Battery performance prediction method and device and training method of battery performance prediction model

Abstract

The invention discloses a battery performance prediction method and device and a training method of a battery performance prediction model. The prediction method comprises the following steps: acquiring battery production data of a to-be-predicted battery; processing the battery production data in combination with explicit feature crossing and implicit feature crossing to obtain feature crossing data; performing feature learning and feature splicing based on the feature cross data through a multi-model integration architecture, and determining a high-dimensional feature vector corresponding to the feature cross data; and through the multi-model integration architecture, generating prediction performance parameters corresponding to the to-be-predicted battery based on the high-dimensional feature vectors. According to the scheme, the battery production data is processed through organic combination of explicit feature intersection and implicit feature intersection, and a good data basis is provided for model prediction; prediction performance parameters are generated through a multi-model integration architecture, and the prediction capability of multiple models can be comprehensively utilized; therefore, according to the scheme, the accuracy of predicting the battery performance can be improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence technology, and in particular to a method and device for predicting battery performance and a method for training a battery performance prediction model. Background Art

[0002] In recent years, with the rapid development of emerging application scenarios such as electric vehicles and energy storage power stations, the market size of lithium-ion batteries has grown rapidly. In the production and manufacturing process of lithium-ion batteries, the quality of battery performance directly affects the quality of the final product and user experience, so it is very important to predict battery performance.

[0003] In the prior art, a prediction model for battery performance can be established using monitoring data from the battery production process, and the battery performance can be predicted through the prediction model. However, using a single model to predict battery performance may have the following problems: the high-dimensional nature of battery data makes it difficult for the model to select the most relevant features; the nonlinear relationship between battery performance and its influencing factors is complex and difficult for the model to capture effectively. The above problems result in low accuracy in predicting battery performance using existing methods. Summary of the invention

[0004] The present invention provides a method and device for predicting battery performance and a training method for a battery performance prediction model, which can improve the accuracy of predicting battery performance.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting battery performance, comprising:

[0006] Obtain battery production data of the battery to be predicted;

[0007] Processing the battery production data by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data;

[0008] Through a multi-model integration architecture, feature learning and feature concatenation are performed based on the feature cross data to determine a high-dimensional feature vector corresponding to the feature cross data;

[0009] Through the multi-model integrated architecture, the predicted performance parameters corresponding to the battery to be predicted are generated based on the high-dimensional feature vector.

[0010] In a second aspect, an embodiment of the present invention provides a method for training a battery performance prediction model, comprising:

[0011] Acquire a sample set to be trained, wherein each sample to be trained in the sample set to be trained includes production data of a battery to be trained and actual performance parameters of a battery corresponding to the production data of the battery to be trained;

[0012] Processing the battery production data to be trained by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data to be trained;

[0013] Through the multi-model integrated architecture to be trained, feature learning and feature splicing are performed based on the feature cross data to be trained, and the high-dimensional feature vector to be trained corresponding to the feature cross data to be trained is determined;

[0014] Generate training prediction performance parameters of the battery corresponding to the battery production data to be trained based on the high-dimensional feature vector to be trained through the multi-model integrated architecture to be trained;

[0015] A loss function is constructed based on the actual performance parameters and the training prediction performance parameters, and the multi-model integration architecture to be trained is optimized by the loss function to obtain an optimized multi-model integration architecture.

[0016] In a third aspect, an embodiment of the present invention provides a device for predicting battery performance, comprising:

[0017] An acquisition module, used to acquire battery production data of the battery to be predicted;

[0018] A feature crossover module, used to process the battery production data by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data;

[0019] A feature learning and splicing module, used to perform feature learning and feature splicing based on the feature cross data through a multi-model integration architecture, and determine a high-dimensional feature vector corresponding to the feature cross data;

[0020] A prediction module is used to generate predicted performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector through the multi-model integrated architecture.

[0021] In a fourth aspect, an embodiment of the present invention provides a training device for a battery performance prediction model, comprising:

[0022] A first processing module is used to obtain a sample set to be trained, wherein each sample to be trained in the sample set to be trained includes production data of a battery to be trained and actual performance parameters of a battery corresponding to the production data of the battery to be trained;

[0023] A second processing module, used for processing the battery production data to be trained by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data to be trained;

[0024] A third processing module is used to perform feature learning and feature splicing based on the feature cross data to be trained through a multi-model integrated architecture to be trained, and determine a high-dimensional feature vector to be trained corresponding to the feature cross data to be trained;

[0025] A fourth processing module, configured to generate, through the multi-model integrated architecture to be trained, a training prediction performance parameter of a battery corresponding to the battery production data to be trained based on the high-dimensional feature vector to be trained;

[0026] The fifth processing module is used to construct a loss function based on the actual performance parameters and the training prediction performance parameters, and optimize the multi-model integration architecture to be trained through the loss function to obtain an optimized multi-model integration architecture.

[0027] In a fifth aspect, an embodiment of the present invention provides an electronic device, including:

[0028] at least one processor; and

[0029] a memory communicatively connected to the at least one processor; wherein,

[0030] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the method as described in the first aspect, or execute the method as described in the second aspect.

[0031] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect, or implements the method described in the second aspect.

[0032] The technical solution of the embodiment of the present invention processes battery production data through the organic combination of explicit feature crossover and implicit feature crossover, which can effectively mine the logical associations and nonlinear interactive relationships in the battery production data, achieve comprehensiveness and efficiency in feature extraction, and provide a good data foundation for subsequent model prediction; generate predicted performance parameters corresponding to the battery to be predicted through a multi-model integrated architecture, and comprehensively utilize the prediction capabilities of multiple models, which can not only enhance the model's adaptability to different data, effectively avoid the deviations that may occur in a single model under certain special data distributions, but also capture complex dynamic changes in the data; therefore, this solution can improve the accuracy of battery performance prediction.

