Deep-learning-based battery life prediction and dynamic scheduling method for battery of battery replacement cabinet of rider

Abstract

The invention discloses a deep learning-based battery life prediction and dynamic scheduling method for a battery in a rider battery changing cabinet, and the method comprises the following steps: S1, collecting the multi-source time sequence data of the battery in the rider battery changing cabinet, and carrying out the preprocessing of the multi-source time sequence data; s2, constructing a variational auto-encoder model, and training the model by using the preprocessed multi-source time sequence data; s3, outputting future state data through a decoder, wherein the future state data comprises health state and life prediction; s4, introducing an improved butterfly optimization algorithm, and optimizing model hyper-parameters; s5, generating a scheduling strategy based on a battery life prediction result, and optimizing charging and discharging strategies and replacement time; and S6, performing dynamic scheduling on the battery by applying a scheduling strategy, updating the model regularly and optimizing hyper-parameters. According to the method, the service life and the health state of the battery in the battery replacing cabinet of the rider are accurately predicted through the variational auto-encoder of deep learning and the improved butterfly optimization algorithm, the scheduling strategy is optimized, and the efficiency and the reliability of the battery management system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning, and particularly to a method for predicting the battery life and dynamic scheduling of a rider's battery swapping cabinet based on deep learning. Background Art

[0002] With the continuous development of new energy and intelligent transportation technologies, electric vehicles are being more and more widely used in urban travel. Compared with traditional fuel vehicles, electric vehicles have the advantages of environmental protection and energy conservation. However, as the core component of electric vehicles, the performance and life of the battery directly affect the operation efficiency and safety of the whole vehicle. In recent years, as a new battery replacement method, the rider's battery swapping cabinet has gradually become an important carrier for the energy supply of electric vehicles. The management, life prediction and dynamic scheduling of the batteries in the rider's battery swapping cabinet have become the key technologies to improve the overall system operation efficiency and reduce the maintenance cost.

[0003] At present, there have been many research results in the field of battery management at home and abroad. Traditional battery management systems mainly rely on empirical rules and simple physical models to estimate the remaining life and health state of the battery. However, due to the comprehensive influence of various factors such as environmental temperature, charge-discharge cycles, and usage load in the actual application process of the battery, traditional methods often fail to accurately reflect the actual usage state of the battery, resulting in large prediction errors and affecting the maintenance and replacement decisions of the battery. In addition, the existing technologies do not adequately consider the dynamic scheduling problem of the batteries in the rider's battery swapping cabinet and fail to make full use of real-time data for intelligent scheduling, thus unable to achieve the optimal arrangement of the battery usage cycle, charge and discharge strategies, and replacement time.

[0004] In recent years, deep learning technologies have developed rapidly in various fields, and their powerful feature extraction and data modeling capabilities provide new ideas for battery life prediction. In particular, the variational autoencoder (VAE), as a generative model, can map high-dimensional, multi-source time-series data to a low-dimensional latent space and capture the implicit features in the data, thereby realizing the prediction of the future state of the battery. There are studies in the existing technologies that use deep learning models to predict the battery state, but most of them focus on the application of a single deep neural network or convolutional neural network and do not fully consider the complex relationship between the battery health state and the remaining life under the influence of multiple factors. In addition, the existing deep learning models are often affected by factors such as hyperparameter settings, model structure design, and training data noise in actual applications, resulting in the prediction accuracy not meeting the actual requirements.

[0005] In practical application scenarios, the batteries in the rider's battery swapping cabinet are not only affected by the physical and chemical property changes of the batteries themselves, but also by charging and discharging strategies, usage cycles, and environmental factors. Therefore, how to comprehensively utilize multi-source time series data, accurately predict the remaining life and health status of the batteries through deep learning technology, and generate reasonable scheduling strategies based on the prediction results has become a key technical problem that needs to be solved urgently. However, there are obvious defects in the existing technologies in this regard. First, traditional prediction methods lack the ability to comprehensively process multi-source time series data and are difficult to accurately capture the state changes of the batteries in different usage environments. Second, although deep learning models have shown great advantages in feature extraction and prediction, their performance highly depends on the hyperparameter settings of the models, such as the latent space dimension, learning rate, number of layers of the encoder and decoder, etc. In the existing methods, the optimization of these hyperparameters mainly relies on experience or simple search algorithms, lacking a targeted and intelligent adjustment mechanism. Third, due to the dynamic changes in environmental conditions and battery usage status, the existing scheduling strategies are often static and difficult to respond to the fluctuations in battery status in real time, resulting in lagging or inaccurate scheduling decisions.

[0006] Therefore, how to provide a method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose a method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning. The present invention makes full use of the variational autoencoder and the improved butterfly optimization algorithm, combines multi-source time series data, and intelligently processes the historical usage data and environmental data of the batteries in the rider's battery swapping cabinet to achieve accurate prediction of the remaining life and health status of the batteries in the rider's battery swapping cabinet. Further, based on the optimized prediction results, the present invention generates and optimizes the scheduling strategy, thereby dynamically adjusting the usage cycle, charging, discharging strategy, and replacement time of the batteries in the rider's battery swapping cabinet. The present invention has the advantages of high prediction accuracy, real-time scheduling response, strong system stability, and high intelligence.

[0008] The method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning according to the embodiment of the present invention includes the following steps:

[0009] S1. Collect multi-source time series data of the batteries in the rider's battery swapping cabinet and preprocess the multi-source time series data;

[0010] S2. Construct a variational autoencoder model, use the preprocessed multi-source time series data to train the variational autoencoder model, and map the multi-source time series data to the latent space through the encoder;

[0011] S3. Reconstruct the historical usage data and environmental data in the latent space into the future state data of the batteries in the rider's battery swapping cabinet through a decoder, and output the remaining life and health state distribution of the batteries in the rider's battery swapping cabinet in the future;

[0012] S4. Introduce an improved butterfly optimization algorithm to optimize the hyperparameters of the variational autoencoder model, and use the optimized variational autoencoder model to predict the life of the batteries in the rider's battery swapping cabinet, and output the remaining life and health state distribution of the batteries in each rider's battery swapping cabinet after optimization;

[0013] S5. Generate a scheduling strategy based on the battery life prediction results in the rider's battery swapping cabinet, and optimize the usage cycle, charging and discharging strategies, and replacement time of the batteries in the rider's battery swapping cabinet;

[0014] S6. Apply the scheduling strategy to dynamically schedule the batteries in the rider's battery swapping cabinet, and regularly update the variational autoencoder model and re-optimize the hyperparameters according to the real-time data feedback.

