Storage battery defect monitoring identification and service life prediction method

Abstract

The invention discloses a storage battery defect monitoring identification and life prediction method, and relates to the technical field of battery management, and the method comprises the steps: obtaining a storage battery operation monitoring index set, randomly configuring weights, and constructing an initial weight particle swarm; traversing the initial weight particle swarm, training the storage battery life prediction model, and obtaining the verification accuracy of the initial weight particle swarm; when all the accuracy rates are smaller than or equal to the convergence accuracy rate, the initial weight particle swarm verification accuracy rate is used as a fitness function, particle optimization is carried out, an updated particle set is obtained, and a storage battery life prediction model is trained; and when any particle of the initial weight particle swarm verification accuracy is greater than the convergence accuracy, outputting a corresponding storage battery life prediction model, and executing storage battery defect monitoring identification and life prediction tasks. The technical problem that the operation state and the residual service life of the storage battery cannot be accurately evaluated in the prior art is solved, and the technical effects of storage battery defect monitoring and residual service life accurate prediction are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and in particular to a method for monitoring and identifying battery defects and predicting battery life. Background Art

[0002] With the widespread application of energy storage technology, the application scale of batteries in new energy power generation, communication base stations and data centers continues to expand. However, batteries may be affected by many factors during long-term operation, such as charge and discharge frequency, operating environment temperature, internal resistance changes, etc., resulting in performance degradation and shortened life. Therefore, how to accurately monitor the operating status of batteries and predict their remaining service life has become an important technical requirement to ensure the safe and stable operation of batteries.

[0003] Existing technologies usually use a single parameter (such as voltage or internal resistance) to evaluate the battery status, and fail to fully consider the comprehensive impact of multi-dimensional operating parameters on life prediction. This method not only has insufficient monitoring accuracy, but also fails to fully reflect the health status of the battery. In addition, some methods rely on fixed models when predicting life, and lack the ability to dynamically adapt to changes in the battery's operating status, resulting in a large deviation between the life prediction results and the actual situation, making it difficult to meet the needs of complex application scenarios. Summary of the invention

[0004] The present application provides a battery defect monitoring and identification and life prediction method, which is used to solve the technical problem that the prior art cannot accurately evaluate the battery operating status and remaining service life.

[0005] In view of the above problems, the present application provides a method for monitoring and identifying battery defects and predicting battery life.

[0006] The present application provides a method for monitoring and identifying battery defects and predicting battery life, the method comprising: Obtain a battery operation monitoring indicator set; randomly configure weights for the battery operation monitoring indicator set to construct an initial weight particle group, wherein any particle of the initial weight particle group represents a set of battery operation monitoring indicator weights, and the set of battery operation monitoring indicator weights corresponds one-to-one to the battery operation monitoring indicator set; traverse the initial weight particle group, train a battery life prediction model based on the battery operation monitoring indicator set, and obtain the initial weight particle group verification accuracy, wherein each particle of the initial weight particle group corresponds one-to-one to the initial weight particle group verification accuracy; when the initial weight particle group verification accuracy is less than or equal to the convergence accuracy, the initial weight particle group is optimized with the initial weight particle group verification accuracy as the fitness function to obtain an updated particle set; according to the updated particle set, the battery life prediction model is trained based on the battery operation monitoring indicator set; when any particle of the initial weight particle group verification accuracy is greater than the convergence accuracy, the battery life prediction model of the corresponding particle is output, and the battery defect monitoring identification and life prediction tasks are performed.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application obtains a battery operation monitoring indicator set; randomly configures weights for the battery operation monitoring indicator set to construct an initial weight particle group, wherein any particle of the initial weight particle group represents a set of battery operation monitoring indicator weights, and the set of battery operation monitoring indicator weights corresponds one-to-one to the battery operation monitoring indicator set; traverses the initial weight particle group, trains a battery life prediction model based on the battery operation monitoring indicator set, and obtains the initial weight particle group verification accuracy, wherein each particle of the initial weight particle group corresponds one-to-one to the initial weight particle group verification accuracy; when the initial weight particle group verification accuracy is less than or equal to the convergence accuracy, the initial weight particle group is optimized with the initial weight particle group verification accuracy as the fitness function to obtain an updated particle set; according to the updated particle set, the battery life prediction model is trained based on the battery operation monitoring indicator set; when any particle of the initial weight particle group verification accuracy is greater than the convergence accuracy, the battery life prediction model of the corresponding particle is output, and the battery defect monitoring identification and life prediction tasks are performed. The present invention solves the technical problem that the prior art cannot accurately evaluate the operating status and remaining service life of the battery. The sensor module collects battery operation monitoring indicators, constructs an initial weighted particle group and optimizes the weight configuration, uses a neural network to train a life prediction model, evaluates the particle effect by verifying the accuracy, and outputs the best model after iterative optimization, thereby achieving the technical effect of battery defect monitoring and accurate prediction of remaining service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] 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.