[0033] It should be understood that the contents described in this section are not intended to identify 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

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0035] Figure 1 is a flow chart of a method for predicting battery performance according to Embodiment 1 of the present invention;

[0036] Figure 2 is a flow chart of a method for predicting battery performance according to Embodiment 2 of the present invention;

[0037] Figure 3 is a flowchart of a method for training a battery performance prediction model provided in accordance with Embodiment 3 of the present invention;

[0038] Figure 4 is a flow chart of a method for predicting battery performance according to a fourth embodiment of the present invention;

[0039] Figure 5 is a schematic diagram of the structure of a battery performance prediction device provided according to Embodiment 5 of the present invention;

[0040] Figure 6 is a structural schematic diagram of a battery performance prediction model training device provided in accordance with Embodiment 6 of the present invention;

[0041] Figure 7 It is a schematic diagram of the structure of an electronic device implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme 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 only 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 should fall within the scope of protection of the present invention.

[0043] It should be noted that the terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] Embodiment 1

[0045] Figure 1 This is a flow chart of a method for predicting battery performance according to Embodiment 1 of the present invention. This embodiment is applicable to the case of predicting battery performance. The method can be executed by a battery performance prediction device, which can be implemented in the form of software and / or hardware and integrated in an electronic device. Further, the electronic device includes but is not limited to: a computer, a laptop computer, etc.

[0046] like Figure 1 As shown, the method includes:

[0047] S110, obtaining battery production data of the battery to be predicted.

[0048] The battery to be predicted may be a battery for which battery performance prediction is to be performed. The battery production data may be data generated during the battery production process. The battery performance prediction may include, but is not limited to, battery capacity prediction, capacity classification prediction, self-discharge abnormality prediction, etc.

[0049] In this step, the data generated during the production process of the battery to be predicted can be obtained, which can include complete process data from liquid injection to detection and sorting; the acquired data is cleaned to remove abnormal data and irrelevant data in the acquired data; due to the differences in the numerical ranges between different features in the data, all numerical features in the cleaned data can be normalized to obtain the battery production data of the battery to be predicted.

[0050] S120. Process the battery production data by combining explicit feature crossover with implicit feature crossover to obtain feature crossover data.

[0051] Explicit feature crossover can be a feature crossover method that combines key features according to set rules to generate new features with physical meaning. For example, explicit feature crossover can be used to logically associate related parameters based on the battery manufacturing process to obtain new features.

[0052] Implicit feature crossover can be a feature crossover method that automatically captures nonlinear high-order interactions between features through a multi-layer perceptron (MLP) to generate complex features that are difficult to define explicitly.

[0053] The feature cross-linking data may be data obtained by combining explicit feature cross-linking and implicit feature cross-linking on battery production data.

[0054] In this step, explicit feature crossover and implicit feature crossover can be organically combined to perform feature crossover processing on the battery production data to obtain feature crossover data. Set the original feature corresponding to the battery production data before feature crossover to X∈R n×d , where n is the number of samples and d is the feature dimension, then X * =f cross (X) obtains feature crossover data X*, and fcross represents a feature crossover operation that organically combines explicit feature crossover with implicit feature crossover.

[0055] The explicit feature crossover and implicit feature crossover are organically combined for feature crossover processing. The advantage of this setting is that it can not only make full use of domain knowledge, but also mine potential feature patterns through deep learning, thereby achieving comprehensiveness and efficiency of feature extraction. It is particularly suitable for processing highly complex and nonlinear multidimensional data in battery production data.

[0056] S130. Through a multi-model integration architecture, feature learning and feature concatenation are performed based on feature cross-data to determine a high-dimensional feature vector corresponding to the feature cross-data.

[0057] The multi-model integrated architecture can be an architecture obtained by integrating multiple models and learning, and can be used to predict battery performance. There is no limit on the number and structure of the multiple models involved in the integration. For example, the multi-model integrated architecture can be obtained by integrating three different recursive neural network models.

[0058] In this step, the feature cross data can be input into each model in the multi-model integration architecture, and feature learning can be performed on the feature cross data to obtain the result of feature learning of each model; the result of feature learning of each model can be feature spliced ​​according to the dimension to obtain the high-dimensional feature vector corresponding to the feature cross data. Among them, the high-dimensional feature vector can be understood as the result obtained after feature learning and feature splicing of the feature cross data.

[0059] Feature learning can be understood as the process of automatically learning and extracting effective features from data that help solve problems. Feature concatenation can be understood as the process of combining multiple different features or feature representations into a new single representation.

[0060] S140, generating predicted performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector through a multi-model integrated architecture.

[0061] In this step, the high-dimensional feature vector can be processed through a multi-model integration architecture to further integrate the features of multiple models in the multi-model integration architecture; after processing the high-dimensional feature vector, the processing result can be input into a function for outputting the performance prediction result to obtain the predicted performance parameters corresponding to the battery to be predicted.

[0062] The predicted performance parameter may be a parameter obtained by predicting the performance of the battery to be predicted through a multi-model integrated architecture. The predicted performance parameter may be a parameter related to the performance of the battery, which is not limited here, and may be a specific value, or a classification category corresponding to the value, or a combination of the above two.

[0063] The function used to output the performance prediction result can be determined based on the actual application needs. For example, when the prediction needs to output a specific numerical value, it can be a nonlinear activation function; for example, when the prediction needs to output the classification category corresponding to the numerical value, it can be a classification function.

[0064] Among them, the nonlinear activation function is a key component of the nonlinear neural network, which can introduce nonlinear factors into the neural network, so that the network has the ability to learn complex patterns. The classification function can be a function that can achieve the classification task and obtain the classification category. There is no limitation on what the nonlinear activation function or classification function is, as long as it can achieve the corresponding function.