[0015] Optionally, the multi-source time-series data specifically includes battery voltage, temperature, charge and discharge cycle times, battery capacity, charging rate, environmental data, and rider usage behavior data, which are used to analyze the health state of the battery, life prediction, and optimize the battery scheduling strategy.

[0016] Optionally, the preprocessing of the multi-source time-series data specifically includes removing invalid data, removing outliers, and performing data standardization processing, which is used to improve the effect of battery life prediction and dynamic scheduling.

[0017] Optionally, the S2 specifically includes:

[0018] S21. Perform feature selection on the preprocessed multi-source time-series data, and identify the features most relevant to the battery life prediction in the rider's battery swapping cabinet through the principal component analysis method, including battery voltage, temperature, charge and discharge cycle times, battery capacity, charging rate, environmental data, and rider usage behavior data;

[0019] S22. Construct a variational autoencoder model according to the identified features, define the number of layers of the encoder and decoder and the number of neurons in each layer, and determine the dimension of the latent space;

[0020] S23. Use the preprocessed multi-source time-series data for preliminary training of the variational autoencoder model, map the input multi-source time-series data to the latent space through the encoder, and the decoder reconstructs the input data through the output of the latent space;

[0021] S24. Apply the backpropagation algorithm to train the variational autoencoder model, and optimize the weight parameters of the variational autoencoder model by minimizing the reconstruction error and KL divergence;

[0022] S25. Evaluate the training effect of the variational autoencoder model. Use cross-validation to analyze the performance of the variational autoencoder model on the training set. If there are situations such as large training errors or overfitting, adjust the structures of the encoder and decoder.

[0023] S26. After completing the training, use the variational autoencoder model to process new multi-source time-series data. Map the data to the latent space through the encoder to generate prediction data for the future state of the batteries in the rider's battery swapping cabinet.

[0024] Optionally, the specific steps of S3 are as follows:

[0025] S31. Input the preprocessed multi-source time-series data, including the historical usage data and environmental data of the batteries in the rider's battery swapping cabinet, through the trained variational autoencoder model.

[0026] S32. Map the input multi-source time-series data to the latent space through the encoder to generate a latent variable z ∈ R d , where R represents the set of real numbers and d represents the dimension of the latent space.

[0027] S33. Decode the latent variable z into the future state data of the batteries in the rider's battery swapping cabinet through the decoder to obtain the future health status and life prediction results of the batteries in the rider's battery swapping cabinet.

[0028] S34. Use the future state data of the batteries in the rider's battery swapping cabinet output by the decoder to further calculate the performance changes of the batteries in the rider's battery swapping cabinet under different usage conditions in the future, and generate the future behavior trajectory of the batteries in the rider's battery swapping cabinet.

[0029] S35. Generate a prediction error based on the predicted future health status value and remaining life prediction value of the batteries in the rider's battery swapping cabinet, and adjust the mapping of the latent space according to the prediction error.

[0030] Optionally, the latent variable specifically includes the mean, variance, and vector of the latent space, which are used to capture the implicit features of the data.

[0031] Optionally, the specific steps of S4 are as follows:

[0032] S41. When initializing the butterfly optimization algorithm, set the positions and velocities of the butterfly individuals, which represent the solutions of the hyperparameters of the variational autoencoder model. The position x i = {x i1 , x i2 , …, x id} is the hyperparameter vector, where each hyperparameter x ij represents a hyperparameter in the variational autoencoder model. According to the predicted future health status value h predand the predicted remaining life value t pred Based on the dynamic changes of

[0033]

[0034] where β1 represents the acceleration coefficient that affects the individual butterfly's self-approach to the optimal position, and β2 represents the acceleration coefficient that the individual butterfly approaches the global optimal position, h max is the maximum value of the health state of the battery in the rider's battery replacement cabinet, t max is the predicted remaining life value of the battery in the rider's battery replacement cabinet, and are the acceleration coefficients at the current iteration time t, and represent the updated values of the acceleration coefficients;

[0035] S42. Adjust the flight direction of the butterfly according to the current position, global optimal solution, and local optimal solution:

[0036]

[0037] where is the speed of the butterfly, g i is the position of the global optimal solution, is the position vector of the i-th butterfly individual in the (t + 1)-th generation, the position vector of the i-th butterfly individual in the t-th generation, min is the operation of finding the minimum value. Considering the predicted future health state value h pred of the battery in the rider's battery replacement cabinet and the actual usage load L load , an adaptive distance limit mechanism is introduced to control the maximum distance d between butterfly individuals max , and the adaptive distance limit mechanism can dynamically adjust the size of the search space according to the current usage state of the battery in the rider's battery replacement cabinet:

[0038]

[0039] where d initial is the initially set maximum distance, L max is the maximum value of the battery load in the rider's battery replacement cabinet, and α is the adjustment coefficient of the load factor;

[0040] S43. Update the speed of the butterfly individual according to the current position, historical best position, and global optimal solution position:

[0041]

[0042] where ω is the inertia weight, c1 and c2 are learning factors, rand1 and rand2 are random numbers, p iis the historical best position of the i-th butterfly, g i is the global best position, is the velocity vector of the i-th butterfly individual in the (t + 1)-th generation, is the velocity vector of the i-th butterfly individual in the t-th generation. In the case of large prediction errors in the battery health state and remaining life in the rider's battery swapping cabinet, increase the learning factors c1 and c2 to accelerate global search, while when the prediction error is small, decrease the learning factors;

[0043] S44. Evaluate the fitness of each butterfly individual and calculate the fitness of the individual according to the prediction results of the variational autoencoder model. Adopt a composite fitness function and simultaneously consider the prediction errors of the battery health state and remaining life in the rider's battery swapping cabinet:

[0044]

[0045] where, f(x i ) is the fitness value of the i-th butterfly individual, n is the number of samples, h true is the true health state of the battery in the rider's battery swapping cabinet, t true is the true remaining life of the battery in the rider's battery swapping cabinet, w1 and w2 are weights. If the prediction error of the battery health state in the rider's battery swapping cabinet is large, increase the weight of w1; if the prediction error of the remaining life is large, increase the weight of w2, and flexibly adjust the weights at different stages;

[0046] S45. Through multiple iterative optimizations, the butterfly optimization algorithm converges to the optimal hyperparameters The optimal hyperparameters will be used to train the variational autoencoder model;

[0047] S46. Use the optimized variational autoencoder model to predict the remaining life and health state of the battery in the rider's battery swapping cabinet. The variational autoencoder outputs the future state data of the battery in the rider's battery swapping cabinet according to the historical usage data and environmental data of the battery in the rider's battery swapping cabinet, and obtains the predicted value of the remaining life of the battery in each optimized rider's battery swapping cabinet and the predicted value of the health state

[0048] S47. Based on the prediction results, combined with the current usage load and temperature environmental factors of the battery in the rider's battery swapping cabinet, further fine-tune the prediction results of the remaining life and health state. If the battery in the rider's battery swapping cabinet is used under high load, appropriately adjust the remaining life according to the load situation; if the temperature of the battery in the rider's battery swapping cabinet is too high, correct the health state.

