervised contrastive representation learning stage, the feature extraction model is trained using a triplet loss function to optimize the time series features extracted by the network, and the learned feature representation is used to find semantically meaningful nearest neighbors and farthest neighbors for each sample; In the self-supervised classification stage, classification is performed based on the nearest neighbor and farthest neighbor information of all feature samples, and the classification model is trained through a custom loss function to achieve the best classification effect; Use the trained model to classify and predict the new time series data of the battery and output the battery health status.

2. The method according to claim 1, characterized in that in, Preprocessing of the time series data of various characteristics and status information of the battery at historical time points includes: Use instance normalization to normalize the time series window; Point anomalies and subsequence anomalies are used in the standardized time series data, and the dimensions and starting time are randomly selected and anomaly injected to construct abnormal samples.

3. The method according to claim 2, characterized in that in, The temporal mixer uses convolution kernels of different sizes to capture short-term and long-term dependencies and extract variable time series information; the spatial mixer performs feature aggregation based on the maximum variance framework to extract multi-scale information on the variable dimension.

4. The method according to claim 3, characterized in that in, The temporal mixer uses convolution kernels of different sizes to capture short-term and long-term dependencies and extract variable time series information, including: The time mixer extracts variable time series information by using linear layers and multi-layer perceptrons along the time dimension, and adds the input and output through short connections or residual connections. The specific formula is: ， in, is the input time series feature; It is a linear layer, which is used to perform linear layer normalization on the input time series features; It is a multi-layer perceptron used for feature transformation and extraction, and performs nonlinear mapping on the normalized data to learn deeper representations; is the final output feature, that is, the extracted variable time series information.

5. The method according to claim 3, characterized in that: in, The spatial mixer performs feature aggregation based on the maximum variance framework and extracts multi-scale information on the variable dimension including: Data cycles were analyzed using a maximum variance framework to identify periodic patterns in the data; Based on the results of data period analysis, multi-scale patch segmentation is performed, including: determining the patch sizes of different scales according to the data period; dividing the original pre-processed time series data into patches of multiple scales, each patch containing data within a specific time range; For each scale of patch, a parameter-sharing multilayer perceptron is used to extract features along the variable dimension; The multi-scale feature aggregation weight is determined based on the amplitude calculated based on the maximum variance framework, and the multi-scale features are fused according to the calculated multi-scale feature aggregation weight, and added to the output of the time mixer through a short connection. The specific formula is: ， in, is the i-th scale patch determined based on the data periodicity analysis; n represents the number of scales; represents the features extracted by applying a multi-layer perceptron to the i-th scale patch; is the aggregation weight of the i-th scale feature determined by the magnitude calculated by the maximum variance framework; is the output of the time mixer; S is the final output of the space mixer.

6. The method according to claim 5, characterized in that in, The use of a maximum variance framework to analyze data cycles to identify periodic patterns in the data includes: The maximum variance framework matches the modified circulant matrix, that is, finds a modified circulant matrix that maximizes the variance of the data CX obtained after the modified circulant matrix C transforms the original data X. By maximizing the variance, the most important feature in the time series data, namely the data periodic pattern, is extracted.

7. The method according to claim 3, characterized in that in, In the self-supervised contrastive representation learning stage, the feature extraction model is trained using a triplet loss function to optimize the time series features extracted by the network, including: The feature extraction model is trained using a triplet loss function to minimize the distance between the anchor sample and the positive sample, while maximizing the distance between the anchor sample and the negative sample, and the difference is at least a preset interval. ; If the distance between the negative sample and the anchor point does not exceed the distance between the positive sample and the anchor point plus the preset interval , a loss will occur, otherwise the loss is zero, thereby optimizing the time series features extracted by the feature extraction model to better distinguish normal and abnormal samples.

8. The method according to claim 5, characterized in that in, In the self-supervised classification stage, classification is performed based on the nearest neighbor and farthest neighbor information of all feature samples, and the classification model is trained through a custom loss function to achieve the best classification effect, including: The classification model is trained by a custom loss function so that the classification model can accurately classify the nearest neighbor and farthest neighbor information of all feature samples; the loss function maximizes the similarity between the window representation and its nearest neighbor and minimizes the similarity between the window representation and its farthest neighbor. The loss function formula is: ， in, Indicates windows; represents the nearest neighbor; represents the farthest neighbor; represents a similarity measure; Control the weight of nearest neighbor similarity optimization; Control the weight of the farthest neighbor similarity optimization; Indicates Windows Its nearest neighbors The similarity between Indicates Windows Its furthest neighbor ; α is a hyperparameter used to control the weight of the regularization term; H(x) represents entropy.

9. The method according to claim 7, characterized in that: in, Use the trained model to classify and predict the new battery time series data and output the battery health status, including: After model training is complete, the class assignment and majority class of the dataset are determined through the inference process, including: First, use the trained classification model to classify the battery data, identify normal and abnormal battery data, and then count the proportion of the majority class, that is, normal battery data; Calculate the probability that the new window belongs to the majority class, determine whether the new window is abnormal based on the probability, and output the abnormal score of the battery health status indicator.

10. The method according to claim 1, characterized in that in, The method further comprises: The trained model is used to extract features from the battery health status data of the battery, and the battery health status data after feature extraction is input into the remaining life prediction model for prediction, and the remaining life prediction value of the battery is output. If the remaining life prediction value is less than a certain threshold, an alarm is triggered or a suggestion to replace the battery is made.

Images