[0009] Figure 1 A schematic diagram of a battery defect monitoring, identification and life prediction method provided in an embodiment of the present application; Figure 2 A schematic diagram of a process for obtaining the initial weighted particle swarm verification accuracy in the battery defect monitoring, identification and life prediction method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The present application provides a battery defect monitoring and identification and life prediction method, which is used to solve the technical problem that the existing technology cannot accurately evaluate the operating status and remaining service life of the battery. The battery operation monitoring indicators are collected through the sensor module, an initial weighted particle group is constructed and the weight configuration is optimized, and a life prediction model is trained using a neural network. The particle effect is evaluated by verifying the accuracy rate, and the best model is output after iterative optimization, thereby achieving the technical effect of battery defect monitoring and accurate prediction of remaining life.

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0012] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0013] Examples, such as Figure 1 As shown, the present application provides a method for monitoring and identifying battery defects and predicting battery life, the method comprising: Step S100: Obtain a battery operation monitoring indicator set.

[0014] In the embodiment of the present application, data is first collected through sensor modules installed in the battery pack and its operating environment. The sensor module includes a voltage sensor, an internal resistance sensor, a temperature sensor, a charge and discharge current sensor, etc., which monitor the key physical parameters of the battery operation in real time. The battery operation monitoring index refers to a specific set of parameters collected by these sensors, including but not limited to battery voltage, internal resistance value, operating temperature, charge and discharge current, and discharge depth.

[0015] The collected data is stored and processed to form a historical database. Finally, by extracting these stored historical data, a complete set of battery operation monitoring indicators is obtained.

[0016] Step S200: randomly configuring weights for the battery operation monitoring indicator set to construct an initial weight particle group, wherein any particle of the initial weight particle group represents a group of battery operation monitoring indicator weights, and the group of battery operation monitoring indicator weights corresponds one-to-one to the battery operation monitoring indicator set.

[0017] In the embodiment of the present application, for each indicator in the battery operation monitoring indicator set (such as voltage, internal resistance, temperature, charge and discharge current), a random proportion generation method is used to assign a weight value to each indicator, and ensure that these weight values ​​meet the constraint that the sum is 1. In this way, a set of weight values ​​is formed, each weight value corresponds to a monitoring indicator in the set, indicating the importance of the indicator in the subsequent life prediction task.

[0018] Each group of weight values ​​assigned in this way constitutes a particle. For example, assuming that the battery operation monitoring indicator set includes four indicators: voltage, internal resistance, temperature and current, then a particle may be represented as 0.3, 0.4, 0.2, 0.1, where 0.3 represents the weight of voltage, 0.4 represents the weight of internal resistance, and 0.2 and 0.1 correspond to the weights of temperature and current respectively. The significance of particle representation is that it fully reflects the importance distribution of each indicator in the battery operation monitoring indicator set in the form of weight.

[0019] By repeating the above steps, multiple particles are generated to construct an initial weighted particle swarm. Each particle in the particle swarm represents a complete set of monitoring index weight configurations and corresponds one-to-one to each index in the battery operation monitoring index set.

[0020] In this particle swarm, each particle corresponds to a complete set of monitoring indicator weight configurations. These weight values ​​strictly satisfy a one-to-one correspondence, which not only reflects the relative importance of each monitoring indicator in the set, but also provides a variety of initial weight configuration schemes for subsequent model training and optimization.

[0021] Step S300: traverse the initial weighted particle swarm, train the battery life prediction model based on the battery operation monitoring indicator set, and obtain the initial weighted particle swarm verification accuracy, wherein each particle of the initial weighted particle swarm corresponds one-to-one to the initial weighted particle swarm verification accuracy.

[0022] In an embodiment of the present application, each particle in the initial weighted particle group is extracted in turn, wherein each particle contains a set of weight values ​​corresponding one to one with a set of battery operation monitoring indicators. Based on the extracted particle weight set, the neural network topology of the battery life prediction model is configured, wherein the input node corresponds to each indicator of the monitoring indicator set, and the output node is used to predict the battery life, and the life prediction model after weight configuration is obtained. Subsequently, the data of the battery operation monitoring indicator set is used as input features, and supervised learning is performed in combination with the life identification data. The model is trained to fit the actual life data, and the trained life prediction model corresponding to the current particle is obtained. Among them, the battery operation monitoring indicator set and the corresponding life identification data are collected and obtained from the historical database, and the life identification data indicates the remaining usable time of the battery in the current state.

[0023] Next, an independent validation data set is used to evaluate the performance of the model and calculate its validation accuracy to reflect the prediction effect of the current particle weight configuration. Repeat the above steps for all particles in the particle swarm, and finally obtain a validation accuracy set containing all particles, that is, the validation accuracy of the initial weighted particle swarm. This validation accuracy set achieves a one-to-one correspondence between particles and their weight configuration effects.

