Lithium ion battery energy storage system state monitoring method and device

Abstract

The invention discloses a lithium ion battery energy storage system state monitoring method and device. The method comprises the following steps: collecting operation parameter historical data under each state type in a lithium ion battery energy storage system; preprocessing the collected historical data to obtain an expanded training data set; performing feature extraction on the expanded training data set to obtain a training data set after feature extraction; performing feature dimension reduction on the training data set after feature extraction to obtain a training data set after dimension reduction; performing optimization training on the constructed classifier model by combining an extreme learning machine algorithm based on quantum information entanglement measurement on the basis of the dimensionality-reduced training data set; and performing feature extraction and feature dimension reduction on the obtained real-time data of each operation parameter, inputting the data into the trained classifier model for processing, and outputting a state monitoring result of the lithium ion battery energy storage system. According to the invention, the operation state type of the lithium ion battery energy storage system can be timely and accurately monitored, so that corresponding measures can be taken as soon as possible.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion battery management, and in particular to a method and device for monitoring the state of a lithium-ion battery energy storage system. Background Art

[0002] With the widespread application of battery technology, lithium-ion batteries have become one of the key technologies in the field of energy storage and electric vehicles. It is particularly important to ensure the safe operation of lithium-ion battery energy storage systems. The battery management system (BMS) needs to be able to monitor the status of the battery in real time to prevent and respond to potential safety risks such as overheating, overcharging or battery failure.

[0003] In traditional battery status monitoring methods, battery management systems usually use a single neural network for feature extraction and prediction when processing complex battery monitoring data, such as BP neural network, ConvNet deep convolutional neural network and LSTM long short-term memory neural network. However, these traditional monitoring methods have many deficiencies in data preprocessing, model construction and training; among them, data preprocessing often fails to fully expand samples and simulate data characteristics in actual operating environments, resulting in training data that cannot fully reflect the actual battery usage, increasing safety risks; in addition, these neural network models often have problems such as gradient disappearance, overfitting and insufficient generalization ability during training, which limits the model's adaptability to new situations and the accuracy of prediction. Ultimately, due to the limitations of training efficiency and model performance, the responsiveness and accuracy of monitoring results cannot meet the requirements of high safety, and there are potential safety risks. Therefore, a new solution is needed that can efficiently analyze and process complex data while improving monitoring accuracy to enhance the safety and reliability of lithium-ion battery energy storage systems. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a lithium-ion battery energy storage system status monitoring method and device, so as to solve the problem that the prior method fails to make full use of existing data resources to improve the accuracy and timeliness of monitoring results, and further enhance the safety and reliability of the lithium-ion battery energy storage system.

[0005] The present invention adopts the following technical solution.

[0006] In a first aspect, the present invention provides a method for monitoring the state of a lithium-ion battery energy storage system, the method comprising: Collect historical data of operating parameters in various states of lithium-ion battery energy storage systems; Preprocess the collected historical data to obtain an expanded training data set; Perform feature extraction on the expanded training data set to obtain a training data set after feature extraction; Performing feature dimensionality reduction on the training data set after feature extraction to obtain a training data set after dimensionality reduction; Based on the reduced-dimensional training data set, the constructed classifier model is optimized and trained in combination with the extreme learning machine algorithm based on quantum information entanglement metric; The real-time data of each operating parameter is extracted and reduced in dimension, and then input into the trained classifier model for processing, and the status monitoring results of the lithium-ion battery energy storage system are output.

[0007] Optionally, the operating parameters include at least one of the battery voltage, current, temperature, charging status, vibration frequency, installation location, ambient humidity and ambient temperature; the status types include normal operation, abnormal status and danger warning.

[0008] Optionally, the step of preprocessing the collected historical data includes: Count the number of historical data collected in each state type respectively; The historical data of the corresponding type that is less than the set number is regarded as the minority sample set, and the historical data of the remaining types is regarded as the majority sample set; For each minority sample set, the SMOTE oversampling algorithm based on nonlinear spatial mapping is used to expand the sample to obtain the expanded synthetic sample set; The expanded synthetic sample sets and the majority sample sets are used as expanded training data sets.

[0009] Optionally, the step of performing sample expansion using a SMOTE oversampling algorithm based on nonlinear space mapping includes: Step A: For the minority class sample set Any sample in , based on the current scaling factor , the corresponding new samples are generated through the SMOTE oversampling algorithm ; Step B: For the new sample currently generated The following nonlinear mapping function processing is performed: , In the formula, Indicated in The synthetic samples after nonlinear mapping at the iteration; Representation sample In the New samples generated by iterations; is the Sigmoid activation function; It is a sample No. Features, is the number of features; It is in The nonlinear mapping weight at the iteration; Step C: After updating the nonlinear mapping weights and proportional coefficients, let , return to step A until the first termination condition is met, and output the synthetic sample after the current nonlinear mapping ; Step D: Repeat steps A to C until the minority class sample set is traversed. All samples in the final output are the corresponding nonlinearly mapped synthetic sample sets. .

[0010] Optionally, the step of extracting features from the expanded training data set includes: A 3-layer fully connected neural network is used to build a feature extraction model; Based on the expanded training data set, the constructed feature extraction model is trained in combination with an optimization algorithm based on plant growth photosensitivity response, and the parameter selection of the neural network is optimized using a sparse constraint strategy, so that the neural network can automatically remove unimportant connections during the learning process, thereby obtaining a trained and optimized feature extraction model; The expanded training data set is input into the trained and optimized feature extraction model, and the training data set after feature extraction is output.