[0065] The technical solution of the embodiment of the present invention processes battery production data through the organic combination of explicit feature crossover and implicit feature crossover, which can effectively mine the logical associations and nonlinear interactive relationships in the battery production data, achieve comprehensiveness and efficiency in feature extraction, and provide a good data foundation for subsequent model prediction; generate predicted performance parameters corresponding to the battery to be predicted through a multi-model integrated architecture, and comprehensively utilize the prediction capabilities of multiple models, which can not only enhance the model's adaptability to different data, effectively avoid the deviations that may occur in a single model under certain special data distributions, but also capture complex dynamic changes in the data; therefore, this solution can improve the accuracy of battery performance prediction.

[0066] Embodiment 2

[0067] Figure 2It is a flow chart of a method for predicting battery performance provided according to the second embodiment of the present invention. This embodiment is based on the above-mentioned first embodiment, and further refines the high-dimensional feature vector corresponding to the feature cross data by performing feature learning and feature splicing based on feature cross data through a multi-model integration architecture; and further refines the predicted performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector generated through a multi-model integration architecture.

[0068] like Figure 2 As shown, the method includes:

[0069] S110, obtaining battery production data of the battery to be predicted.

[0070] This step is consistent with the description of S110 in the first embodiment and will not be repeated here.

[0071] S120. Process the battery production data by combining explicit feature crossover with implicit feature crossover to obtain feature crossover data.

[0072] This step is consistent with the description of S120 in Example 1 and will not be repeated here.

[0073] S131. Input the feature cross-data into each model included in the multi-model integration architecture for feature learning, and obtain multiple feature learning data at different time scales.

[0074] In the embodiment of the present invention, the battery production data may be data recorded in time series, and the feature cross-data may be data based on the time series.

[0075] In one embodiment, among the multiple models included in the multi-model integration architecture, different models correspond to different time dependency ranges. The time dependency range can be understood as the degree or span of historical information that the model can effectively consider in time series analysis. By having multiple models focus on different time dependency ranges, the ability of the model to capture features can be improved.

[0076] In this step, the feature cross data based on the time series can be input into each model focusing on different time-dependent ranges in the multi-model integration architecture for feature learning, and feature learning data of multiple different time scales can be obtained. Among them, the feature learning data can be the data for feature learning of the feature cross data. Exemplarily, taking the multi-model integration architecture including three models as an example, the above process can be expressed by the following formula:

[0077]

[0078] Among them, X *It can be feature cross data; M1, M2, and M3 can be feature learning operations corresponding to three different recursive neural network models with different time dependency ranges; f1 can be feature learning data corresponding to M1, f2 can be feature learning data corresponding to M2, and f3 can be feature learning data corresponding to M3. n can be the number of samples, and h1, h2, and h3 are the hidden dimensions of three different recurrent neural network models.

[0079] Optionally, the three different recurrent neural network models can be a Long Short-Term Memory (LSTM) model, a Gated Recurrent Unit (GRU) model, and a Recurrent Neural Network (RNN) model. Among them, LSTM is suitable for long-term dependency modeling, GRU performs well in short- to medium-term dynamic modeling, and RNN can quickly capture short-term dynamic features.

[0080] The fusion design of multiple network structures (such as LSTM, GRU and RNN) has stronger robustness when processing time series data, can capture complex dynamic changes, and can improve the adaptability to different batches of data.

[0081] S132. Concatenate multiple feature learning data according to their dimensions to obtain a high-dimensional feature vector corresponding to the feature cross data.

[0082] In this step, a feature concatenation (also called concat) strategy can be used to concatenate multiple feature learning data according to the dimension to obtain a high-dimensional feature vector corresponding to the feature cross data. For example, taking the multi-model integration architecture including three models as an example, the above process can be expressed by the following formula:

[0083]

[0084] Among them, f all It can be a high-dimensional feature vector obtained by concatenating multiple feature learning data f1, f2 and f3 according to the dimension. The concatenated dimension can be represented by h1+h2+h3.

[0085] The advantage of this setting is that it enables feature sharing among multiple models in the multi-model integration architecture, which can further optimize the parameter efficiency of the network and make the multi-model integration architecture have stronger generalization ability and prediction stability.

[0086] S141. Through a multi-model integration architecture, the high-dimensional feature vector is processed through a fully connected network to obtain a fully connected network processing result.

[0087] The fully connected network processing may be a process for further integrating the features of multiple models in the multi-model integration architecture. The fully connected network processing result may be a result of processing a high-dimensional feature vector through the fully connected network. In this step, the fully connected network processing result may be expressed as M share (f all ), M share is the operation processed by the fully connected network, f all is a high-dimensional feature vector.

[0088] S142. Based on the fully connected network processing result, a prediction performance category corresponding to the battery to be predicted is generated through a classification function.

[0089] S143. Based on the fully connected network processing result, a predicted performance value corresponding to the battery to be predicted is generated through a nonlinear activation function.

[0090] The execution order of S142 and S143 is not limited, and S142 and S143 are explained by the following content:

[0091] In an embodiment of the present invention, the predicted performance parameter may include a predicted performance value and a predicted performance category. The predicted performance value may be a form in which the predicted performance is expressed in a specific numerical value. The predicted performance category may be a form in which the predicted performance is expressed in a classification category. The predicted performance value and the predicted performance category may be obtained for the same performance prediction of the battery, such as both may be related to the capacity of the battery.

[0092] In one embodiment, the predicted performance value includes a predicted battery capacity, and the predicted performance category includes a predicted capacity category. The predicted battery capacity may be a predicted capacity of the battery. The predicted capacity category may be a classification category corresponding to the predicted battery capacity, such as a classification category that may include "high capacity", "normal" and "low capacity".