[0024] Further, such as Figure 2 As shown, in the method provided in the embodiment of the application, the initial weighted particle swarm is traversed, and the battery life prediction model is trained based on the battery operation monitoring indicator set to obtain the initial weighted particle swarm verification accuracy, and further includes: According to the initial weighted particle group, a first particle is extracted, wherein the first particle has a first battery operation monitoring indicator weight set; a battery life prediction model topology is configured, wherein the battery life prediction model topology is a neural network topology, an input node of the neural network topology corresponds one-to-one to each node of the battery operation monitoring indicator set, and the neural network topology has one input node for outputting predicted life; according to the first battery operation monitoring indicator weight set, the input nodes of the battery life prediction model topology are weighted to obtain an initial battery life prediction model; according to the battery operation monitoring indicator set, the initial battery life prediction model is trained to obtain a first particle battery life prediction model; the first particle battery life prediction model is verified to obtain a verification accuracy rate and add it to the initial weighted particle group training accuracy rate.

[0025] In the embodiment of the present application, the first particle is first extracted from the initial weight particle group, and the first particle contains the first battery operation monitoring indicator weight set. Each weight value in the weight set corresponds to each indicator (such as voltage, internal resistance, temperature, current, etc.) in the battery operation monitoring indicator set, indicating the relative importance of each monitoring indicator in the prediction task. The first particle is extracted in sequence by the list index method to obtain the first battery operation monitoring indicator weight set.

[0026] Then, according to the weight set of the first battery operation monitoring indicator, the battery life prediction model topology is configured. The model topology is a neural network structure, in which the number of input nodes is consistent with the number of indicators in the battery operation monitoring indicator set, each input node corresponds to a monitoring indicator, and the output node is unique, which is used to predict the service life of the battery. The neural network topology is constructed through a multi-layer perceptron (MLP), and the input layer nodes, hidden layer structure and output nodes are set to obtain a complete battery life prediction model topology.

[0027] Then, according to the weight set of the first battery operation monitoring index, the input nodes of the battery life prediction model topology are weighted. The weight configuration adopts a linear weighting method to multiply each weight value of the weight set in the first particle by the input feature corresponding to the monitoring index. For example, if the weight set of the first battery operation monitoring index is 0.3, 0.4, 0.2, 0.1, the input feature voltage, internal resistance, temperature and current are weighted to 0.3, 0.4, 0.2 and 0.1 respectively. After completing the weight configuration, the initial battery life prediction model based on the first battery operation monitoring index weight set is obtained.

[0028] Next, the initial battery life prediction model is trained using the data in the battery operation monitoring indicator set (such as voltage, internal resistance, temperature and current data in the form of time series) and the corresponding life identification data (such as actual service life). The training process uses the mean square error (MSE) as the loss function, and optimizes the model parameters through the batch gradient descent algorithm to minimize the gap between the predicted life and the actual life. After the training is completed, the first particle battery life prediction model based on the first battery operation monitoring indicator weight set is obtained.

[0029] The first particle battery life prediction model is then verified using an independent verification data set, where the independent verification data set is prepared in advance. The verification data set is input into the model, the error between the predicted value and the actual value is calculated, and the verification accuracy is calculated based on the error result. The verification accuracy quantifies the adaptability of the first battery operation monitoring index weight set in the life prediction task. After the verification is completed, the verification accuracy of the first particle battery life prediction model is obtained.

[0030] Finally, the verification accuracy of the first particle battery life prediction model is recorded and added to the verification accuracy set of the initial weighted particle swarm.

[0031] Furthermore, in the method provided in the embodiment of the application, the initial battery life prediction model is trained according to the battery operation monitoring indicator set to obtain the first particle battery life prediction model, and further includes: The battery operation monitoring indicator set is used as input data, and the battery life data is used as output data, and a battery operation monitoring indicator characteristic value set and battery life identification data are collected; the battery life identification data is used as supervision, and the battery operation monitoring indicator characteristic value set is used as input to train a first-level first-particle battery life prediction model; when the data amount of the first-level loss data set that does not meet the first accuracy threshold of the first-level first-particle battery life prediction model is less than the data amount threshold, the first-level first-particle battery life prediction model is set to the first-particle battery life prediction model.

[0032] In the embodiment of the present application, firstly, the battery operation monitoring index set is used as input data, and the set includes characteristic values ​​of monitoring indicators such as voltage, internal resistance, temperature, and charge and discharge current collected from sensors; at the same time, the battery life data is used as output data, and the data represents the actual battery life value corresponding to each set of monitoring indicators (for example, represented by the number of cycles or working hours). Through the integration of historical data, the battery operation monitoring index characteristic value set and the corresponding battery life identification data are collected and constructed.