[0011] Optionally, the training optimization step of the feature extraction model specifically includes: S3.1. Initialize the neural network parameters of the feature extraction model: weights and bias , to simulate the initial state of plant growth; S3.2. Update the neural network parameters by simulating the plant's response to light intensity. The expression is as follows: , In the formula, For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the loss function of the neural network, where the loss function The predicted label corresponding to the input sample is the category with the highest probability calculated by the preset Softmax classification function after feature extraction. is the loss function About Neural Network Weights The gradient of For the The learning rate of the iteration; is the adjustment coefficient of the photosensitivity effect; represents the function of simulating photosynthesis efficiency; Indicates A function of the simulated photosynthesis efficiency at each iteration; is the sparse regularization term; S3.3. Based on the currently updated neural network parameters, evaluate the contribution of each layer of the neural network to the model output results; when the contribution metrics of all network layers are greater than the preset threshold and the loss function converges, stop the iteration and obtain the final trained feature extraction model; otherwise, adjust the learning rate After updating, , return to step S3.2.

[0012] Optionally, the function for simulating photosynthesis efficiency in step S3.2 and the sparse regularization term The calculation formulas are: , , In the formula, is the Sigmoid activation function; Represents the output features of the middle layer of the neural network; is the weight matrix number of rows; For the The weight parameters of the row neurons; is the adjustment factor, used for L1 regularization of balancing weights; Represents the weight matrix Middle The L1 norm of the row, used to simulate the energy consumption of photosynthesis; is the weight matrix Middle L2 norm of the column; is the sparse regularization adjustment factor; is the weight matrix The number of columns.

[0013] Optionally, the step of performing feature dimension reduction on the training data set after feature extraction includes: A feature dimensionality reduction model is constructed using an autoencoder, wherein the autoencoder includes an encoder and a decoder; Based on the training data set after feature extraction, the constructed feature dimension reduction model is optimized and trained in combination with an optimization algorithm based on nonlinear dynamic programming, including: using an encoder to compress high-dimensional input data into a low-dimensional feature vector, a decoder to reconstruct the original input according to the low-dimensional feature vector, and optimizing the network parameters of the feature dimension reduction model by minimizing the reconstruction error between input and output, thereby obtaining a trained and optimized feature dimension reduction model; The training data set after feature extraction is input into the trained and optimized feature dimension reduction model, and the training data set after dimension reduction is output.

[0014] Optionally, the specific steps of optimizing the constructed feature dimension reduction model by combining the optimization algorithm based on nonlinear dynamic programming include: S4.1. Initialize the weights of the autoencoder and bias ; S4.2, using the forward propagation method, the encoder processes the training data samples after feature extraction, maps the high-dimensional data to a low-dimensional space, and generates its corresponding low-dimensional feature vector , expressed as: , In the formula, For the original input sample The encoded low-dimensional feature vector; and are the weights and biases of the encoder respectively; is the ReLNU activation function; S4.3. Utilizing low-dimensional feature vectors through the decoder To reconstruct the original input , expressed as: , In the formula, is the reconstructed input data; and are the weights and biases of the decoder respectively; S4.4. The mean square error is used to measure the desired minimization of the difference between the original input and the reconstructed output. The expression for calculating the loss function based on the reconstruction error is as follows: , In the formula, is the loss function; is the number of data samples; Indicates The original input samples, Indicates reconstructed input samples; S4.5. Use gradient descent to update the weights and biases in the autoencoder to reduce the reconstruction error. S4.6. Based on the weights and biases of the currently updated autoencoder, repeat steps S4.2 to S4.5 until the set second termination iteration condition is met to obtain the final trained feature dimensionality reduction model.

[0015] Optionally, the optimizing training of the constructed classifier model by combining an extreme learning machine algorithm based on quantum information entanglement metric includes: The reduced-dimensional features are converted into the form of qubits, and the superposition and entanglement states of qubits are used to enhance the ability of feature representation to improve the classification accuracy of the classifier model.

[0016] Optionally, the specific steps of optimizing the constructed classifier model by combining the extreme learning machine algorithm based on quantum information entanglement metric include: S501, the input dimensionality-reduced training data set is transformed into Medium Features Converting to qubits In the form of: , , In the formula, For the The qubit representation of each feature; Indicates quantum ground state; For the training data set The number of sample features is equal to the number of quantum ground states; Features In the quantum ground state The superposition coefficient of For training data set The Input features; A hyperparameter to control the sensitivity of the mapping; Indicates The center vector of the quantum ground state; Represents input features Quantum Ground State Center The Euclidean distance between Indicates The center vector of the quantum ground state; Represents input features Quantum Ground State Center The Euclidean distance between S502, respectively calculate the entanglement metric between each qubit using the following formula: , In the formula, Respectively represent Input features and Input features The entanglement measure between For qubit In the input feature The probability distribution of For qubit In the input feature The probability distribution of S503, based on the entanglement metric between the qubits, respectively recombining and optimizing the input features to obtain the corresponding reconstructed feature vectors; S504, the reconstructed feature vectors The reconstruction feature set composed of Input into the extreme learning machine for training to obtain a trained classifier model.

[0017] Optionally, in step S502, the quantum bit In the input feature The probability distribution of The calculation method is expressed as: , In the formula, Representation characteristics In the quantum ground state The square of the superposition coefficient, Representation characteristics In the quantum ground state The square of the superposition coefficient.

[0018] Optionally, in step S503, the expression for recombining and optimizing each input feature is as follows: , In the formula, For the Input features The reconstructed feature vector.