[0093] In practical applications, the classification function such as softmax function can be used to process the result M based on the fully connected network. share (f all ) generates the predicted performance category, that is, the predicted performance category is y class =softmax(M share (f all )). The softmax function may be a function for implementing a classification task, i.e., a task for generating a prediction performance category.

[0094] In practical applications, the result M can be processed based on the fully connected network through nonlinear activation functions such as Rectified Linear Unit (RELU). share (f all ) generates a predicted performance value, that is, the predicted performance value ypred =RELU(M share (f all )). The task of generating a prediction performance value can be understood as a regression task.

[0095] The technical solution of the embodiment of the present invention performs feature learning on feature cross data to obtain multiple feature learning data of different time scales, which can improve the model's ability to capture features; obtains a high-dimensional feature vector by feature concatenating multiple feature learning data by dimension, and predicts the model based on the high-dimensional feature vector, which can improve the generalization ability and prediction stability of the model; through the advantages of multi-model fusion, generates the predicted performance category and predicted performance value corresponding to the battery to be predicted, which can effectively improve the prediction accuracy and production efficiency, and significantly reduce the testing cost and time.

[0096] In one embodiment, the predicted performance value is generated by a regression task, and the predicted performance category is generated by a classification task; the method further comprises:

[0097] When a deviation between the predicted performance value and the reference performance value is greater than a set threshold, determining that the predicted performance value is a first predicted anomaly detection result corresponding to the multi-model integrated architecture, and the first predicted anomaly detection result indicates that the regression task is abnormal;

[0098] In a case where the predicted performance category does not match the performance category corresponding to the predicted performance value, determining that the predicted performance value and the predicted performance category are a second predicted anomaly detection result corresponding to the multi-model integrated architecture, the second predicted anomaly detection result indicating a joint anomaly of the regression task and the classification task;

[0099] The first predicted anomaly detection result and / or the second predicted anomaly detection result are fed back to the multi-model integration architecture.

[0100] In the prediction of battery performance, classification tasks and regression tasks have a natural correlation. The detection of predicted anomalies in a multi-model integrated architecture can be realized by combining classification tasks with regression tasks. The predicted anomaly detection result is the result of predicted anomaly detection.

[0101] Compare the predicted performance value with the reference performance value. If the deviation between the two is greater than a set threshold, the predicted performance value is determined to be a first predicted anomaly detection result, and the first predicted anomaly detection result indicates that the predicted performance value generated by the regression task is abnormal. The reference performance value may be a reference value set for the performance value to be predicted. There is no limitation on the reference performance value and the set threshold, and they can be selected based on actual application needs.

[0102] When the predicted performance category does not match the performance category corresponding to the predicted performance value, that is, the predicted performance category and the performance category corresponding to the predicted performance value are contradictory, such as the predicted performance category is category one and the performance category corresponding to the predicted performance value is category two, then the predicted performance value and the predicted performance category are determined to be the second predicted anomaly detection result, and the second predicted anomaly detection result indicates a joint anomaly of the predicted performance value generated by the regression task and the predicted performance category generated by the classification task, and the joint anomaly may indicate that the result output of at least one of the classification task and the regression task is abnormal.

[0103] Exemplarily, when the predicted performance value includes predicted battery capacity, and the predicted performance category includes predicted capacity category: if the deviation between the predicted battery capacity and the reference value of the battery capacity is greater than a set threshold, the predicted battery capacity is determined to be a first predicted anomaly detection result, and the first predicted anomaly detection result indicates that the regression task is abnormal; if the predicted capacity category is "normal", but the predicted battery capacity falls within the range of "low capacity" or "high capacity", the predicted battery capacity and the predicted capacity category are determined to be a second predicted anomaly detection result, and the second predicted anomaly detection result indicates a joint anomaly of the classification task and the regression task.

[0104] The first predicted anomaly detection result and / or the second predicted anomaly detection result are fed back to the multi-model integration framework, so that the multi-model integration framework performs further model optimization based on the existing anomalies.

[0105] By combining classification tasks with regression tasks, the multi-model integrated architecture can detect predictive anomalies. This collaborative analysis strategy can more accurately discover potential problems in the production process while reducing the missed detection rate.

[0106] Embodiment 3

[0107] Figure 3 This is a flowchart of a method for training a battery performance prediction model according to Embodiment 3 of the present invention. This embodiment is applicable to the case of training a model for battery performance prediction. The method can be performed by a training device for a battery performance prediction model, which can be implemented in the form of software and / or hardware and integrated in an electronic device. Furthermore, the electronic device includes but is not limited to: a computer, a laptop, etc.

[0108] It should be noted that the battery performance prediction model in the embodiment of the present invention is the multi-model integrated architecture in the above embodiment, that is, the embodiment of the present invention is a detailed description of the training process of the multi-model integrated architecture. For the contents not fully described in this embodiment, reference can be made to the above embodiments.

[0109] like Figure 3 As shown, the method includes:

[0110] S310, obtaining a set of samples to be trained, wherein each sample to be trained in the set of samples to be trained includes production data of a battery to be trained and actual performance parameters of a battery corresponding to the production data of the battery to be trained.

[0111] The sample set to be trained may be a set of multiple samples to be trained. Each sample to be trained may include production data of the battery to be trained and actual performance parameters of the battery corresponding to the production data of the battery to be trained, wherein the production data of the battery to be trained may be production data of the battery to be trained for model training, and the actual performance parameters may be performance parameters actually corresponding to the production data of the battery to be trained, such as the actual capacity of the battery, the classification of the capacity, etc.