[0033] Then, supervised learning method was used to train the first-level first-particle battery life prediction model, with battery life identification data as the supervision signal and the battery operation monitoring index feature value set as the input feature. The topology of the model is based on a neural network, where the number of nodes in the input layer corresponds to the number of features in the monitoring index set, and the output layer is a node used to predict the battery life value. The mean square error (MSE) was used as the loss function during the training process, and the model parameters were adjusted by the gradient descent optimization algorithm to minimize the gap between the predicted life and the actual life. After training, the first-level first-particle battery life prediction model was obtained.

[0034] The first-level first-particle battery life prediction model is then verified, and the prediction accuracy of the model is calculated using an independent verification data set. The size of the first-level loss data set is counted, that is, the number of data points in the verification data set where the model prediction error exceeds the allowable range. The amount of loss data is compared with the preset data volume threshold. If the amount of data in the first-level loss data set is less than the preset data volume threshold, it means that the prediction performance of the model has met the requirements in most cases, and the error points only occupy a small range, and the overall performance of the model is considered acceptable. In this case, the first-level first-particle battery life prediction model is set as the final first-particle battery life prediction model.

[0035] Furthermore, the method provided in the application embodiment also includes: When the data volume of the first-level loss data set for which the first-level first-particle battery life prediction model does not meet the first accuracy threshold is greater than or equal to the data volume threshold, extract the life deviation vector set of the first-level first-particle battery life prediction model; perform system error analysis on the life deviation vector set to obtain a system error vector; use the opposite of the system error vector as supervision data and the battery operation monitoring index characteristic value set as input to train the second-level first-particle battery life prediction model; until the data volume of the N-level loss data set for which the N-level first-particle battery life prediction model does not meet the first accuracy threshold is less than the data volume threshold, add and fully connect the outputs of the first-level first-particle battery life prediction model, the second-level first-particle battery life prediction model, and the N-level first-particle battery life prediction model to obtain the first particle battery life prediction model.

[0036] In the embodiment of the present application, a life deviation vector set is first extracted from the validation data set. The difference between the model prediction value and the actual life identification data is calculated by traversing the validation data set point by point, and a deviation vector set is generated. For example, if the model predicts a life of 10 years and the actual life is 9 years, the deviation is 1. After the extraction is completed, a life deviation vector set reflecting the model error distribution is obtained.

[0037] Then, the extracted life deviation vector set is subjected to systematic error analysis. By calculating the mean of the deviation vector, the central tendency and systematic characteristics of the model error are identified. For example, if the mean is 0.2, −0.1, or 0.15, it indicates that the model's prediction results on certain monitoring indicators are generally too high or too low. After the analysis is completed, a systematic error vector describing the direction and magnitude of the model error is obtained.

[0038] Then, the system error vector is negated to generate supervisory data for correcting the model. The supervisory data reflects the direction and magnitude that need to be adjusted. For example, if the system error vector is 0.2, −0.1, 0.15, the negation is −0.2, 0.1, −0.15, which is used to guide the model to correct the error. Through this step, the negation of the system error vector for optimization is obtained.

[0039] Next, the negative number of the system error vector is used as the supervision data, and the characteristic value set of the battery operation monitoring index is used as the input to train the secondary first-particle battery life prediction model. Through the supervised learning method, the mean square error (MSE) is used as the loss function to optimize the model parameters to minimize the error between the output and the negative number. After the training is completed, the secondary first-particle battery life prediction model that corrects the system error of the primary model is obtained.

[0040] If the secondary first-particle battery life prediction model still does not meet the accuracy threshold, and the amount of data in the loss data set is greater than or equal to the data volume threshold, repeat the above steps. Extract a new set of deviation vectors, perform system error analysis, generate the opposite of a new system error vector, and continue to train the tertiary first-particle battery life prediction model. Through multiple iterative optimizations, the system error is continuously corrected until the amount of data in the loss data set of a certain level of model is less than the data volume threshold.

[0041] Finally, the outputs of all levels of models (such as level 1, level 2, and up to level N) are added and integrated through the fully connected layer. The fully connected layer combines the correction results of each level of models into the final prediction value through linear combination. After the integration is completed, a fully optimized first-particle battery life prediction model is obtained.

[0042] Step S400: When the verification accuracy of the initial weighted particle swarm is less than or equal to the convergence accuracy, the verification accuracy of the initial weighted particle swarm is used as the fitness function, and the initial weighted particle swarm is subjected to particle optimization to obtain an updated particle set.

[0043] In an embodiment of the present application, the verification accuracy of the initial weighted particle swarm is compared with the preset convergence accuracy. When the verification accuracy of the initial weighted particle swarm is less than or equal to the convergence accuracy, particle optimization is performed on the initial weighted particle swarm to obtain an updated particle set.