[0019] In a second aspect, the present invention provides a lithium-ion battery energy storage system status monitoring device, which runs the steps of any method described in the first aspect of the present invention, and the device includes: A data collection module is used to collect historical data of operating parameters in various states of the lithium-ion battery energy storage system; A preprocessing module is used to preprocess the collected historical data to obtain an expanded training data set; A feature extraction module is used to extract features from the expanded training data set to obtain a training data set after feature extraction; A feature dimension reduction module is used to perform feature dimension reduction on the training data set after feature extraction to obtain a training data set after dimension reduction; The classifier training module is used to optimize the constructed classifier model based on the training data set after dimensionality reduction and the extreme learning machine algorithm based on quantum information entanglement measurement; The status monitoring module is used to extract features from the real-time data of various operating parameters, reduce the feature dimension, and then input them into the trained classifier model for processing, and output the status monitoring results of the lithium-ion battery energy storage system.

[0020] In a third aspect, the present invention provides a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of any one of the methods described in the first aspect of the present invention.

[0021] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect of the present invention.

[0022] The beneficial effects of the present invention are as follows: 1. In order to solve the problem that insufficient training samples in existing data resources easily lead to poor model generalization ability and affect the accuracy of the model, the present invention adopts the SMOTE oversampling algorithm based on nonlinear spatial mapping to generate samples and expand the data samples; and the present invention uses a nonlinear mapping layer to perform nonlinear mapping enhancement on the generated samples to more finely understand the complexity of the lithium-ion battery energy storage system data, thereby generating synthetic data that is more in line with the actual situation, so as to improve the monitoring accuracy.

[0023] 2. The present invention aims to solve the shortcomings of existing feature extraction network models, such as gradient vanishing, gradient exploding or falling into local optimal solutions, which affect the stability of training and model performance. The present invention adopts an optimization algorithm based on plant growth photosensitivity to optimize the constructed feature extraction model, thereby solving the problems of overfitting and slow learning rate of the neural network during the feature extraction process, and adopts a sparse constraint strategy to optimize the parameter selection of the neural network, so that the network can automatically remove unimportant connections during the learning process, thereby improving the generalization ability and training efficiency of the model.

[0024] 3. The present invention uses an optimization algorithm based on nonlinear dynamic programming to optimize the constructed feature dimensionality reduction model, so that the autoencoder of the feature dimensionality reduction model can more efficiently select and combine input features during feature dimensionality reduction, optimize the learning process of the network, and improve the model's ability to capture the intrinsic structure of the data.

[0025] 4. The present invention uses an extreme learning machine algorithm based on quantum information entanglement measurement to optimize the constructed classifier model, converts the reduced-dimensional features into the form of quantum bits, and uses the superposition and entanglement states of quantum bits to enhance the ability of feature representation, thereby improving the classification accuracy of the classifier model.

[0026] 5. In general, the present invention significantly improves the processing and prediction capabilities of the lithium-ion battery energy storage system in a complex data environment by expanding the sampled data, combining an optimization algorithm based on plant growth photosensitivity, an optimization algorithm based on nonlinear dynamic programming, and an extreme learning machine classifier based on quantum information entanglement measurement. It not only enhances the representativeness of the data and the accuracy of the features, but also improves the generalization ability and training efficiency of the model, thereby effectively improving the timeliness and accuracy of the system's safety monitoring and risk warning, so that corresponding measures can be taken as early as possible, greatly enhancing the safety and reliability of the lithium-ion battery energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the method for monitoring the state of a lithium-ion battery energy storage system in the present invention; Figure 2 It is a schematic diagram of the flow of training and application of each model in the present invention; Figure 3 This is a structural principle block diagram of the lithium-ion battery energy storage system status monitoring device in the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in the present invention are only part of the embodiments of the present invention, not all of the embodiments. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0029] Embodiment 1: Reference Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a lithium-ion battery energy storage system state monitoring method, which specifically includes the following steps: Step S1: Collecting historical data of operating parameters in various state types of the lithium-ion battery energy storage system; The data in this embodiment are derived from the real-time monitoring system of the battery, the historical fault database and the real-time environmental monitoring sensor, and the data is stored in JSON format. Among them, the operating parameters include at least one of the battery's voltage, current, temperature, charging status, vibration frequency, installation location, ambient humidity and ambient temperature. It should be noted that this embodiment is only used to illustrate a data format and type of the present invention. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of attributes of the data may reach dozens or even hundreds. Furthermore, the collected data is labeled, and the labeled categories of the present invention include: normal operation, abnormal state, and danger warning, a total of 3 categories.

[0030] Step S2: preprocessing the collected historical data to obtain an expanded training data set; It is further explained that, in order to address the problem that insufficient existing training samples easily lead to poor model generalization ability and affect the accuracy of the model, the present invention adopts the SMOTE oversampling algorithm based on nonlinear spatial mapping to generate samples and thus achieve data expansion; on the basis of the traditional SMOTE oversampling algorithm, the present invention adopts a nonlinear mapping layer to perform nonlinear mapping enhancement to more finely simulate the complexity of the battery management system data, thereby generating synthetic data that is more in line with the actual situation.

[0031] In a preferred but non-limiting embodiment, the step of preprocessing the collected historical data in step S2 includes: S2.1. Count the number of historical data collected in each state type respectively; S2.2, historical data of the corresponding type that is less than the set number is regarded as the minority sample set, and historical data of the remaining types is regarded as the majority sample set; S2.3, for each minority sample set, a SMOTE oversampling algorithm based on nonlinear spatial mapping is used to perform sample expansion to obtain an expanded synthetic sample set; S2.4. Using the expanded synthetic sample sets and the expanded majority sample sets as expanded training data sets.