[0112] In this step, the production data of the battery to be trained can be obtained during the production process of the battery, and the actual performance parameters of the battery corresponding to the production data of the battery to be trained can be recorded, and the production data of the battery to be trained and the corresponding actual performance parameters can be used as a sample to be trained; multiple samples to be trained can be collected in the same way to obtain a sample set to be trained.

[0113] S320, processing the battery production data to be trained by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data to be trained.

[0114] In this step, explicit feature crossover and implicit feature crossover can be organically combined to perform feature crossover processing on the battery production data to be trained to obtain feature crossover data to be trained. The feature crossover data to be trained can be data obtained by combining explicit feature crossover and implicit feature crossover processing on the battery production data to be trained.

[0115] S330. Through the integrated architecture of the multi-model to be trained, feature learning and feature concatenation are performed based on the feature cross data to be trained, and the high-dimensional feature vector to be trained corresponding to the feature cross data to be trained is determined.

[0116] The multi-model integrated architecture to be trained can be an integrated learning architecture that integrates multiple models to be trained, which can be used to predict battery performance. There is no limitation on the number and structure of the multiple models involved in the integration, such as integrating three different recursive neural network models, exemplarily including a long short-term memory network (LSTM) model, a gated recurrent unit (GRU) model, and a recurrent neural network (RNN) model.

[0117] In this step, the feature cross data to be trained can be input into each model in the multi-model integrated architecture to be trained, and feature learning can be performed on the feature cross data to be trained to obtain the result of feature learning of each model; the result of feature learning of each model can be feature spliced ​​according to the dimension to obtain the high-dimensional feature vector to be trained corresponding to the feature cross data to be trained. Among them, the high-dimensional feature vector to be trained can be understood as the result obtained after feature learning and feature splicing of the feature cross data to be trained.

[0118] S340, generating training prediction performance parameters of the battery corresponding to the battery production data to be trained based on the high-dimensional feature vector to be trained through the integrated architecture of the multi-model to be trained.

[0119] In this step, the high-dimensional feature vector to be trained can be processed through the multi-model integrated architecture to be trained, so that the features of multiple models in the multi-model integrated architecture to be trained can be further integrated; after the high-dimensional feature vector to be trained is processed, the processing result can be input into the function for outputting the result of performance prediction, so as to obtain the training prediction performance parameters of the battery corresponding to the battery production data to be trained.

[0120] The training prediction performance parameter can be the performance parameter of the battery corresponding to the production data of the battery to be trained predicted by the multi-model integrated architecture to be trained, which can be a specific value, or a classification category corresponding to the value, or a combination of the two. The function used to output the result of performance prediction can be a classification function such as a softmax function, or a nonlinear activation function such as RELU.

[0121] S350, constructing a loss function based on the actual performance parameters and the training prediction performance parameters, optimizing the multi-model integration architecture to be trained through the loss function, and obtaining an optimized multi-model integration architecture.

[0122] The loss function can be a function used to evaluate the difference between the actual performance parameters of the battery corresponding to the battery production data to be trained and the training predicted performance parameters predicted by the multi-model integrated architecture to be trained. The present invention does not limit the specific loss function to be used, as long as the corresponding function can be achieved.

[0123] In this step, a loss function can be constructed based on the actual performance parameters and the training prediction performance parameters. The multi-model integration architecture to be trained can be continuously optimized by adjusting the loss function, so that the training prediction performance parameters generated by the multi-model integration architecture to be trained are continuously close to the corresponding actual performance parameters, thereby obtaining the optimized multi-model integration architecture.

[0124] The technical solution of the embodiment of the present invention achieves comprehensiveness and efficiency of feature extraction through the organic combination of explicit feature crossover and implicit feature crossover during the training of the multi-model integrated architecture to be trained; the prediction capabilities of multiple models are comprehensively utilized for prediction, which can enhance the model's adaptability to different data and capture complex dynamic changes in the data; the multi-model integrated architecture to be trained is optimized through the loss function, so that the results predicted by the trained multi-model integrated architecture are continuously close to the actual results; therefore, the prediction accuracy of the multi-model integrated architecture trained by this solution is high.

[0125] In one embodiment, a loss function is constructed based on the actual performance parameters and the training predicted performance parameters, including:

[0126] Based on the actual performance parameters and the training prediction performance parameters, construct a classification loss corresponding to the classification task and a regression loss corresponding to the regression task; wherein the classification task is used to generate the training prediction performance category included in the training prediction performance parameters, and the regression task is used to generate the training prediction performance value included in the training prediction performance parameters;

[0127] The sum of the product of the classification loss and the first weight of the classification loss and the product of the regression loss and the second weight of the regression loss is determined as the loss function.

[0128] The training prediction performance parameters may include training prediction performance categories and training prediction performance values. The training prediction performance categories may be classification categories of battery performance predicted during the model training process, and the training prediction performance values ​​may be performance values ​​of battery performance predicted during the model training process, i.e., specific numerical values.

[0129] The classification loss Focal_loss corresponding to the classification task can be expressed by the following formula:

[0130] Focal_loss=-a t (1-p t )glog(p t )

[0131] Among them, p t is the model’s predicted probability for the correct classification category, α t is the weight of the category, and γ is the adjustment factor. Specifically, α t It can be set based on the type of classification category and the actual performance parameter. For example, if the types of classification categories include "low tolerance", "normal" and "high tolerance", and the actual performance parameter indicates that the actual classification categories in multiple samples are unbalanced, then α can be set t It is set to [1, 0.2, 1] to deal with the sample imbalance problem with abnormal capacity.

[0132] The regression loss Smooth_L1_Loss corresponding to the regression task can be expressed by the following formula:

[0133]

[0134] Among them, y pred Can be a training prediction performance parameter; y true It can be an actual performance parameter; δ can be a threshold of prediction error. Using this regression loss can achieve both the accuracy of mean square error and the robustness of absolute error.