[0049] Optionally, the hyperparameters specifically include the dimension of the latent space, the learning rate, and the number of layers of the encoder and decoder, which are used to optimize the structure and training process of the variational autoencoder model.

[0050] Optionally, the S5 specifically includes:

[0051] S51. Calculate the usage cycle, charging strategy, and discharging strategy of the battery in each rider's battery swapping cabinet according to the predicted remaining battery life and predicted health status values output by the variational autoencoder model, and determine the optimal replacement time of the battery in the rider's battery swapping cabinet based on the predicted remaining battery life value in the rider's battery swapping cabinet; and the predicted health status value Calculate the usage cycle, charging strategy, and discharging strategy of the battery in each rider's battery swapping cabinet, and determine the optimal replacement time of the battery in the rider's battery swapping cabinet according to the predicted remaining battery life value in the rider's battery swapping cabinet;

[0052] S52. Based on the optimized predicted health status value of the battery in the rider's battery swapping cabinet, evaluate the current health status of the battery in the rider's battery swapping cabinet, and classify the battery in the rider's battery swapping cabinet by setting a health status threshold h thresh to generate scheduling strategies for different health status categories:

[0053]

[0054] S53. According to the optimized predicted remaining battery life value of the battery in the rider's battery swapping cabinet When the remaining battery life in the rider's battery swapping cabinet is less than the set threshold t thresh trigger a battery replacement reminder in the rider's battery swapping cabinet:

[0055]

[0056] S54. Based on the predicted remaining battery life value and the predicted health status value, optimize the charging and discharging strategies of the battery in the rider's battery swapping cabinet and adjust the usage cycle of the battery in the rider's battery swapping cabinet through a scheduling strategy optimization algorithm;

[0057] S55. According to the optimized health status and predicted remaining battery life value of the battery in the rider's battery swapping cabinet, combined with historical usage data and environmental data, generate the specific replacement time of the battery in each rider's battery swapping cabinet through an optimized scheduling strategy, and regularly update the scheduling strategy to cope with the real-time changing battery data and environmental conditions in the rider's battery swapping cabinet;

[0058] S56. Update the scheduling strategy in real time, and dynamically adjust the usage cycle, charging and discharging strategies, and replacement time of the battery in the rider's battery swapping cabinet according to the real-time battery feedback data in the rider's battery swapping cabinet.

[0059] The beneficial effects of the present invention are:

[0060] The present invention uses a variational autoencoder to deeply extract features from the historical usage data and environmental data of the batteries in the rider's battery swapping cabinet, and combines an improved butterfly optimization algorithm to intelligently adjust the hyperparameters, achieving accurate prediction of the remaining life and health status of the batteries in the rider's battery swapping cabinet. In the case where traditional methods are difficult to simultaneously consider the influence of multiple factors and the dynamic changes of the battery state, the present invention makes full use of deep learning technology to map high-dimensional and multi-source data to a low-dimensional latent space, thereby capturing the inherent laws of battery performance changes. At the same time, with the help of the improved butterfly optimization algorithm, through innovative mechanisms such as adaptive parameter adjustment, inter-individual distance control, and improved fitness evaluation, the hyperparameter configuration of the model is optimized, making the prediction results more stable and accurate.

[0061] Furthermore, the present invention generates a scheduling strategy based on the optimized remaining life prediction value and health status prediction value, and dynamically manages the usage cycle, charging, discharging strategy, and replacement time of the batteries in the rider's battery swapping cabinet. This intelligent scheduling method can respond in real time to the changes in the battery usage environment and state. By comprehensively considering the health status and remaining life prediction of the battery, it realizes the precision and intelligence of battery scheduling and maintenance, thereby reducing the maintenance cost, improving the battery utilization efficiency, and extending the overall life of the battery. Compared with the prior art, the present invention not only improves the prediction accuracy and response speed, but also makes the system more intelligent and stable, and can achieve efficient management and optimized operation of the batteries in the rider's battery swapping cabinet in complex and dynamic actual application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0063] Figure 1 is a flowchart of the method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning proposed by the present invention;

[0064] Figure 2 is a schematic structural diagram of the variational autoencoder model of the method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning proposed by the present invention;

[0065] Figure 3 is a flowchart of the working process of the improved butterfly optimization algorithm of the method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0067] Reference Figure 1 、 Figure 2 and Figure 3 , a method for predicting the battery life and dynamic scheduling of a rider's battery swapping cabinet based on deep learning, includes the following steps:

[0068] S1. Collect multi-source time-series data of the batteries in the rider's battery swapping cabinet and preprocess the multi-source time-series data;

[0069] S2. Build a variational autoencoder model, use the preprocessed multi-source time-series data to train the variational autoencoder model, and map the multi-source time-series data to the latent space through the encoder;

[0070] S3. Reconstruct the historical usage data and environmental data in the latent space into the future state data of the batteries in the rider's battery swapping cabinet by the decoder, and output the remaining life and health state distribution of the batteries in the rider's battery swapping cabinet in the future;

[0071] S4. Introduce an improved butterfly optimization algorithm to optimize the hyperparameters of the variational autoencoder model, use the optimized variational autoencoder model to predict the life of the batteries in the rider's battery swapping cabinet, and output the remaining life and health state distribution of the batteries in each rider's battery swapping cabinet after optimization;

[0072] S5. Generate a scheduling strategy based on the battery life prediction results in the rider's battery swapping cabinet, and optimize the usage cycle, charging and discharging strategies, and replacement time of the batteries in the rider's battery swapping cabinet;

[0073] S6. Apply the scheduling strategy to dynamically schedule the batteries in the rider's battery swapping cabinet, and regularly update the variational autoencoder model and re-optimize the hyperparameters according to the real-time data feedback.

[0074] By introducing a variational autoencoder model and an improved butterfly optimization algorithm, the present invention realizes accurate life prediction and dynamic scheduling management of the batteries in the rider's battery swapping cabinet. Compared with traditional battery management technologies, the present invention makes full use of multi-source time-series data and combines deep learning technologies, which not only improves the prediction accuracy of the remaining life and health state of the batteries, but also can respond to changes in the battery usage environment in real time, thereby optimizing the usage cycle, charging, discharging strategies, and replacement time of the batteries.