[0032] In the embodiment of the present invention, the process of using the SMOTE oversampling algorithm based on nonlinear space mapping to perform sample expansion in step S2.3 is as follows: Step A: For the minority class sample set Any sample in , the corresponding new samples are generated by the SMOTE oversampling algorithm, and the expression is as follows , In the formula, Representation sample In the New samples generated by iterations; Represents the minority class sample set The samples; for The neighboring samples of Represents a random number between 0 and 1; Indicated in The ratio coefficient of samples generated in the iteration; In this embodiment, the neighboring samples It is based on K-nearest neighbor algorithm or Euclidean distance to sample A sample randomly selected from the set of generated neighboring samples.

[0033] Step B: The following nonlinear mapping function processing is performed to adjust its feature distribution to better mimic the data characteristics in the actual operating environment. The expression is as follows: , In the formula, Indicated in The synthetic samples after nonlinear mapping at the iteration; is the Sigmoid activation function; It is a sample No. Features, is the number of features; It is in The nonlinear mapping weight at the iteration; Step C: After updating the nonlinear mapping weight and proportional coefficient according to the following formula, let , return to step A until the first termination condition is met, and output the synthetic sample after the current nonlinear mapping ; , , , , In the formula, and Respectively expressed in The nonlinear mapping weights and scaling coefficients for the iterations; and Respectively expressed in The nonlinear mapping weights and scaling coefficients for the iterations; represents the momentum update rate; and Respectively and The gradient of Indicated in The nonlinear mapping weight at the iteration; Represents the learning rate. In this embodiment, Set to 0.01; indicates sample With Samples at iteration The reconstruction error loss function between (i.e., L2 norm); Represents the loss function Gradients with respect to weight parameters; Among them, the gradient and The calculation methods are as follows: , , In the formula, is the number of synthetic samples; and They are the partial derivatives of the nonlinear mapping function with respect to the weight parameter and the proportional coefficient respectively; Furthermore, the first termination condition set in this embodiment is to reach a preset maximum number of iterations or a loss function. Convergence; preferably, the preset maximum number of iterations is set to 1000 times.

[0034] Step D: Repeat steps A to C until the minority class sample set is traversed. All samples in the final output are the corresponding nonlinearly mapped synthetic sample sets. .

[0035] Step S3: performing feature extraction on the expanded training data set to obtain a training data set after feature extraction; In this embodiment, the expanded data is input into the feature extraction model to train the feature extraction model, and a 3-layer fully connected neural network is used for feature extraction. In view of the prior art, some schemes use neural networks for feature extraction, which may encounter problems such as gradient vanishing, gradient explosion, or falling into a local optimal solution, affecting the stability of the training and the performance of the model; step S3 of this embodiment uses an optimized algorithm based on the photosensitive response of plant growth to train the feature extraction model. In addition, in order to solve the problems of overfitting and slow learning rate of neural networks during feature extraction, the present invention adopts a sparse constraint strategy to optimize the parameter selection of the neural network, so that the network can automatically remove unimportant connections during the learning process, thereby improving the generalization ability and training efficiency of the model.

[0036] As an embodiment of the present invention, the specific steps of optimizing and training the constructed feature extraction model in combination with the optimization algorithm based on plant growth photosensitive response include: S3.1. Initialize the neural network parameters of the feature extraction model; among them, the model weight and bias Initialized to a small random value to simulate the initial state of plant growth, its expression is as follows: , , In the formula, and are the initial weights and initial biases of the neural network, is a set constant; represents a normal distribution with mean 0 and unit variance; preferably, Set to 0.0001.

[0037] S3.2. Update the neural network parameters by simulating the plant's response to light intensity. The expression is as follows: , In the formula, For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the loss function of the neural network, where the loss function The predicted label corresponding to the input sample is the category with the highest probability calculated by the preset Softmax classification function after feature extraction. is the loss function About Neural Network Weights The gradient of For the The learning rate of the iteration; is the adjustment coefficient of the photosensitivity effect; preferably, It is set in a dynamic adjustment mode, and its initial value is set to 0.001. Set to 0.3; represents the function of simulating photosynthesis efficiency; Indicates A function of the simulated photosynthesis efficiency at each iteration; is the sparse regularization term; Among them, the function simulating photosynthesis efficiency and the sparse regularization term The calculation formulas are: , , In the formula, is the Sigmoid activation function; Represents the output features of the middle layer of the neural network; is the weight matrix number of rows; For the The weight parameters of the row neurons; is the adjustment factor, used for L1 regularization of balancing weights; Represents the weight matrix Middle The L1 norm of the row, used to simulate the energy consumption of photosynthesis; is the weight matrix Middle L2 norm of the column; is the sparse regularization adjustment factor; is the weight matrix The number of columns; preferably, Set to 0.4.

[0038] It is further explained that, during the training process, this embodiment updates the model parameters by simulating the plant's response to light intensity, and the weight adjustment of a specific layer is not only based on error back propagation, but also optimized according to the performance of the specific layer in the overall network (equivalent to photosynthesis efficiency), so that the weights of key features are more prominent.

[0039] S3.3. Based on the currently updated neural network parameters, evaluate the contribution of each layer of the neural network to the model output results; when the contribution metrics of all network layers are greater than the preset threshold and the loss function converges, stop the iteration and obtain the final trained feature extraction model; otherwise, adjust the learning rate After updating, , return to step S3.2.