[0044] Specifically, firstly, the distances between particles of the initial weighted particle group are evaluated pairwise, and the distance of each pair of particles is calculated to form a set of particle distance evaluation values. Then, based on the particle distance threshold, the particles are clustered according to the distance relationship to obtain multiple groups of initial weighted particles. Then, according to the verification accuracy of each group of particles, they are sorted from high to low to generate multiple groups of initial weighted particle sorting results. Finally, based on the sorting results, the particle positions are updated using the collaborative optimization strategy to generate an updated particle set.

[0045] Further, in the method provided in the embodiment of the application, when the verification accuracy of the initial weighted particle swarm is less than or equal to the convergence accuracy, the verification accuracy of the initial weighted particle swarm is used as the fitness function, the initial weighted particle swarm is subjected to particle optimization to obtain an updated particle set, and further includes: The particle distances of the initial weighted particle swarm are evaluated in pairs to obtain a set of particle distance evaluation values; clustering is performed based on the particle distance threshold according to the set of particle distance evaluation values ​​to obtain multiple groups of initial weighted particles; according to the verification accuracy of the initial weighted particle swarm, the multiple groups of initial weighted particles are sorted from large to small according to the verification accuracy to obtain multiple groups of initial weighted particle sorting results; according to the sorting results of the multiple groups of initial weighted particles, collaborative particle optimization is performed on the initial weighted particle swarm to obtain an updated particle set.

[0046] In the embodiment of the present application, the initial weighted particle group is first evaluated for the particle distance between two particles. Two particles in the particle group are randomly extracted, and their first battery operation monitoring index weight set and second battery operation monitoring index weight set are obtained respectively, and then the Euclidean distance between the two particle weight sets is calculated. The calculation result is used as the particle distance evaluation value and added to the particle distance evaluation value set. By comparing all particle combinations in pairs, a complete particle distance evaluation value set is finally obtained.

[0047] Then, according to the set of particle distance evaluation values, the particle group is clustered based on the preset particle distance threshold, and particles with closer distances are grouped together to form multiple groups of initial weighted particles. For example, if the distance threshold is 0.1, all particles with a distance less than 0.1 are grouped together. After clustering is completed, multiple groups of initial weighted particles are obtained.

[0048] Then, according to the verification accuracy of the initial weighted particle group, each group of initial weighted particles is sorted from high to low according to the verification accuracy, generating multiple groups of initial weighted particle sorting results. This sorting result identifies the particles with better performance in each group, providing a reference for guiding particle optimization. For example, a group may be sorted as particle 1, particle 2, particle 3, among which particle 1 has the highest verification accuracy.

[0049] Then, according to the results of the particle sorting of multiple groups of initial weights, the particle swarm is subjected to collaborative particle optimization. Two groups of particle sorting results are randomly extracted, and the first particles of the first and second groups are obtained respectively, and the first particles are used to guide the last particles of other groups to mutate. For example, the first particle of the first group guides the last particle of the second group, and the particle position is updated by reducing the particle distance evaluation value between the two to generate the first particle update result. Similarly, the first particle of the second group guides the last particle of the first group to generate the second particle update result. All update results are added to the update particle set.

[0050] Finally, after the above steps, the optimization of the particle swarm is completed and the updated particle set is obtained.

[0051] Furthermore, in the method provided in the embodiment of the application, the distance between particles of the initial weighted particle group is evaluated pairwise to obtain a set of particle distance evaluation values, and the method further includes: A first particle and a second particle of the initial weighted particle group are randomly obtained, wherein the first particle has a first battery operation monitoring indicator weight set, and the second particle has a second battery operation monitoring indicator weight set; the Euclidean distance between the first battery operation monitoring indicator weight set and the second battery operation monitoring indicator weight set is calculated, set as a first particle distance evaluation value, and added to the particle distance evaluation value set.

[0052] In the embodiment of the present application, first, two particles are randomly extracted from the initial weighted particle group, which are called the first particle and the second particle. Each particle contains a set of weight values ​​corresponding to the battery operation monitoring index set, and these weight values ​​respectively reflect the importance of each monitoring index in the model.

[0053] Then, the weight sets of the two particles are compared using the Euclidean distance method. Euclidean distance is a common method for measuring the distance between points in multidimensional space. It can effectively represent the difference in weight distribution between two particles. During the calculation, the differences in the corresponding weight values ​​of each monitoring indicator in the first particle and the second particle are compared one by one. These differences are combined to obtain the distance between the particles. This distance is called the first particle distance evaluation value, which indicates the similarity of the two particles in weight configuration.

[0054] Finally, the calculated first particle distance evaluation value is recorded and added to the particle distance evaluation value set.