[0135] Loss function L total It can be expressed by the following formula:

[0136] L total =w cls Focal_loss+w reg Smooth_L1_Loss

[0137] Among them, w cls is the first weight of the classification loss; w reg is the second weight of regression loss. There is no limit on the first weight and the second weight, which can be adjusted continuously in combination with the actual situation during the actual training process.

[0138] In one embodiment, the method further includes: dynamically adjusting the first weight and the second weight in the loss function at different stages of training the multi-model integrated architecture to be trained.

[0139] In the initial stage of training, in order to ensure the accuracy of the classification task, a larger first weight can be assigned to the classification loss; as the accuracy of the classification task gradually increases, the first weight of the classification task can be gradually reduced, while the second weight of the regression task can be increased to focus on the accuracy optimization of the training prediction performance value.

[0140] Due to the design of the loss function, the classification task and the regression task are highly coupled. Therefore, when the weight of the classification task is reduced and the weight of the regression task is increased, the accuracy of the classification task will not decrease. On the contrary, the shared learning ability brought about by the increase in the weight of the regression task may further enhance the robustness and generalization ability of the classification task.

[0141] By dynamically adjusting the weights of the classification task and the regression task through the joint loss, this design not only achieves the collaborative optimization of multi-objective tasks, but also amplifies the importance of accurate classification tasks, significantly improves the accuracy of classification tasks and regression tasks, and solves the common task conflict problem in multi-objective learning.

[0142] Embodiment 4

[0143] This embodiment is an exemplary description of the above-mentioned embodiment. Figure 4 1 is a flow chart of a method for predicting battery performance according to a fourth embodiment of the present invention, and is described by taking the predicted performance parameters including predicted battery capacity and predicted capacity category as an example, including the following steps:

[0144] 1. Data preprocessing: that is, obtaining the battery production data of the battery to be predicted;

[0145] 2. Feature crossover: that is, combining explicit feature crossover with implicit feature crossover to process battery production data and obtain feature crossover data;

[0146] 3. Multi-model ensemble learning: feature cross data is input into each model included in the multi-model ensemble architecture for feature learning, and multiple feature learning data of different time scales are obtained; each model included in the multi-model ensemble architecture is LSTM, GRU and RNN;

[0147] 4. Multi-model fusion and deep feature sharing to obtain predicted capacity category and predicted battery capacity: that is, multiple feature learning data are spliced ​​by dimension to obtain a high-dimensional feature vector corresponding to the feature cross data; through the multi-model integration architecture, the high-dimensional feature vector is processed through a fully connected network to obtain a fully connected network processing result; based on the fully connected network processing result, the predicted capacity category corresponding to the battery to be predicted is generated through a classification function; based on the fully connected network processing result, the predicted battery capacity corresponding to the battery to be predicted is generated through a nonlinear activation function.

[0148] 5. Constrain and optimize the multi-model integration architecture through correlation loss function and anomaly detection:

[0149] Correlation loss function: The loss function L in the above embodiment is total As the loss function of the multi-model integration architecture, the first weight of the classification task and the second weight of the regression task in the loss function are dynamically adjusted through the multi-task dynamic adjustment strategy;

[0150] Anomaly detection: that is, when the deviation between the predicted battery capacity and the reference value of the battery capacity is greater than the set threshold, the predicted battery capacity is determined to be the first predicted anomaly detection result, and the first predicted anomaly detection result indicates that the regression task is abnormal; when the predicted capacity category does not match the capacity category corresponding to the predicted battery capacity, the predicted battery capacity and the predicted capacity category are determined to be the second predicted anomaly detection result, and the second predicted anomaly detection result indicates a joint anomaly of the regression task and the classification task; the first predicted anomaly detection result and / or the second predicted anomaly detection result are fed back to the multi-model integration architecture.

[0151] The technical solution of the embodiment of the present invention proposes a method for capacity classification and regression prediction of battery capacity division process based on multi-objective deep learning, and realizes accurate and efficient capacity prediction and classification through hybrid model structure and feature extraction technology. The solution utilizes battery production data in the battery production process, integrates multi-task learning, attention mechanism and dynamic weight adjustment strategy, significantly improves the model's ability to capture key features and adapt to complex production data, and realizes accurate capacity classification and capacity prediction in the battery capacity division process.

[0152] Embodiment 5

[0153] Figure 5 FIG. 1 is a schematic diagram of a battery performance prediction device according to Embodiment 5 of the present invention. This embodiment is applicable to the case of predicting battery performance. Figure 5 As shown, the specific structure of the device includes:

[0154] An acquisition module 51 is used to acquire battery production data of a battery to be predicted;

[0155] A feature crossover module 52 is used to process the battery production data by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data;

[0156] A feature learning and splicing module 53 is used to perform feature learning and feature splicing based on feature cross data through a multi-model integration architecture, and determine a high-dimensional feature vector corresponding to the feature cross data;

[0157] The prediction module 54 is used to generate the prediction performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector through a multi-model integration architecture.

[0158] The battery performance prediction device provided in this embodiment obtains the battery production data of the battery to be predicted through the acquisition module; the battery production data is processed by the feature cross module in combination with explicit feature cross and implicit feature cross to obtain feature cross data; the feature learning and feature splicing are performed based on the feature cross data through the multi-model integrated architecture through the feature learning and splicing module, and the high-dimensional feature vector corresponding to the feature cross data is determined; the prediction module generates the prediction performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector through the multi-model integrated architecture. The scheme processes the battery production data through the organic combination of explicit feature cross and implicit feature cross, which can effectively mine the logical association and nonlinear interaction relationship in the battery production data, realize the comprehensiveness and efficiency of feature extraction, and provide a good data foundation for subsequent model prediction; the prediction performance parameters corresponding to the battery to be predicted are generated through the multi-model integrated architecture, and the prediction capabilities of multiple models are comprehensively utilized, which can not only enhance the adaptability of the model to different data, effectively avoid the deviation that may occur in a single model under certain special data distributions, but also capture the complex dynamic changes in the data; therefore, the scheme can improve the accuracy of battery performance prediction.