[0040] Specifically, when the model is close to training completion, the impact of each layer of the neural network on the final result is evaluated, and the parameter settings of the layers that have a significant impact on the results are further adjusted and optimized; the calculation formula for evaluating the contribution measure of each layer of the neural network to the model output result is as follows: , In the formula, For the Contribution measure of layer neural network; is the output feature vector of the feature extraction model; For the The output feature vector of the layer neural network; Express The partial derivative of is the first The output feature vector of the layer neural network is used to improve the robustness of the model; in, The generation method is expressed as: , , In the formula, is the noise injection intensity, which is dynamically adjusted according to the current training error of the model; represents normal distribution noise with mean 0 and covariance as the unit matrix; is the baseline intensity of the noise injection, preferably, Set to 0.3; Is the current The learning rate of the iteration; For the model The loss function value of the iteration; It is further explained that after each round of training, the learning rate is adjusted according to the training results, similar to how plants adjust their growth strategies according to environmental conditions at different growth stages. The dynamic adjustment of the learning rate is based on the changes in the loss function.

[0041] In a preferred but non-limiting embodiment, in step S3.3, the learning rate The update expression is as follows: , In the formula, For the The learning rate of the iteration; is the regulating factor; is the loss variation; preferably, It is set in a dynamic adjustment mode, with the initial value set to 0.4 and the adjustment factor The calculation method is expressed as: , , In the formula, For the model The loss function value of the iteration; The function is a hyperbolic tangent function, which is used to smooth the loss changes , ensure that the learning rate is adjusted within a reasonable range to avoid shocks during the learning process.

[0042] Furthermore, the expanded training data set is input into the trained feature extraction model to obtain a training data set after feature extraction.

[0043] Step S4: performing feature dimensionality reduction on the training data set after feature extraction to obtain a training data set after dimensionality reduction; Specifically, this embodiment inputs the data after feature extraction into the feature dimension reduction model to train the feature dimension reduction model, and uses an autoencoder as the feature dimension reduction model. The autoencoder consists of an encoder and a decoder. The encoder is responsible for compressing the high-dimensional input data into a low-dimensional feature representation, and the decoder reconstructs the original input from these low-dimensional features. The network is trained by minimizing the reconstruction error between the input and output, thereby learning an effective low-dimensional representation of the data. In addition, this embodiment uses an optimization algorithm based on nonlinear dynamic programming to improve the autoencoder to more efficiently select and combine input features during feature dimension reduction, optimize the network's learning process, and improve the model's ability to capture the intrinsic structure of the data.

[0044] As an embodiment of the present invention, the specific process steps of optimizing the constructed feature dimension reduction model by combining the optimization algorithm based on nonlinear dynamic programming include: S4.1. Initialize the weights of the autoencoder and bias , which is expressed as follows: , , In the formula, and Represent the initial weight and bias of the autoencoder respectively; is the set standard deviation coefficient, represents a normal distribution with a mean of 0 and a covariance of the unit matrix; wherein all weights and bias parameters of the autoencoder in this embodiment are initialized by small-scale random numbers.

[0045] S4.2, using the forward propagation method, the encoder processes the training data samples after feature extraction, maps the high-dimensional data to a low-dimensional space, and generates its corresponding low-dimensional feature vector , expressed as: , In the formula, For the original input sample The encoded low-dimensional feature vector; and are the weights and biases of the encoder respectively; is the ReLNU activation function; S4.3. Utilizing low-dimensional feature vectors through the decoder To reconstruct the original input , expressed as: , In the formula, is the reconstructed input data; and are the weight and bias of the decoder respectively; it is further explained that, in this embodiment, the low-dimensional feature representation is passed to the decoder, and the decoder attempts to reconstruct the original input data, and the desired minimization of the difference between the original input and the reconstructed output is measured by the mean square error.

[0046] S4.4. The mean square error is used to measure the desired minimization of the difference between the original input and the reconstructed output. The expression for calculating the loss function based on the reconstruction error is as follows: , In the formula, is the loss function; is the number of data samples; Indicates The original input samples, Indicates reconstructed input samples; S4.5. Use the gradient descent method to update the weights and biases in the autoencoder to reduce the reconstruction error. The expression is as follows: , , In the formula, and Respectively represent Second and The weight at the iteration; For the The learning rate of the autoencoder at iterations; preferably, The initial value of is set to 0.001; It is the energy consumption monitoring function; and They are loss functions The calculation formulas for the gradients of weights and biases are as follows: , , In the formula, represents the derivative of the ReLU activation function, It is for The low-dimensional feature vector after encoding the input samples; The energy consumption monitoring function The calculation method is: , In the formula, is the weight coefficient of energy consumption, preferably, Set to 0.35. represents the square norm; S4.6. Based on the weights and biases of the currently updated autoencoder, repeat steps S4.2 to S4.5 until the set second termination iteration condition is met to obtain the final trained feature dimensionality reduction model.

[0047] It should also be noted that in each iteration of the above steps, the weights and biases of the encoder and decoder can be updated using the conventional gradient descent method, which will not be described in detail here.

[0048] In a preferred but non-limiting embodiment, the present invention uses a nonlinear dynamic programming greedy strategy to update the learning rate of the autoencoder , thereby accelerating the learning process and improving model performance, which is expressed as follows: , In the formula, Indicates The learning rate of the autoencoder at the iteration; is the sensitivity coefficient of learning rate adjustment, preferably, Set to 0.3; and Respectively expressed in Second and The loss function at the iteration Gradients with respect to weights and biases.

[0049] In the embodiment of the present invention, the second termination iteration condition set in step S4.6 is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0050] Furthermore, the training data set after feature extraction is input into the trained feature dimension reduction model to obtain the training data set after dimension reduction.