[0075] Specifically, through the variational autoencoder model, in-depth analysis of multi-source time-series data can be carried out to generate latent variables and capture the changing rules of the battery health state and remaining life. By optimizing the hyperparameters of the model with the improved butterfly optimization algorithm, the accuracy of the prediction results is further improved, making the battery life prediction more refined. When generating a scheduling strategy, the charging, discharging, and replacement strategies of the batteries are dynamically adjusted based on the prediction results to ensure the optimal operation of the batteries at each stage and effectively reduce the risk of overuse and loss of the batteries.

[0076] In addition, the introduction of a real-time data feedback mechanism enables the battery management system to have an adaptive adjustment ability, and it can regularly update the model and scheduling strategy according to the actual operating state of the battery, further improving the stability and efficiency of the system. Generally speaking, the present invention provides an efficient and intelligent battery management method, which has the advantages of accurate prediction, dynamic scheduling and adaptive optimization, and significantly improves the usage efficiency and safety of the batteries in the rider's battery swapping cabinet.

[0077] In this embodiment, the multi-source time series data specifically includes battery voltage, temperature, charge and discharge cycle times, battery capacity, charging rate, environmental data and rider usage behavior data, which are used to analyze the health status of the battery, life prediction and optimize the battery scheduling strategy.

[0078] In this embodiment, the preprocessing of the multi-source time series data specifically includes removing invalid data, removing outliers, and performing data standardization processing, which is used to improve the effects of battery life prediction and dynamic scheduling.

[0079] In this embodiment, S2 specifically includes:

[0080] S21. Perform feature selection on the preprocessed multi-source time series data, and identify the features most relevant to the battery life prediction in the rider's battery swapping cabinet through the principal component analysis method, including battery voltage, temperature, charge and discharge cycle times, battery capacity, charging rate, environmental data and rider usage behavior data;

[0081] S22. According to the identified features, construct a variational autoencoder model, define the number of layers of the encoder and decoder and the number of neurons in each layer, and determine the dimension of the latent space;

[0082] S23. Use the preprocessed multi-source time series data for preliminary training of the variational autoencoder model. The input multi-source time series data is mapped to the latent space through the encoder, and the decoder reconstructs the input data through the output of the latent space;

[0083] S24. Apply the backpropagation algorithm to train the variational autoencoder model, and optimize the weight parameters of the variational autoencoder model by minimizing the reconstruction error and KL divergence;

[0084] S25. Evaluate the training effect of the variational autoencoder model, adopt cross-validation, analyze the performance of the variational autoencoder model on the training set. If there are problems such as large training errors or overfitting, adjust the structures of the encoder and decoder;

[0085] S26. After the training is completed, use the variational autoencoder model to process the new multi-source time series data, map the data to the latent space through the encoder, and generate prediction data on the future state of the batteries in the rider's battery swapping cabinet.

[0086] Through the variational autoencoder model and feature selection method of the present invention, accurate prediction and optimization of the battery life in the rider's battery swapping cabinet are achieved. First, the principal component analysis (PCA) method is used to perform feature selection on the preprocessed multi-source time series data, and key features closely related to the battery life are identified, such as battery voltage, temperature, charge and discharge cycle times, etc. This feature selection process not only reduces the data dimension and improves the efficiency of model training, but also ensures that the model can focus on the factors most influential on battery life prediction, thereby improving the accuracy of the prediction results.

[0087] By constructing a variational autoencoder model, defining appropriate encoder and decoder structures, and determining the dimension of the latent space, the present invention can effectively perform deep learning processing on multi-source time series data and extract the latent features of battery performance changes. During the model training process, the weight parameters are optimized through the backpropagation algorithm, and the reconstruction error and KL divergence are minimized to ensure that the model's prediction of the future state of the battery is more accurate.

[0088] Furthermore, by evaluating the model training effect through cross-validation, the present invention can timely discover and solve problems such as large training errors or overfitting, and ensure the stability and reliability of the model on different data sets. Finally, the fully trained variational autoencoder model can process new multi-source time series data and output the prediction data of the future state of the battery in the rider's battery swapping cabinet, thereby providing a scientific basis for battery life management.

[0089] Through precise feature selection and optimization of the deep learning model, the present invention solves the problems of data complexity and model generalization ability in battery life prediction, and has high efficiency, accuracy and stability, and can effectively improve the performance and reliability of the battery management system.

[0090] In this embodiment, the specific content of S3 includes:

[0091] S31. Input the preprocessed multi-source time series data through the trained variational autoencoder model, including the historical usage data and environmental data of the battery in the rider's battery swapping cabinet;

[0092] S32. Map the input multi-source time series data to the latent space through the encoder to generate a latent variable z ∈ R d , where R represents the set of real numbers and d represents the dimension of the latent space;

[0093] S33. Decode the latent variable z into the future state data of the battery in the rider's battery swapping cabinet through the decoder to obtain the future health state and life prediction results of the battery in the rider's battery swapping cabinet;

[0094] S34. Further calculate the performance change of the battery in the rider's battery replacement cabinet under different future usage conditions by using the future state data of the battery in the rider's battery replacement cabinet output by the decoder, and generate the future behavior trajectory of the battery in the rider's battery replacement cabinet.

[0095] S35. Generate a prediction error based on the predicted future health state value and the predicted remaining life value of the battery in the rider's battery replacement cabinet, and adjust the mapping of the latent space according to the prediction error.

[0096] Through the training and optimization of the variational autoencoder model, the present invention realizes the accurate prediction of the future state of the battery in the rider's battery replacement cabinet and the generation of the behavior trajectory. In traditional battery management technologies, due to the data complexity and the ever-changing battery usage environment, the prediction accuracy of battery life and health state is relatively low. However, the present invention overcomes these limitations through the deep learning processing of multi-source time-series data.

[0097] First of all, through the trained variational autoencoder model, by inputting the historical usage data and environmental data of the battery in the rider's battery replacement cabinet, the model can effectively map this data to the latent space. The mapping of the latent space not only reduces the data dimension, but also can extract the implicit features of the battery performance, providing a more accurate description of the battery state. The encoder transforms these complex time-series data into latent variables, laying a solid foundation for the subsequent prediction of the battery health state and remaining life.

[0098] Through the decoder, the latent variables are reconstructed into the future state data of the battery in the rider's battery replacement cabinet, including the predicted health state and remaining life results. The innovation of this process lies in that the model can generate the future behavior trajectory of the battery under different usage conditions and predict the performance change of the battery in the future period of time. In this way, the present invention can provide more comprehensive and accurate battery management data to help realize the intelligent scheduling and maintenance of the battery.

[0099] Furthermore, by calculating the prediction error and adjusting the mapping of the latent space according to the error, the present invention improves the adaptive ability of the model, making the prediction of battery life and health state more accurate and dynamic. Real-time adjustment of the mapping of the latent space can effectively reduce error accumulation and improve the stability and reliability of the model in different usage environments and conditions.