[0159] Furthermore, the battery production data is recorded in time series; the feature learning and splicing module 53 is specifically used for:

[0160] Inputting the feature cross data into each model included in the multi-model integrated architecture for feature learning, to obtain a plurality of feature learning data at different time scales;

[0161] The multiple feature learning data are feature concatenated according to dimension to obtain a high-dimensional feature vector corresponding to the feature intersection data.

[0162] Furthermore, among the multiple models included in the multi-model integration architecture, different models correspond to different time dependency ranges.

[0163] Furthermore, the predicted performance parameters include predicted performance values ​​and predicted performance categories; the prediction module 54 is specifically used for:

[0164] Through the multi-model integrated architecture, the high-dimensional feature vector is processed through a fully connected network to obtain a fully connected network processing result;

[0165] Based on the fully connected network processing result, generating a predicted performance category corresponding to the battery to be predicted through a classification function;

[0166] Based on the fully connected network processing result, a predicted performance value corresponding to the battery to be predicted is generated through a nonlinear activation function.

[0167] Further, the predicted performance value is generated by a regression task, and the predicted performance category is generated by a classification task; the device also includes:

[0168] A first determination module is used to determine, when a deviation between the predicted performance value and the reference performance value is greater than a set threshold, that the predicted performance value is a first predicted anomaly detection result corresponding to the multi-model integrated architecture, wherein the first predicted anomaly detection result indicates that the regression task is abnormal;

[0169] a second determination module, configured to determine, when the predicted performance category does not match the performance category corresponding to the predicted performance value, that the predicted performance value and the predicted performance category are a second predicted anomaly detection result corresponding to the multi-model integrated architecture, the second predicted anomaly detection result indicating a joint anomaly of the regression task and the classification task;

[0170] A feedback module is used to feed back the first predicted anomaly detection result and / or the second predicted anomaly detection result to the multi-model integrated architecture.

[0171] Further, the predicted performance value includes a predicted battery capacity, and the predicted performance category includes a predicted capacity category.

[0172] The battery performance prediction device provided in the embodiment of the present invention can execute the battery performance prediction method provided in the embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0173] Embodiment 6

[0174] Figure 6 1 is a schematic diagram of a structure of a training device for a battery performance prediction model according to Embodiment 6 of the present invention. This embodiment is applicable to the case of training a model for battery performance prediction. Figure 6 As shown, the specific structure of the device includes:

[0175] A first processing module 61 is used to obtain a sample set to be trained, wherein each sample to be trained in the sample set to be trained includes production data of a battery to be trained and actual performance parameters of a battery corresponding to the production data of the battery to be trained;

[0176] A second processing module 62 is used to process the battery production data to be trained by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data to be trained;

[0177] The third processing module 63 is used to perform feature learning and feature concatenation based on the feature cross data to be trained through the multi-model integrated architecture to be trained, and determine the high-dimensional feature vector to be trained corresponding to the feature cross data to be trained;

[0178] A fourth processing module 64 is used to generate a training prediction performance parameter of a battery corresponding to the battery production data to be trained based on the high-dimensional feature vector to be trained through the multi-model integrated architecture to be trained;

[0179] The fifth processing module 65 is used to construct a loss function based on the actual performance parameters and the training prediction performance parameters, and optimize the multi-model integrated architecture to be trained through the loss function to obtain an optimized multi-model integrated architecture.

[0180] The training device of the battery performance prediction model provided in the present embodiment obtains a sample set to be trained through a first processing module, wherein each sample to be trained in the sample set to be trained includes production data of the battery to be trained and actual performance parameters of the battery corresponding to the production data of the battery to be trained; the production data of the battery to be trained is processed by a second processing module in combination with explicit feature crossover and implicit feature crossover to obtain feature crossover data to be trained; the third processing module performs feature learning and feature concatenation based on the feature crossover data to be trained through a multi-model integrated architecture to be trained, and determines a high-dimensional feature vector to be trained corresponding to the feature crossover data to be trained; the fourth processing module generates training prediction performance parameters of the battery corresponding to the production data of the battery to be trained based on the high-dimensional feature vector to be trained through the multi-model integrated architecture to be trained; the fifth processing module constructs a loss function based on the actual performance parameters and the training prediction performance parameters, optimizes the multi-model integrated architecture to be trained through the loss function, and obtains an optimized multi-model integrated architecture. In the process of training the multi-model integrated architecture to be trained, this scheme achieves comprehensiveness and efficiency of feature extraction through the organic combination of explicit feature crossing and implicit feature crossing; the comprehensive use of the predictive capabilities of multiple models for prediction can enhance the model's adaptability to different data and capture complex dynamic changes in the data; the multi-model integrated architecture to be trained is optimized through the loss function, so that the predicted results of the trained multi-model integrated architecture are constantly close to the actual results; therefore, the prediction accuracy of the multi-model integrated architecture trained by this scheme is high.

[0181] Furthermore, the fifth processing module 65 is specifically configured to:

[0182] Based on the actual performance parameters and the training prediction performance parameters, constructing a classification loss corresponding to a classification task and a regression loss corresponding to a regression task; wherein the classification task is used to generate a training prediction performance category included in the training prediction performance parameters, and the regression task is used to generate a training prediction performance value included in the training prediction performance parameters;

[0183] The sum of the product of the classification loss and the first weight of the classification loss and the product of the regression loss and the second weight of the regression loss is determined as the loss function.