[0051] Step S5: Based on the training data set after dimensionality reduction, the constructed classifier model is optimized and trained in combination with an extreme learning machine algorithm based on quantum information entanglement metric; Specifically, the reduced-dimensional data is input into a classifier model for training the classifier model. The present invention adopts an extreme learning machine algorithm based on quantum information entanglement metric as a classification algorithm, converts the reduced-dimensional features into the form of quantum bits, and utilizes the superposition and entanglement states of quantum bits to enhance the ability of feature representation.

[0052] As an embodiment of the present invention, the specific process steps of optimizing the constructed classifier model in step S5 in combination with the extreme learning machine algorithm based on quantum information entanglement metric include: S501, the input dimensionality-reduced training data set is transformed into Medium Features Converting to qubits In the form of: , , In the formula, For the The qubit representation of each feature; Indicates quantum ground state; For the training data set The number of sample features is equal to the number of quantum ground states; Features In the quantum ground state The superposition coefficient is used to represent the input features The degree of superposition of each quantum ground state; For training data set The Input features; A hyperparameter to control the sensitivity of the mapping; Indicates The center vector of the quantum ground state; Represents input features Quantum Ground State Center The Euclidean distance between Indicates The center vector of the quantum ground state; Represents input features Quantum Ground State Center The Euclidean distance between ; S502, respectively calculating the entanglement metric between each qubit by the following formula to determine the relationship between each feature and the importance of information, and the expression is as follows: , In the formula, Respectively represent Input features and Input features The entanglement measure between For quantum bits In the input feature The probability distribution of For qubit In the input feature The probability distribution of Among them, quantum bits In the input feature The probability distribution of The calculation method is expressed as: , In the formula, Representation characteristics In the quantum ground state The square of the superposition coefficient, Representation characteristics In the quantum ground state The square of the superposition coefficient; S503, based on the entanglement metric between the qubits, the input features are recombined and optimized respectively, and the expressions are as follows: , In the formula, For the Input features The reconstructed feature vector; S504, the reconstructed feature vectors The reconstruction feature set composed of Input into the extreme learning machine for training to obtain a trained classifier model; among them, the output layer weight of the extreme learning machine The calculation method is expressed as: , In the formula, Reconstruct the feature set The calculated hidden layer output is for The transpose of is the target label matrix, is the regularization parameter, is the identity matrix.

[0053] Step S6: Extract features from the acquired real-time data of each operating parameter, reduce the feature dimension, and then input it into the trained classifier model for processing, and output the status monitoring result of the lithium-ion battery energy storage system.

[0054] In real-time monitoring, trained feature extraction models, feature dimension reduction models, and classifier models are used to process new samples to implement fire linkage decisions of the battery management system. Based on the classification results, the battery management system may trigger corresponding safety protocols, such as increasing the monitoring frequency when a warning state is detected, or notifying the maintenance team when maintenance status is required. The classification results and any triggered responses are recorded in the system log for future analysis and system optimization.

[0055] The beneficial effects of the present invention are as follows: 1. In order to solve the problem that insufficient training samples in existing data resources easily lead to poor model generalization ability and affect the accuracy of the model, the present invention adopts the SMOTE oversampling algorithm based on nonlinear spatial mapping to generate samples and expand the data samples; and the present invention uses a nonlinear mapping layer to perform nonlinear mapping enhancement on the generated samples to more finely understand the complexity of the lithium-ion battery energy storage system data, thereby generating synthetic data that is more in line with the actual situation, so as to improve the monitoring accuracy.

[0056] 2. The present invention aims to solve the shortcomings of existing feature extraction network models, such as gradient vanishing, gradient exploding or falling into local optimal solutions, which affect the stability of training and model performance. The present invention adopts an optimization algorithm based on plant growth photosensitivity to optimize the constructed feature extraction model, thereby solving the problems of overfitting and slow learning rate of the neural network during the feature extraction process, and adopts a sparse constraint strategy to optimize the parameter selection of the neural network, so that the network can automatically remove unimportant connections during the learning process, thereby improving the generalization ability and training efficiency of the model.

[0057] 3. The present invention uses an optimization algorithm based on nonlinear dynamic programming to optimize the constructed feature dimensionality reduction model, so that the autoencoder of the feature dimensionality reduction model can more efficiently select and combine input features during feature dimensionality reduction, optimize the learning process of the network, and improve the model's ability to capture the intrinsic structure of the data.

[0058] 4. The present invention uses an extreme learning machine algorithm based on quantum information entanglement measurement to optimize the constructed classifier model, converts the reduced-dimensional features into the form of quantum bits, and uses the superposition and entanglement states of quantum bits to enhance the ability of feature representation, thereby improving the classification accuracy of the classifier model.

[0059] 5. In general, the present invention significantly improves the processing and prediction capabilities of the lithium-ion battery energy storage system in a complex data environment by expanding the sampled data, combining an optimization algorithm based on plant growth photosensitivity, an optimization algorithm based on nonlinear dynamic programming, and an extreme learning machine classifier based on quantum information entanglement measurement. It not only enhances the representativeness of the data and the accuracy of the features, but also improves the generalization ability and training efficiency of the model, thereby effectively improving the timeliness and accuracy of the system's safety monitoring and risk warning, so that corresponding measures can be taken as early as possible, greatly enhancing the safety and reliability of the lithium-ion battery energy storage system.