[0055] Furthermore, in the method provided in the embodiment of the application, according to the sorting results of the multiple groups of initial weighted particles, the initial weighted particle group is subjected to collaborative particle optimization to obtain an updated particle set, and further includes: Randomly extract the first group of initial weighted particle sorting results and the second group of initial weighted particle sorting results of the multiple groups of initial weighted particle sorting results; obtain the first initial weighted particle of the first group of initial weighted particle sorting results, guide the initial weighted particles at the end of a preset number of the second group of initial weighted particle sorting results to mutate, reduce the particle distance evaluation value with the first initial weighted particle, and obtain a first particle update result; obtain the second initial weighted particle of the second group of initial weighted particle sorting results, guide the initial weighted particles at the end of a preset number of the first group of initial weighted particle sorting results to mutate, reduce the particle distance evaluation value with the second initial weighted particle, and obtain a second particle update result; add the first particle update result and the second particle update result into the update particle set.

[0056] In the embodiment of the present application, two groups of the multiple groups of initial weighted particle sorting results are first randomly extracted, which are respectively referred to as the first group of initial weighted particle sorting results and the second group of initial weighted particle sorting results. The first particle, i.e., the first initial weighted particle, is selected from the first group of sorting results as the optimization target.

[0057] Then, the number of dimensions k used for mutation is determined by randomly taking values ​​according to the weight dimension, and the value is used to mutate the preset number of tail particles in the second group of sorting results. The mutation adjusts the partial weight values ​​of these tail particles so that they gradually approach the first initial weight particle, thereby reducing the distance evaluation value between particles. After the mutation is completed, the first particle update result is generated.

[0058] Then, the first particle is selected from the second group of initial weight particle sorting results, that is, the second first initial weight particle, and similar mutation operations are performed on the preset number of last particles in the first group of sorting results. By adjusting the weight value, these particles are gradually brought closer to the second first initial weight particle, thereby optimizing their configuration. After completion, the second particle update result is generated.

[0059] Finally, the updated results of the first particle and the second particle are added to the updated particle set to complete a collaborative particle optimization process.

[0060] Furthermore, in the method provided in the embodiment of the application, the first initial weighted particle of the first group of initial weighted particle sorting results is obtained, and the initial weighted particles at the end of the preset number of the second group of initial weighted particle sorting results are guided to mutate, and the particle distance evaluation value with the first initial weighted particle is reduced to obtain the first particle update result, and the method also includes: Based on the weight dimension, a value is randomly taken to obtain a k value, 1≤k≤weight dimension; according to the k value, the first initial weight particle is taken as the target, the initial weight particles at the end of the preset number are mutated, and the particle distance evaluation value with the first initial weight particle is reduced to obtain the first particle update result.

[0061] In an embodiment of the present application, first, the weight dimension of the particle is determined according to the weight configuration characteristics of the particle, that is, the number of weight values ​​contained in each particle. The weight dimension comes from the number of features of the monitoring indicator set. For example, if the battery operation monitoring indicator set includes voltage, internal resistance, temperature and current, the weight dimension is 4. After determining the weight dimension, a random number generation method (such as a uniformly distributed random number generator) is used to randomly generate an integer value k in the range of 1 to the weight dimension. The value k represents the number of weight dimensions that will be adjusted in the mutation operation. After this process is completed, the number of adjustment dimensions k used for mutation is randomly obtained, which lays the foundation for subsequent operations.

[0062] Then, the first initial weight particle is extracted from the particle sorting result. This particle is the particle with the best performance in the current group, and its weight set (such as 0.3, 0.4, 0.2, 0.1) represents the optimal weight distribution scheme. This particle is selected as the target particle to guide other particles to mutate. Through this operation, the target particle for mutation is identified, providing a benchmark for the optimization of the last particle.

[0063] Then, a preset number of end particles (e.g., 3 particles) are selected from the sorting results and mutated. For each end particle, k dimensions (e.g., the 1st and 3rd dimensions) are randomly selected from its weight set, and then the weight difference between the end particle and the target particle in these dimensions is calculated, and these weight values ​​are gradually adjusted according to the dynamic scaling factor (set to 0.5) to gradually approach the target particle. For example, if the weight set of a certain end particle is 0.2, 0.4, 0.3, 0.1, and the weight set of the target particle is 0.3, 0.4, 0.2, 0.1, the 1st and 3rd dimensions are selected for mutation, and the mutated weight set is adjusted to 0.25, 0.4, 0.25, 0.1. Through this step, the weight of each end particle gradually approaches the target particle in the selected dimension, thereby reducing the distance between particles and optimizing the weight distribution.

[0064] After the mutation is completed, the new weight configurations of all the tail particles form an optimized particle set, which is called the first particle update result.

[0065] Step S500: training a battery life prediction model according to the updated particle set and based on the battery operation monitoring indicator set.

[0066] In an embodiment of the present application, a battery life prediction model is trained based on a battery operation monitoring indicator set according to an updated particle set. Specifically, each indicator feature in the battery operation monitoring indicator set is weighted using the particle weight configuration in the updated particle set to highlight the relative importance of different monitoring indicators in life prediction. Subsequently, a battery life prediction model is constructed, whose input layer corresponds one-to-one to each monitoring indicator in the battery operation monitoring indicator set, and whose output layer is used to predict the service life of the battery. By inputting historical monitoring data and life identification data, the model parameters are optimized using a supervised learning method, and finally a battery life prediction model that can perform accurate life prediction based on the battery operation monitoring indicator set is obtained.