[0100] Generally speaking, through the deep learning of the variational autoencoder and the optimization of the latent space, the present invention realizes the accurate prediction of the future state of the battery in the rider's battery replacement cabinet, with the advantages of high efficiency, accuracy and adaptability, providing strong technical support for the battery management system.

[0101] In this embodiment, the latent variables specifically include the mean, variance and vectors in the latent space, which are used to capture the implicit features of the data.

[0102] In this embodiment, step S4 specifically includes:

[0103] S41. When initializing the butterfly optimization algorithm, set the positions and velocities of butterfly individuals, which represent the hyperparameter solutions of the variational autoencoder model. The position x of each butterfly individual i ={x i1 ,x i2 ,…,x id} is a hyperparameter vector, where each hyperparameter x ij represents a hyperparameter in the variational autoencoder model. According to the dynamic changes of the predicted future health state h pred and the predicted remaining life t pred of the battery in the rider's battery replacement cabinet, adjust the acceleration coefficients β1 and β2:

[0104]

[0105] Among them, β1 represents the acceleration coefficient that affects the butterfly individual to approach the optimal position by itself, β2 represents the acceleration coefficient that the butterfly individual approaches the global optimal position, h max is the maximum value of the health state of the battery in the rider's battery replacement cabinet, t max is the predicted remaining life of the battery in the rider's battery replacement cabinet, and are the acceleration coefficients at the current iteration time t, and represent the updated values of the acceleration coefficients;

[0106] S42. Adjust the flight direction of the butterfly according to the current position, the global optimal solution, and the local optimal solution:

[0107]

[0108] Among them, v i t is the velocity of the butterfly, g i is the position of the global optimal solution, is the position vector of the i-th butterfly individual in the (t + 1)-th generation, the position vector of the i-th butterfly individual in the t-th generation, min is the operation of finding the minimum value. Considering the predicted future health state h pred and the actual usage load L load of the battery in the rider's battery replacement cabinet, introduce an adaptive distance limit mechanism to control the maximum distance d max between butterfly individuals. The adaptive distance limit mechanism can dynamically adjust the size of the search space according to the current usage state of the battery in the rider's battery replacement cabinet:

[0109]

[0110] Among them, d initial is the initially set maximum distance, and L max is the maximum value of the battery load in the rider's battery replacement cabinet, and α is the adjustment coefficient of the load factor;

[0111] S43. The speed of the butterfly individual is updated according to the current position, historical best position, and global optimal solution position:

[0112]

[0113] Among them, ω is the inertia weight, c1 and c2 are learning factors, rand1 and rand2 are random numbers, and p i is the historical best position of the i-th butterfly, and g i is the global best position, is the speed vector of the i-th butterfly individual in the (t + 1)-th generation, is the speed vector of the i-th butterfly individual in the t-th generation. In the case of large prediction errors in the battery health state and remaining life in the rider's battery replacement cabinet, increase the learning factors c1 and c2 to accelerate global search, and when the prediction error is small, reduce the learning factors;

[0114] S44. Perform fitness evaluation on each butterfly individual, and calculate the fitness of the individual according to the prediction results of the variational autoencoder model. Adopt a composite fitness function, and simultaneously consider the prediction errors of the battery health state and remaining life in the rider's battery replacement cabinet:

[0115]

[0116] Among them, f(x i ) is the fitness value of the i-th butterfly individual, n is the number of samples, h true is the true health state of the battery in the rider's battery replacement cabinet, t true is the true remaining life of the battery in the rider's battery replacement cabinet, and w1 and w2 are weights. If the prediction error of the battery health state in the rider's battery replacement cabinet is large, increase the weight of w1; if the prediction error of the remaining life is large, increase the weight of w2, and flexibly adjust the weights at different stages;

[0117] S45. Through multiple iterative optimizations, the butterfly optimization algorithm converges to the optimal hyperparameters The optimal hyperparameters will be used to train the variational autoencoder model;

[0118] S46. Use the optimized variational autoencoder model to predict the remaining life and health state of the battery in the rider's battery replacement cabinet. The variational autoencoder outputs the future state data of the battery in the rider's battery replacement cabinet according to the historical usage data and environmental data of the battery in the rider's battery replacement cabinet, and obtains the predicted value of the remaining life of each battery in the rider's battery replacement cabinet after optimization and the predicted value of the health state

[0119] S47. Based on the prediction results, in combination with the current usage load and temperature environment factors of the battery in the rider's battery replacement cabinet, further fine-tune the prediction results of the remaining life and health state. If the battery in the rider's battery replacement cabinet is used under high load, appropriately adjust the remaining life according to the load situation; if the temperature of the battery in the rider's battery replacement cabinet is too high, correct the health state.

[0120] The present invention optimizes the hyperparameters of the variational autoencoder model by introducing an improved butterfly optimization algorithm, significantly improving the accuracy of battery life prediction and health state assessment in the rider's battery replacement cabinet. The improved butterfly optimization algorithm adds a dynamic adjustment acceleration coefficient, an adaptive distance limit mechanism, a speed update mechanism, and a composite fitness function on the basis of the traditional butterfly optimization algorithm, ensuring that the optimal solution can be more efficiently searched when the prediction errors of the battery health state and remaining life are large.

[0121] First, the improved butterfly optimization algorithm dynamically adjusts the acceleration coefficient, so that when the prediction error of the health state or remaining life of the battery in the rider's battery replacement cabinet is large, the learning factor is increased to enhance the global search ability, so as to explore the optimal hyperparameters faster. On the contrary, when the prediction error is small, the learning factor is reduced to improve the local development ability, which can effectively avoid premature convergence to the local optimal solution and improve the stability of the optimization process.

[0122] Secondly, the maximum distance between butterfly individuals is dynamically adjusted through an adaptive distance limit mechanism, and the size of the search space is optimized according to the actual usage load and environmental factors of the battery. In this way, not only can the search be prevented from being too concentrated, but also the search range can be flexibly adjusted according to the current state of the battery, improving the search efficiency and accuracy.

[0123] In addition, the improved fitness evaluation mechanism combines the prediction errors of the health state and remaining life, adopts a composite fitness function, and flexibly adjusts the weights. When a certain prediction error is large, the weight in the fitness evaluation is enhanced, so that the optimization process pays more attention to the high-error prediction part, further improving the prediction accuracy.