[0060] Embodiment 2: like Figure 3 As shown, the present invention provides a lithium-ion battery energy storage system state monitoring device, which is used to implement the steps of the method in the above embodiment 1, and the device specifically includes: A data collection module is used to collect historical data of operating parameters in various states of the lithium-ion battery energy storage system; A preprocessing module is used to preprocess the collected historical data to obtain an expanded training data set; A feature extraction module is used to extract features from the expanded training data set to obtain a training data set after feature extraction; A feature dimension reduction module is used to perform feature dimension reduction on the training data set after feature extraction to obtain a training data set after dimension reduction; The classifier training module is used to optimize the constructed classifier model based on the training data set after dimensionality reduction and the extreme learning machine algorithm based on quantum information entanglement measurement; The status monitoring module is used to extract features from the real-time data of various operating parameters, reduce the feature dimension, and then input them into the trained classifier model for processing, and output the status monitoring results of the lithium-ion battery energy storage system.

[0061] The lithium-ion battery energy storage system state monitoring device provided in the embodiment of the present invention and the lithium-ion battery energy storage system state monitoring method provided in Example 1 are based on the same technical concept and can produce the beneficial effects described in Example 1. For the contents not described in detail in this embodiment, please refer to Example 1.

[0062] Embodiment three: An embodiment of the present invention provides a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instruction to execute the steps of the method according to any one of the embodiments.

[0063] Embodiment 4: An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of any method described in Embodiment 1 are implemented.

[0064] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0065] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0066] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0067] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring the state of a lithium-ion battery energy storage system, characterized in that: Methods include: Collect historical data of operating parameters in various states of lithium-ion battery energy storage systems; Preprocess the collected historical data to obtain an expanded training data set; Perform feature extraction on the expanded training data set to obtain a training data set after feature extraction; Performing feature dimensionality reduction on the training data set after feature extraction to obtain a training data set after dimensionality reduction; Based on the reduced-dimensional training data set, the constructed classifier model is optimized and trained in combination with the extreme learning machine algorithm based on quantum information entanglement metric; The real-time data of each operating parameter is extracted and reduced in dimension, and then input into the trained classifier model for processing, and the status monitoring results of the lithium-ion battery energy storage system are output.

2. The lithium-ion battery energy storage system state monitoring method according to claim 1, characterized in that: The operating parameters include at least one of the battery voltage, current, temperature, charging status, vibration frequency, installation location, ambient humidity and ambient temperature; the status types include normal operation, abnormal status and danger warning.

3. The lithium-ion battery energy storage system state monitoring method according to claim 1, characterized in that: The step of preprocessing the collected historical data includes: Count the number of historical data collected in each state type respectively; The historical data of the corresponding type that is less than the set number is regarded as the minority sample set, and the historical data of the remaining types is regarded as the majority sample set; For each minority sample set, the SMOTE oversampling algorithm based on nonlinear spatial mapping is used to expand the sample to obtain the expanded synthetic sample set; The expanded synthetic sample sets and the majority sample sets are used as expanded training data sets.

4. The lithium-ion battery energy storage system state monitoring method according to claim 3, characterized in that: The step of using the SMOTE oversampling algorithm based on nonlinear space mapping to perform sample expansion includes: Step A: For the minority class sample set Any sample in , based on the current scaling factor , the corresponding new samples are generated through the SMOTE oversampling algorithm ; Step B: For the new sample currently generated The following nonlinear mapping function processing is performed: ， In the formula, Indicated in The synthetic samples after nonlinear mapping at the iteration; Representation sample In the New samples generated by iterations; is the Sigmoid activation function; It is a sample No. Features, is the number of features; It is in The nonlinear mapping weight at the iteration; Step C: After updating the nonlinear mapping weights and proportional coefficients, let , return to step A until the first termination condition is met, and output the synthetic sample after the current nonlinear mapping ; Step D: Repeat steps A to C until the minority class sample set is traversed. All samples in the final output are the corresponding nonlinearly mapped synthetic sample sets. .

5. The lithium-ion battery energy storage system state monitoring method according to claim 1, characterized in that: The step of extracting features from the expanded training data set includes: A 3-layer fully connected neural network is used to build a feature extraction model; Based on the expanded training data set, the constructed feature extraction model is trained in combination with an optimization algorithm based on plant growth photosensitivity response, and the parameter selection of the neural network is optimized using a sparse constraint strategy, so that the neural network can automatically remove unimportant connections during the learning process, thereby obtaining a trained and optimized feature extraction model; The expanded training data set is input into the trained and optimized feature extraction model, and the training data set after feature extraction is output.

6. The lithium-ion battery energy storage system state monitoring method according to claim 5, characterized in that: The training and optimization steps of the feature extraction model specifically include: S3.

1. Initialize the neural network parameters of the feature extraction model: weights and bias , to simulate the initial state of plant growth; S3.

2. Update the neural network parameters by simulating the plant's response to light intensity. The expression is as follows: ， In the formula, For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the loss function of the neural network, where the loss function The predicted label corresponding to the input sample is the category with the highest probability calculated by the preset Softmax classification function after feature extraction. is the loss function About Neural Network Weights The gradient of For the The learning rate of the iteration; is the adjustment coefficient of the photosensitivity effect; represents the function of simulating photosynthesis efficiency; Indicates A function of the simulated photosynthesis efficiency at each iteration; is the sparse regularization term; S3.

3. Based on the currently updated neural network parameters, evaluate the contribution of each layer of the neural network to the model output results; when the contribution metrics of all network layers are greater than the preset threshold and the loss function converges, stop the iteration and obtain the final trained feature extraction model; otherwise, adjust the learning rate After updating, , return to step S3.

2.

7. The lithium-ion battery energy storage system state monitoring method according to claim 6, characterized in that: Function for simulating photosynthesis efficiency in step S3.2 and the sparse regularization term The calculation formulas are: ， ， In the formula, is the Sigmoid activation function; Represents the output features of the middle layer of the neural network; is the weight matrix number of rows; For the The weight parameters of the row neurons; is the adjustment factor, used for L1 regularization of balancing weights; Represents the weight matrix Middle The L1 norm of the row, used to simulate the energy consumption of photosynthesis; is the weight matrix Middle L2 norm of the column; is the sparse regularization adjustment factor; is the weight matrix The number of columns.