[0067] Step S600: When the verification accuracy of any particle of the initial weighted particle swarm is greater than the convergence accuracy, the battery life prediction model of the corresponding particle is output to perform battery defect monitoring and identification and life prediction tasks.

[0068] In an embodiment of the present application, first, the initial weighted particle group is screened to find particles whose verification accuracy is greater than the convergence accuracy. This process is the same as the calculation of the verification accuracy mentioned above, by applying the weight configuration of each particle to the model and calculating its prediction accuracy on the verification set. After screening, the particles that meet the conditions and their corresponding weight configurations are extracted. Then, the particles with the highest verification accuracy are selected from the particles that meet the conditions, and their optimal weight configurations are extracted. This operation is the same as the process of screening and sorting particles mentioned above, by sorting the verification accuracy from high to low, and selecting the optimal particles and their weight configurations. The final battery life prediction model is then constructed using the screened optimal weight configuration.

[0069] After the model is built, the real-time execution stage of the battery defect monitoring and life prediction task begins. Specifically, the real-time collected battery operation monitoring index data is input into the model, and after weighted processing and model reasoning, the predicted battery life value is output. At the same time, the life prediction results are analyzed in real time in combination with the anomaly detection algorithm. If the predicted value is lower than the preset safety threshold, it is judged as a potential defect and an alarm is triggered, and a corresponding monitoring report is generated.

[0070] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects: The present application obtains a battery operation monitoring indicator set; randomly configures weights for the battery operation monitoring indicator set to construct an initial weight particle group, wherein any particle of the initial weight particle group represents a set of battery operation monitoring indicator weights, and the set of battery operation monitoring indicator weights corresponds one-to-one to the battery operation monitoring indicator set; traverses the initial weight particle group, trains a battery life prediction model based on the battery operation monitoring indicator set, and obtains the initial weight particle group verification accuracy, wherein each particle of the initial weight particle group corresponds one-to-one to the initial weight particle group verification accuracy; when the initial weight particle group verification accuracy is less than or equal to the convergence accuracy, the initial weight particle group is optimized with the initial weight particle group verification accuracy as the fitness function to obtain an updated particle set; according to the updated particle set, the battery life prediction model is trained based on the battery operation monitoring indicator set; when any particle of the initial weight particle group verification accuracy is greater than the convergence accuracy, the battery life prediction model of the corresponding particle is output, and the battery defect monitoring identification and life prediction tasks are performed. The present invention solves the technical problem that the prior art cannot accurately evaluate the operating status and remaining service life of the battery. The sensor module collects battery operation monitoring indicators, constructs an initial weighted particle group and optimizes the weight configuration, uses a neural network to train a life prediction model, evaluates the particle effect by verifying the accuracy, and outputs the best model after iterative optimization, thereby achieving the technical effect of battery defect monitoring and accurate prediction of remaining service life.

[0071] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0073] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for monitoring and identifying battery defects and predicting battery life, characterized in that: include: Obtain a set of battery operation monitoring indicators; Randomly configuring weights for the battery operation monitoring indicator set to construct an initial weight particle group, wherein any particle of the initial weight particle group represents a group of battery operation monitoring indicator weights, and the group of battery operation monitoring indicator weights corresponds one-to-one to the battery operation monitoring indicator set; Traversing the initial weighted particle swarm, training a battery life prediction model based on the battery operation monitoring indicator set, and obtaining a verification accuracy rate of the initial weighted particle swarm, wherein each particle of the initial weighted particle swarm corresponds one-to-one to the verification accuracy rate of the initial weighted particle swarm; When the verification accuracy of the initial weighted particle swarm is less than or equal to the convergence accuracy, the verification accuracy of the initial weighted particle swarm is used as the fitness function, and the initial weighted particle swarm is subjected to particle optimization to obtain an updated particle set; According to the updated particle set, based on the battery operation monitoring indicator set, training a battery life prediction model; When any particle of the initial weighted particle swarm verification accuracy is greater than the convergence accuracy, the battery life prediction model of the corresponding particle is output to perform battery defect monitoring and identification and life prediction tasks.