[0184] Furthermore, the device also includes:

[0185] A weight adjustment module is used to dynamically adjust the first weight and the second weight in the loss function at different stages of training the multi-model integrated architecture to be trained.

[0186] The training device for the battery performance prediction model provided in the embodiment of the present invention can execute the training method for the battery performance prediction model provided in the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method.

[0187] Embodiment 7

[0188] Figure 7 Schematic diagram of the structure of the electronic device implementing the embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0189] like Figure 7 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0190] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0191] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method for predicting battery performance or a method for training a battery performance prediction model.

[0192] In some embodiments, the method for predicting battery performance or the method for training a battery performance prediction model may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for predicting battery performance or the method for training a battery performance prediction model described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the method for predicting battery performance or the method for training a battery performance prediction model by any other appropriate means (e.g., by means of firmware).

[0193] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0194] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0195] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0196] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0197] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0198] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0199] 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 of the present invention can be achieved, and this document does not limit this.

[0200] 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. A method for predicting battery performance, characterized in that: include: Obtain battery production data of the battery to be predicted; Processing the battery production data by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data; Through a multi-model integration architecture, feature learning and feature concatenation are performed based on the feature cross data to determine a high-dimensional feature vector corresponding to the feature cross data; Through the multi-model integrated architecture, the predicted performance parameters corresponding to the battery to be predicted are generated based on the high-dimensional feature vector.

2. The method according to claim 1, characterized in that The battery production data is recorded in time series; feature learning and feature concatenation are performed based on the feature cross data through a multi-model integration architecture to determine a high-dimensional feature vector corresponding to the feature cross data, including: Inputting the feature cross data into each model included in the multi-model integrated architecture for feature learning, to obtain a plurality of feature learning data at different time scales; The multiple feature learning data are feature concatenated according to dimension to obtain a high-dimensional feature vector corresponding to the feature intersection data.

3. The method according to claim 2, characterized in that Among the multiple models included in the multi-model integration architecture, different models correspond to different time dependency ranges.

4. The method according to claim 1, characterized in that: The predicted performance parameters include predicted performance values ​​and predicted performance categories; generating predicted performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector through the multi-model integrated architecture includes: Through the multi-model integrated architecture, the high-dimensional feature vector is processed through a fully connected network to obtain a fully connected network processing result; Based on the fully connected network processing result, generating a predicted performance category corresponding to the battery to be predicted through a classification function; Based on the fully connected network processing result, a predicted performance value corresponding to the battery to be predicted is generated through a nonlinear activation function.

5. The method according to claim 4, characterized in that The predicted performance value is generated by a regression task, and the predicted performance category is generated by a classification task; the method further comprises: When a deviation between the predicted performance value and the reference performance value is greater than a set threshold, determining that the predicted performance value is a first predicted anomaly detection result corresponding to the multi-model integrated architecture, and the first predicted anomaly detection result indicates that the regression task is abnormal; In a case where the predicted performance category does not match the performance category corresponding to the predicted performance value, determining that the predicted performance value and the predicted performance category are a second predicted anomaly detection result corresponding to the multi-model integrated architecture, the second predicted anomaly detection result indicating a joint anomaly of the regression task and the classification task; Feedback the first predicted anomaly detection result and / or the second predicted anomaly detection result to the multi-model integration architecture.

6. The method according to claim 4, characterized in that The predicted performance value includes a predicted battery capacity, and the predicted performance category includes a predicted capacity category.

7. A method for training a battery performance prediction model, characterized in that: include: Acquire a sample set to be trained, wherein each sample to be trained in the sample set to be trained includes production data of a battery to be trained and actual performance parameters of a battery corresponding to the production data of the battery to be trained; Processing the battery production data to be trained by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data to be trained; Through the multi-model integrated architecture to be trained, feature learning and feature splicing are performed based on the feature cross data to be trained, and the high-dimensional feature vector to be trained corresponding to the feature cross data to be trained is determined; Generate training prediction performance parameters of the battery corresponding to the battery production data to be trained based on the high-dimensional feature vector to be trained through the multi-model integrated architecture to be trained; A loss function is constructed based on the actual performance parameters and the training prediction performance parameters, and the multi-model integration architecture to be trained is optimized by the loss function to obtain an optimized multi-model integration architecture.

8. The method according to claim 7, characterized in that Constructing a loss function based on the actual performance parameter and the training predicted performance parameter includes: Based on the actual performance parameters and the training prediction performance parameters, constructing a classification loss corresponding to a classification task and a regression loss corresponding to a regression task; wherein the classification task is used to generate a training prediction performance category included in the training prediction performance parameters, and the regression task is used to generate a training prediction performance value included in the training prediction performance parameters; The sum of the product of the classification loss and the first weight of the classification loss and the product of the regression loss and the second weight of the regression loss is determined as the loss function.

9. The method according to claim 8, characterized in that Also includes: At different stages of training the multi-model integrated architecture to be trained, the first weight and the second weight in the loss function are dynamically adjusted.

10. A battery performance prediction device, characterized in that: include: An acquisition module, used to acquire battery production data of the battery to be predicted; A feature crossover module, used to process the battery production data by combining explicit feature crossover and implicit feature crossover to obtain feature crossover data; A feature learning and splicing module, used to perform feature learning and feature splicing based on the feature cross data through a multi-model integration architecture, and determine a high-dimensional feature vector corresponding to the feature cross data; A prediction module is used to generate predicted performance parameters corresponding to the battery to be predicted based on the high-dimensional feature vector through the multi-model integrated architecture.

Images