[0124] Through these improvements, the butterfly optimization algorithm can better adapt to the dynamic changes of the battery life and health state, improve the training efficiency and prediction accuracy of the variational autoencoder model, making the battery management system more intelligent and reliable in practical applications. The optimized model can provide more accurate support for the dynamic scheduling of the battery in the rider's battery replacement cabinet, effectively extend the battery service life, reduce the maintenance cost, and improve the overall system operation efficiency.

[0125] In this embodiment, the hyperparameters specifically include the dimension of the latent space, the learning rate, and the number of layers of the encoder and decoder, which are used to optimize the structure and training process of the variational autoencoder model.

[0126] In this embodiment, S5 specifically includes:

[0127] S51. Calculate the usage cycle, charging strategy, and discharging strategy of the battery in each rider's battery swapping cabinet according to the predicted remaining battery life value and the predicted health state value output by the variational autoencoder model, and determine the optimal replacement time of the battery in the rider's battery swapping cabinet according to the predicted remaining battery life value of the battery in the rider's battery swapping cabinet; and the predicted health state value Calculate the usage cycle, charging strategy, and discharging strategy of the battery in each rider's battery swapping cabinet, and determine the optimal replacement time of the battery in the rider's battery swapping cabinet according to the predicted remaining battery life value of the battery in the rider's battery swapping cabinet;

[0128] S52. Based on the optimized predicted health state value of the battery in the rider's battery swapping cabinet, evaluate the current health status of the battery in the rider's battery swapping cabinet, and classify the battery in the rider's battery swapping cabinet by setting a health state threshold h thresh to generate scheduling strategies under different health state categories:

[0129]

[0130] S53. According to the optimized predicted remaining battery life value of the battery in the rider's battery swapping cabinet When the remaining battery life in the rider's battery swapping cabinet is less than the set threshold t thresh trigger a battery replacement reminder in the rider's battery swapping cabinet:

[0131]

[0132] S54. Based on the predicted remaining battery life value and the predicted health state value, optimize the charging and discharging strategies of the battery in the rider's battery swapping cabinet and adjust the usage cycle of the battery in the rider's battery swapping cabinet through a scheduling strategy optimization algorithm;

[0133] S55. According to the optimized health state and predicted remaining battery life value of the battery in the rider's battery swapping cabinet, combined with historical usage data and environmental data, generate the specific replacement time of the battery in each rider's battery swapping cabinet through an optimized scheduling strategy, and regularly update the scheduling strategy to cope with the real-time changing battery data and environmental conditions in the rider's battery swapping cabinet;

[0134] S56. Update the scheduling strategy in real time, and dynamically adjust the usage cycle, charging and discharging strategies, and replacement time of the battery in the rider's battery swapping cabinet according to the real-time battery feedback data in the rider's battery swapping cabinet.

[0135] The present invention provides an accurate method for generating a battery scheduling strategy in a rider's battery swapping cabinet by combining the predicted remaining battery life and the predicted health state output by an optimized variational autoencoder model. Through the optimization of the battery usage cycle, charging strategy, and discharging strategy, this method realizes the intelligent management of the battery, effectively improving the battery usage efficiency and the reliability of system operation.

[0136] First, by generating a scheduling strategy based on the predicted results of the remaining battery life and the health state, the usage cycle and charging / discharging strategies of the batteries in the rider's battery swapping cabinet are optimized. Especially when the predicted remaining battery life is low, the present invention can timely trigger a battery replacement reminder to prevent overuse of the battery, ensuring the safety and efficient operation of the battery.

[0137] Second, the batteries are classified based on the predicted health state values, and scheduling strategies for different health state categories are generated, effectively distinguishing the working states and performance changes of the batteries. For batteries with a poor health state, the scheduling strategy can prioritize replacement or charging to ensure the overall health level of the batteries in the system and the optimization of battery usage.

[0138] In addition, the present invention can also combine historical usage data and environmental data, not only making accurate predictions on the battery replacement time, but also regularly updating the scheduling strategy to cope with changes in environmental conditions and real-time data, thereby realizing dynamic scheduling. The real-time updated scheduling strategy ensures the flexibility and intelligence of the battery management system, making the maintenance of the batteries in the rider's battery swapping cabinet more precise and efficient.

[0139] Generally speaking, the scheduling strategy optimization method of the present invention has the advantages of high efficiency, intelligence, real-time response, etc., which can greatly improve the performance of the battery management system, extend the battery service life, reduce the maintenance cost, and improve the overall operation efficiency of the batteries in the rider's battery swapping cabinet.

[0140] Example 1:

[0141] To verify the feasibility of the present invention in implementation, the present invention is applied to the battery management system of shared bicycles in a certain city. This system is equipped with multiple rider battery swapping cabinets, providing battery replacement, charging, and management services for electric bicycles. To improve battery usage efficiency, extend battery life, and reduce maintenance costs, in this embodiment, a precise battery life prediction and dynamic scheduling method is proposed based on deep learning technology and optimization algorithms. The problem of battery management for shared bicycles is that the health status and remaining life of the battery are affected by various factors, including battery usage load, temperature, charge and discharge cycles, etc. Traditional battery management systems often adopt fixed charge and discharge cycles and lack real-time monitoring and dynamic adjustment of the battery health status. The method of the present invention predicts the health status and remaining life of the battery based on a variational autoencoder model and optimizes the hyperparameters by combining an improved butterfly optimization algorithm, making the battery management more intelligent, capable of accurately predicting the future state of the battery, and dynamically adjusting the charging, discharging strategies, and replacement time according to the battery state.

[0142] In the battery management system of this embodiment, first, multi-source time-series data of the batteries in the rider battery swapping cabinets are collected, including battery voltage, temperature, charge and discharge cycles, battery capacity, charging rate, etc., as well as environmental data and rider usage behavior data. After preprocessing these data, the principal component analysis (PCA) method is used to extract the features related to battery life prediction. Then, based on these features, a variational autoencoder model is constructed. After training the model, the model is used to predict the remaining life and health status of the battery.

[0143] During the training process, an improved butterfly optimization algorithm is used to optimize the hyperparameters of the model to ensure obtaining the optimal model configuration. The optimized model can output the predicted values of the future health status and remaining life of the batteries in each rider battery swapping cabinet. Based on these prediction results, the scheduling system dynamically adjusts the charge and discharge strategies and usage cycles of the batteries to ensure the optimal use of the batteries.

[0144] When the remaining life of the battery is lower than the set threshold, the system will automatically trigger a replacement prompt and adjust the usage strategy of the battery; when the health status of the battery is poor, the system will give priority to charging or replacing the battery to avoid excessive battery wear. The scheduling system will also fine-tune the prediction results according to the real-time feedback data of the battery (such as temperature and load, etc.) to further improve the accuracy of the prediction.