8. The lithium-ion battery energy storage system state monitoring method according to claim 1, characterized in that: The step of performing feature dimension reduction on the training data set after feature extraction comprises: A feature dimensionality reduction model is constructed using an autoencoder, wherein the autoencoder includes an encoder and a decoder; Based on the training data set after feature extraction, the constructed feature dimension reduction model is optimized and trained in combination with an optimization algorithm based on nonlinear dynamic programming, including: using an encoder to compress high-dimensional input data into a low-dimensional feature vector, a decoder to reconstruct the original input according to the low-dimensional feature vector, and optimizing the network parameters of the feature dimension reduction model by minimizing the reconstruction error between input and output, thereby obtaining a trained and optimized feature dimension reduction model; The training data set after feature extraction is input into the trained and optimized feature dimension reduction model, and the training data set after dimension reduction is output.

9. The lithium-ion battery energy storage system state monitoring method according to claim 8, characterized in that: The specific steps of optimizing the constructed feature dimension reduction model by combining the optimization algorithm based on nonlinear dynamic programming include: S4.

1. Initialize the weights of the autoencoder and bias ; S4.2, using the forward propagation method, the encoder processes the training data samples after feature extraction, maps the high-dimensional data to a low-dimensional space, and generates its corresponding low-dimensional feature vector , expressed as: ， In the formula, For the original input sample The encoded low-dimensional feature vector; and are the weights and biases of the encoder respectively; is the ReLNU activation function; S4.

3. Utilizing low-dimensional feature vectors through the decoder To reconstruct the original input , expressed as: ， In the formula, is the reconstructed input data; and are the weight and bias of the decoder respectively; S4.

4. The mean square error is used to measure the desired minimization of the difference between the original input and the reconstructed output. The expression for calculating the loss function based on the reconstruction error is as follows: ， In the formula, is the loss function; is the number of data samples; Indicates The original input samples, Indicates reconstructed input samples; S4.

5. Use gradient descent to update the weights and biases in the autoencoder to reduce the reconstruction error. S4.

6. Based on the weights and biases of the currently updated autoencoder, repeat steps S4.2 to S4.5 until the set second termination iteration condition is met to obtain the final trained feature dimensionality reduction model.

10. The lithium-ion battery energy storage system state monitoring method according to claim 1, characterized in that: The optimization training of the constructed classifier model in combination with the extreme learning machine algorithm based on quantum information entanglement measurement includes: The reduced-dimensional features are converted into the form of qubits, and the superposition and entanglement states of qubits are used to enhance the ability of feature representation to improve the classification accuracy of the classifier model.

11. The lithium-ion battery energy storage system state monitoring method according to claim 10, characterized in that: The specific steps of optimizing the constructed classifier model by combining the extreme learning machine algorithm based on quantum information entanglement measurement include: S501, the input dimensionality-reduced training data set is transformed into Medium Features Converting to qubits In the form of: ， ， In the formula, For the The qubit representation of each feature; Indicates quantum ground state; For the training data set The number of sample features is equal to the number of quantum ground states; Features In the quantum ground state The superposition coefficient of For training data set The Input features; A hyperparameter to control the sensitivity of the mapping; Indicates The center vector of the quantum ground state; Represents input features Quantum Ground State Center The Euclidean distance between Indicates The center vector of the quantum ground state; Represents input features Quantum Ground State Center The Euclidean distance between S502, respectively calculate the entanglement metric between each qubit using the following formula: ， In the formula, Respectively represent Input features and Input features The entanglement measure between For qubit In the input feature The probability distribution of For qubit In the input feature The probability distribution of S503, based on the entanglement metric between the qubits, respectively recombining and optimizing the input features to obtain the corresponding reconstructed feature vectors; S504, the reconstructed feature vectors The reconstruction feature set composed of Input into the extreme learning machine for training to obtain a trained classifier model.

12. The lithium-ion battery energy storage system state monitoring method according to claim 11, characterized in that: In step S502, the quantum bit In the input feature The probability distribution of The calculation method is expressed as: ， In the formula, Representation characteristics In the quantum ground state The square of the superposition coefficient, Representation characteristics In the quantum ground state The square of the superposition coefficient.

13. The lithium-ion battery energy storage system state monitoring method according to claim 11, characterized in that: In step S503, the expression for recombining and optimizing each input feature is as follows: ， In the formula, For the Input features The reconstructed feature vector.

14. A lithium-ion battery energy storage system status monitoring device, running the lithium-ion battery energy storage system status monitoring method according to any one of claims 1 to 13, characterized in that: The device includes: A data collection module is used to collect historical data of operating parameters in various states of the lithium-ion battery energy storage system; A preprocessing module is used to preprocess the collected historical data to obtain an expanded training data set; A feature extraction module is used to extract features from the expanded training data set to obtain a training data set after feature extraction; A feature dimension reduction module is used to perform feature dimension reduction on the training data set after feature extraction to obtain a training data set after dimension reduction; The classifier training module is used to optimize the constructed classifier model based on the training data set after dimensionality reduction and the extreme learning machine algorithm based on quantum information entanglement measurement; The status monitoring module is used to extract features from the real-time data of various operating parameters, reduce the feature dimension, and then input them into the trained classifier model for processing, and output the status monitoring results of the lithium-ion battery energy storage system.

15. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-13.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 13 are implemented.

Images