2. The method according to claim 1, characterized in that Traversing the initial weighted particle swarm, training a battery life prediction model based on the battery operation monitoring indicator set, and obtaining the initial weighted particle swarm verification accuracy, including: Extracting a first particle according to the initial weighted particle group, wherein the first particle has a first battery operation monitoring indicator weight set; Configuring a battery life prediction model topology, wherein the battery life prediction model topology is a neural network topology, an input node of the neural network topology corresponds one-to-one to each node of the battery operation monitoring indicator set, and the input node of the neural network topology is one, which is used to output the predicted life; According to the first battery operation monitoring indicator weight set, weight configuration is performed on the input nodes of the battery life prediction model topology to obtain an initial battery life prediction model; According to the battery operation monitoring indicator set, the initial battery life prediction model is trained to obtain a first particle battery life prediction model; The first particle battery life prediction model is verified, and the verification accuracy is obtained and added into the initial weighted particle swarm training accuracy.

3. The method according to claim 2, characterized in that According to the battery operation monitoring indicator set, the initial battery life prediction model is trained to obtain a first particle battery life prediction model, including: Taking the battery operation monitoring indicator set as input data and the battery life data as output data, collecting the battery operation monitoring indicator characteristic value set and the battery life identification data; Using the battery life identification data as supervision and the battery operation monitoring index characteristic value set as input, training a first-level first-particle battery life prediction model; When the amount of data of the first-level loss data set that does not satisfy the first accuracy threshold of the first-level first-particle battery life prediction model is less than the data amount threshold, the first-level first-particle battery life prediction model is set to the first particle battery life prediction model.

4. The method according to claim 3, characterized in that Also includes: When the amount of data of the first-level loss data set for which the first-level first-particle battery life prediction model does not meet the first accuracy threshold is greater than or equal to the data amount threshold, extracting a life deviation vector set of the first-level first-particle battery life prediction model; Performing system error analysis on the life deviation vector set to obtain a system error vector; Using the inverse of the system error vector as supervision data and the battery operation monitoring index characteristic value set as input, training a secondary first particle battery life prediction model; Until the data amount of the N-level loss data set for which the N-level first-particle battery life prediction model does not meet the first accuracy threshold is less than the data amount threshold, the outputs of the first-level first-particle battery life prediction model, the second-level first-particle battery life prediction model, and the N-level first-particle battery life prediction model are added and fully connected to obtain the first particle battery life prediction model.

5. The method according to claim 1, characterized in that When the verification accuracy of the initial weighted particle swarm is less than or equal to the convergence accuracy, the verification accuracy of the initial weighted particle swarm is used as the fitness function, and the initial weighted particle swarm is subjected to particle optimization to obtain an updated particle set, including: Performing pairwise evaluation of particle distances on the initial weighted particle swarm to obtain a set of particle distance evaluation values; According to the particle distance evaluation value set, clustering is performed based on the particle distance threshold to obtain multiple groups of initial weighted particles; According to the verification accuracy of the initial weighted particle group, the multiple groups of initial weighted particles are sorted from large to small in verification accuracy to obtain sorting results of the multiple groups of initial weighted particles; According to the sorting results of the multiple groups of initial weighted particles, collaborative particle optimization is performed on the initial weighted particle group to obtain an updated particle set.

6. The method according to claim 5, characterized in that The particle distances of the initial weighted particle swarm are evaluated pairwise to obtain a set of particle distance evaluation values, including: Randomly obtaining a first particle and a second particle of the initial weighted particle group, wherein the first particle has a first battery operation monitoring indicator weight set, and the second particle has a second battery operation monitoring indicator weight set; The Euclidean distance between the first battery operation monitoring indicator weight set and the second battery operation monitoring indicator weight set is calculated, set as a first particle distance evaluation value, and added to the particle distance evaluation value set.

7. The method according to claim 5, characterized in that According to the sorting results of the multiple groups of initial weighted particles, collaborative particle optimization is performed on the initial weighted particle group to obtain an updated particle set, including: Randomly extracting a first group of initial weighted particle sorting results and a second group of initial weighted particle sorting results from the multiple groups of initial weighted particle sorting results; Obtaining the first initial weighted particle of the first group of initial weighted particle sorting results, guiding the initial weighted particles at the end of a preset number of the second group of initial weighted particle sorting results to mutate, reducing the particle distance evaluation value with the first initial weighted particle, and obtaining a first particle update result; Obtaining the second first initial weight particle of the second group of initial weight particle sorting results, guiding the initial weight particles at the end of a preset number of the first group of initial weight particle sorting results to mutate, reducing the particle distance evaluation value with the second first initial weight particle, and obtaining a second particle update result; The first particle update result and the second particle update result are added to the update particle set.

8. The method according to claim 7, characterized in that Obtaining the first initial weighted particle of the first group of initial weighted particle sorting results, guiding the initial weighted particles at the end of a preset number of the second group of initial weighted particle sorting results to mutate, reducing the particle distance evaluation value with the first initial weighted particle, and obtaining the first particle update result, including: Based on the weight dimension, randomly select a value to obtain the k value, 1≤k≤weight dimension; According to the k value, taking the first first initial weight particle as the target, mutating the preset number of tail initial weight particles, reducing the particle distance evaluation value with the first first initial weight particle, and obtaining the first particle update result.

Images