[0145] To verify the effectiveness of the method of the present invention, a field test was conducted on the battery management system of shared bicycles in a certain city for three months. The test location was the core business district of the city. The test period was from June 1, 2024, to August 31, 2024. During the test, we conducted a comparative test on the batteries using traditional battery management methods and the method of the present invention, mainly comparing the remaining life, health status, and replacement cycle of the batteries.

[0146] Table 1 Comparison of Battery Life between Traditional Method and the Method of the Present Invention

[0147]

[0148] As can be seen from the data in Table 1, compared with the traditional method, the batteries using the method of the present invention generally show a more significant improvement in the remaining life. For example, for the battery numbered E001, the remaining life under the traditional method is 45 days, while under the method of the present invention, the remaining life is 62 days, with an improvement rate of 37.8%; similarly, other batteries also show different degrees of life improvement, and the average improvement rate is 37.1%.

[0149] This difference indicates that by adopting a deep learning model based on variational autoencoder and an improved butterfly optimization algorithm, the present invention can more accurately predict the remaining life of the battery. The traditional method usually adopts fixed charge and discharge cycles and static management strategies, while the present invention realizes the refinement and intelligence of battery management by dynamically adjusting and predicting the battery health state. When the battery health state is poor, it can timely predict and adopt appropriate charge and discharge strategies to extend the battery service life.

[0150] For example, the remaining life of the battery numbered E005 under the traditional method is 50 days, while after adopting the method of the present invention, the remaining life is increased to 68 days, with an improvement rate of 36.0%. This result shows that the method of the present invention has a significant effect in extending the battery service life. Especially for batteries with poor health states, the method of the present invention can effectively reduce unnecessary losses and improve the effective use time of the battery.

[0151] Generally speaking, the present invention significantly extends the battery service life by accurately predicting the remaining life of the battery, optimizing the charge and discharge strategies and usage cycles, and has strong advantages compared with the traditional method. This improvement not only helps to improve the battery use efficiency, but also can reduce the cost of frequent battery replacement, providing strong support for the sustainable development of the electric transportation industry.

[0152] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered within the protection scope of the present invention.

Claims

1. A method for predicting the battery life and dynamic scheduling of a rider's battery swapping cabinet based on deep learning, characterized in that, It includes the following steps: S1. Collect multi-source time-series data of the batteries in the rider's battery replacement cabinet, and preprocess the multi-source time-series data; S2. Build a variational autoencoder model, train the variational autoencoder model using the preprocessed multi-source time-series data, and map the multi-source time-series data to the latent space through the encoder; S3. Reconstruct the historical usage data and environmental data in the latent space into the future state data of the batteries in the rider's battery replacement cabinet through the decoder, and output the remaining life and health state distribution of the batteries in the rider's battery replacement cabinet in the future; S4. Introduce an improved butterfly optimization algorithm to optimize the hyperparameters of the variational autoencoder model, use the optimized variational autoencoder model to predict the life of the batteries in the rider's battery replacement cabinet, and output the remaining life and health state distribution of the batteries in each rider's battery replacement cabinet after optimization; S5. Generate a scheduling strategy based on the battery life prediction results of the rider's battery replacement cabinet, and optimize the usage cycle, charging and discharging strategies, and replacement time of the batteries in the rider's battery replacement cabinet; S6. Apply the scheduling strategy to dynamically schedule the batteries in the rider's battery replacement cabinet, and regularly update the variational autoencoder model and re-optimize the hyperparameters according to the real-time data feedback.

2. The method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning according to claim 1, wherein The multi-source time-series data specifically includes battery voltage, temperature, charge and discharge cycle times, battery capacity, charging rate, environmental data, and rider usage behavior data, which are used to analyze the health state of the battery, life prediction, and optimize the battery scheduling strategy.

3. The method for predicting the battery life and dynamic scheduling of the rider battery replacement cabinet based on deep learning according to claim 1, wherein The preprocessing of the multi-source time-series data specifically includes removing invalid data, removing outliers, and performing data standardization processing, which is used to improve the effect of battery life prediction and dynamic scheduling.

4. The method for predicting battery life and dynamic scheduling of a rider's battery swapping cabinet based on deep learning according to claim 1, wherein The specific content of S2 includes: S21. Perform feature selection on the preprocessed multi-source time-series data, and identify the features most relevant to the battery life prediction in the rider's battery replacement cabinet through the principal component analysis method, including battery voltage, temperature, charge and discharge cycle times, battery capacity, charging rate, environmental data, and rider usage behavior data; S22. Build a variational autoencoder model according to the identified features, define the number of layers of the encoder and decoder and the number of neurons in each layer, and determine the dimension of the latent space; S23. Conduct preliminary training of the variational autoencoder model using the preprocessed multi-source time-series data, map the input multi-source time-series data to the latent space through the encoder, and the decoder reconstructs the input data through the output of the latent space; S24. Train the variational autoencoder model using the backpropagation algorithm, and optimize the weight parameters of the variational autoencoder model by minimizing the reconstruction error and KL divergence; S25. Evaluate the training effect of the variational autoencoder model, adopt cross-validation, analyze the performance of the variational autoencoder model on the training set, and if there are problems such as large training errors or overfitting, adjust the structures of the encoder and decoder; S26. After the training is completed, use the variational autoencoder model to process new multi-source time-series data, map the data to the latent space through the encoder, and generate prediction data for the future state of the batteries in the rider's battery replacement cabinet.

5. The method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning according to claim 1, wherein The specific content of S3 includes: S31. Input the preprocessed multi-source time-series data into the trained variational autoencoder model, including the historical usage data and environmental data of the batteries in the rider's battery replacement cabinet. S32. Map the input multi-source time-series data to the latent space through an encoder to generate a latent variable \(z\in\mathbb{R}\) d , where \(\mathbb{R}\) represents the set of real numbers and \(d\) represents the dimension of the latent space; S33. Decode the latent variable z through the decoder into the future state data of the batteries in the rider's battery replacement cabinet, and obtain the future health status and life prediction results of the batteries in the rider's battery replacement cabinet. S34. Utilize the future state data of the batteries in the rider's battery replacement cabinet output by the decoder to further calculate the performance change of the batteries in the rider's battery replacement cabinet under different usage conditions in the future, and generate the future behavior trajectory of the batteries in the rider's battery replacement cabinet. S35. Based on the predicted future health status value and remaining life prediction value of the batteries in the rider's battery replacement cabinet, generate a prediction error, and adjust the mapping of the latent space according to the prediction error.

6. The method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning according to claim 5, characterized in that, The latent variable specifically includes the mean, variance, and vector of the latent space, which are used to capture the implicit features of the data.

7. The method for predicting battery life and dynamic scheduling of a rider's battery swapping cabinet based on deep learning according to claim 1, wherein The S4 specifically includes: S41. When initializing the butterfly optimization algorithm, set the positions and velocities of butterfly individuals, which represent the solutions of the hyperparameters of the variational autoencoder model. The position \(x\) of each butterfly individual i =\(\{x\) i1 , x i2 , \(\cdots\), x id \} is a hyperparameter vector, where each hyperparameter \(x\) ij represents a hyperparameter in the variational autoencoder model. According to the dynamic changes of the predicted future health state \(h\) pred and the predicted remaining life \(t\) pred of the batteries in the rider's battery swapping cabinet, adjust the acceleration coefficients \(\beta1\) and \(\beta2\): Among them, β1 represents the acceleration coefficient that affects the individual butterfly's movement towards the optimal position, β2 represents the acceleration coefficient that the individual butterfly moves towards the global optimal position, and h max is the maximum value of the health state of the battery in the rider's battery replacement cabinet, and t max is the predicted remaining life of the battery in the rider's battery replacement cabinet, and are the acceleration coefficients at the current iteration time t, and represent the updated values of the acceleration coefficients; S42. The flight direction of the butterfly is adjusted according to the current position, global optimal solution, and local optimal solution. Among them, is the speed of the butterfly, g i is the position of the global optimal solution, is the position vector of the i-th butterfly individual in the (t + 1)-th generation, the position vector of the i-th butterfly individual in the t-th generation, min is the minimum operation, considering the predicted value h of the future health state of the battery in the rider's battery replacement cabinet pred and the actual usage load L load , an adaptive distance limit mechanism is introduced to control the maximum distance d between butterfly individuals max , and the adaptive distance limit mechanism can dynamically adjust the size of the search space according to the current usage state of the battery in the rider's battery replacement cabinet: where d initial is the initially set maximum distance, L max is the maximum value of the battery load in the rider's battery replacement cabinet, and α is the adjustment coefficient of the load factor; S43. The speed of each butterfly individual is updated according to the current position, historical best position, and global optimal solution position. where ω is the inertia weight, c1 and c2 are learning factors, rand1 and rand2 are random numbers, and p i is the historical best position of the i-th butterfly, and g i is the global best position. is the velocity vector of the i-th butterfly individual at the (t + 1)-th generation, is the velocity vector of the i-th butterfly individual at the t-th generation. In the case of large prediction errors in the battery health state and remaining life in the rider battery replacement cabinet, increase the learning factors c1 and c2 to accelerate global search, and when the prediction error is small, decrease the learning factors; S44. Evaluate the fitness of each butterfly individual, and calculate the fitness of the individual according to the prediction result of the variational autoencoder model. Adopt a composite fitness function, and simultaneously consider the prediction errors of the health status and remaining life of the batteries in the rider's battery replacement cabinet. where, f(x i ) is the fitness value of the i-th butterfly individual, n is the number of samples, h true is the true health state of the battery in the rider's battery replacement cabinet, t true is the true remaining life of the battery in the rider's battery replacement cabinet, w1 and w2 are weights. If the prediction error of the health state of the battery in the rider's battery replacement cabinet is large, then increase the weight of w1; if the prediction error of the remaining life is large, then increase the weight of w2, and flexibly adjust the weights at different stages; S45. Through multiple iterative optimizations, the butterfly optimization algorithm converges to the optimal hyperparameters The optimal hyperparameters will be used to train the variational autoencoder model; S46. Use the optimized variational autoencoder model to predict the remaining life and health status of the batteries in the rider's battery replacement cabinet. The variational autoencoder outputs the future state data of the batteries in the rider's battery replacement cabinet based on the historical usage data and environmental data of the batteries in the rider's battery replacement cabinet, and obtains the predicted value of the remaining life of the batteries in each rider's battery replacement cabinet after optimization. and the predicted value of the health status S47. On the basis of the prediction result, combined with the current usage load and temperature environment factors of the batteries in the rider's battery replacement cabinet, further fine-tune the prediction results of the remaining life and health status. If the batteries in the rider's battery replacement cabinet are used under high load, appropriately adjust the remaining life according to the load situation; if the temperature of the batteries in the rider's battery replacement cabinet is too high, correct the health status.

8. The method for predicting the battery life and dynamic scheduling of the rider battery replacement cabinet based on deep learning according to claim 7, characterized in that, The hyperparameters specifically include the dimension of the latent space, learning rate, number of layers of the encoder and decoder, which are used to optimize the structure and training process of the variational autoencoder model.

9. The method for predicting the battery life and dynamic scheduling of the rider's battery swapping cabinet based on deep learning according to claim 1, wherein The S5 specifically includes: S51. Calculate the usage cycle, charging strategy, and discharging strategy of the batteries in each rider's battery replacement cabinet based on the predicted remaining battery life and predicted health status values of the optimized rider's battery replacement cabinet output by the variational autoencoder model, and determine the optimal replacement time of the batteries in the rider's battery replacement cabinet according to the predicted remaining battery life value of the batteries in the rider's battery replacement cabinet; and the predicted health status value Calculate the usage cycle, charging strategy, and discharging strategy of the batteries in each rider's battery replacement cabinet, and determine the optimal replacement time of the batteries in the rider's battery replacement cabinet according to the predicted remaining battery life value of the batteries in the rider's battery replacement cabinet; S52. Based on the optimized state of health prediction value of the batteries in the rider's battery swapping cabinet, evaluate the current health status of the batteries in the rider's battery swapping cabinet, and by setting the state of health threshold h thresh classify the batteries in the rider's battery swapping cabinet and generate scheduling strategies for different health status categories: S53. According to the predicted remaining life value after optimization of the battery in the rider's battery replacement cabinet When the remaining life of the battery in the rider's battery replacement cabinet is less than the set threshold t thresh a battery replacement prompt in the rider's battery replacement cabinet is triggered: S54. According to the predicted remaining life value and health status prediction value, through the scheduling strategy optimization algorithm, optimize the charging and discharging strategies of the batteries in the rider's battery replacement cabinet and adjust the usage cycle of the batteries in the rider's battery replacement cabinet. S55. According to the optimized health status and remaining life prediction values of the batteries in the rider's battery replacement cabinet, combined with the historical usage data and environmental data, generate the specific replacement time of each battery in the rider's battery replacement cabinet through the optimized scheduling strategy, and regularly update the scheduling strategy to cope with the real-time changing battery data and environmental conditions in the rider's battery replacement cabinet. S56. Update the scheduling strategy in real time, and dynamically adjust the usage cycle, charging and discharging strategies, and replacement time of the batteries in the rider's battery replacement cabinet according to the real-time feedback data of the batteries in the rider's battery replacement cabinet.

Images