Multi-sensor battery safety monitoring method and system based on machine learning

Abstract

The invention provides a multi-sensor battery safety monitoring method and system based on machine learning. The method comprises the steps of collecting data through sensors in a battery system; the collected data are stored in a JSON format, and the data are manually annotated; expanding the labeled data by using a generative adversarial network based on robust estimation to form a training sample; performing feature extraction on the training sample through a neural network of an adaptive fluctuation optimization algorithm; carrying out dimensionality reduction on the extracted features based on a self-encoding neural network of manifold mapping learning so as to form a low-dimensional feature space; classifying the data after dimension reduction based on a support vector machine classification algorithm of Landau quantization; and monitoring the safety state of the battery based on the classification result. According to the method, key information is kept in feature dimensionality reduction, meanwhile, the data dimensionality is reduced, the processing efficiency and the operation load are optimized, and the classification precision and stability are improved through accurate quantum state adjustment and data feature combination.

Description

Technical Field

[0001] The present invention relates to the technical field of battery safety monitoring, and in particular to a multi-sensor battery safety monitoring method and system based on machine learning. Background Art

[0002] With the popularization of electronic devices and the rapid development of electric vehicles and electrochemical energy storage industries, battery safety has become an important issue; batteries may have various problems such as overheating, overcharging, and aging during use. If these problems are not discovered and handled in time, they may lead to battery performance degradation and even safety accidents. Therefore, it is crucial to ensure the safe operation of batteries under various working conditions.

[0003] Traditional battery monitoring systems mostly rely on simple sensor data collection and threshold judgment. This method is often incapable of dealing with complex or hidden battery failures. In addition, existing technologies have problems in data collection and processing, such as insufficient sample size, inaccurate feature extraction, and low processing efficiency, which limit the effectiveness and reliability of the system in practical applications. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a multi-sensor battery safety monitoring method based on machine learning.

[0005] The present invention adopts the following technical solution. A first aspect of the present invention provides a multi-sensor battery safety monitoring method based on machine learning, comprising: Collecting data through sensors in the battery system; storing the collected data in a set format, and annotating the collected data according to a preset annotation category; Use a generative adversarial network based on robust estimation to expand the labeled data to form training samples; The obtained training samples are subjected to feature extraction through a neural network of an adaptive fluctuation optimization algorithm, and a feature extraction model is obtained through training; The autoencoder neural network based on manifold mapping learning reduces the dimension of the extracted features to form a low-dimensional feature space, and trains to obtain a feature dimension reduction model; The support vector machine classification algorithm based on Landau quantization is used to classify the reduced-dimensional data and train the classifier model; The trained model is used to process new samples to monitor the battery safety status.

[0006] Preferably, the sensors in the battery system include: a temperature sensor, a humidity sensor, a smoke sensor, a carbon monoxide sensor, a hydrogen sensor and a total organic volatile compound sensor.

[0007] Preferably, the loss function of the generator of the generative adversarial network based on robust estimation is constructed by the following formula (1): (1) Where: is the loss function of the generator; Data generated for the generator The mathematical expectation of represents the uncertainty measurement loss coefficient; is the uncertainty assessment function, Data generated for the generator uncertainty assessment.

[0008] Preferably, extracting features from the obtained training samples by using a neural network of an adaptive fluctuation optimization algorithm comprises: By simulating the propagation and energy distribution characteristics of the wave, the weights and biases in the neural network are iteratively updated according to the following formula (2): (2) Where: are the parameters of the neural network, For the The iteration Neural network parameters, For the The iteration Neural network parameters; Indicates the source of the wave location, Respectively represent the wave source The intensity and initial phase of is the total number of set wave sources; is the angular frequency, which indicates the speed and direction of wave propagation; is the ReLU activation function; For the The fluctuation threshold of each wave source; For the The expected value of the fluctuation threshold of a wave source.

[0009] Preferably, reducing the dimension of the extracted features using an autoencoder neural network based on manifold mapping learning includes: The loss function of the autoencoder neural network is constructed according to the following formula (3): (3) Where: is the preset manifold center; is a parameter that controls the compactness of the manifold; is the regularization coefficient; It is a low-dimensional feature space; is the L2 norm; is the feature data that needs to be reduced in dimension. is the reconstruction error.

[0010] Preferably, the support vector machine classification algorithm based on Landau quantization classifies the reduced-dimensional data including: Initialize the quantum state of the data after dimensionality reduction; Constructing a decision boundary between a quantum state and a hyperplane, wherein the quantized hyperplane is constructed according to the support vector mechanism principle of Landau quantization; Perform fission operations on support vectors based on cluster density and edge distance; For groups of data that exhibit similar characteristics, a fusion operation is performed.

[0011] Preferably, constructing a decision boundary between the quantum state and the hyperplane includes constructing the decision boundary according to the following formula (4): (4) Where: Indicates The sample Features, is the complex amplitude; Indicates the initial sample data vector; express The corresponding quantum state; Indicates the number of features; represents a quantum hyperplane; is the inner product of the quantum state and the hyperplane; is the decision boundary function; is the hyperplane vector; is the inner product of the feature and the hyperplane; is the weight of the hyperplane vector.

[0012] Preferably, the fission operation based on the cluster density and the edge distance includes achieving fission by adjusting the quantum state according to the following formula (5): (5) Where: Indicates quantum states; is the quantum state after the support vector is adjusted, is the fission factor; For the quantum state and the The inverse coefficient of the distance between the quantum states.

[0013] Preferably, for a data group exhibiting similar characteristics, performing a fusion operation comprises performing a fusion operation for a data group exhibiting similar characteristics according to the following formula (6): (6) Where: is the quantum state after the fusion operation; is a collection of quantum states similar to ; is the weight assigned according to the similarity.

[0014] The second aspect of the present invention provides a multi-sensor battery safety monitoring system based on machine learning, which runs the multi-sensor battery safety monitoring method based on machine learning according to the first aspect, including: a data acquisition module, a data marking module, a data expansion module, a feature extraction module, a feature dimension reduction module and a data classification module, characterized in that: A data acquisition module, used to collect data through sensors in the battery system; The data labeling module is used to store the collected data in JSON format and manually label the data. The labeling categories include normal operation, overheating, overcharging and battery aging. The data expansion module is used to expand the labeled data using a generative adversarial network based on robust estimation to form training samples; A feature extraction module, used for extracting features from the obtained training samples through a neural network of an adaptive fluctuation optimization algorithm; A feature dimension reduction module is used to reduce the dimension of the extracted features using an autoencoder neural network based on manifold mapping learning to form a low-dimensional feature space; The data classification module is used to classify the reduced-dimensional data through a support vector machine classification algorithm based on Landau quantization to determine the safety status of the battery.

[0015] A third aspect of the present invention provides a terminal, comprising 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 the multi-sensor battery safety monitoring method based on machine learning according to the first aspect.

[0016] The fourth aspect of 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 the multi-sensor battery safety monitoring method based on machine learning described in the first aspect.

[0017] The beneficial effects of the present invention are that, compared with the prior art, the present invention: 1. The data expanded by generating adversarial networks approximates the real data distribution, enhancing the diversity of the data set and the effectiveness of model training.

[0018] 2. The adaptive volatility optimization algorithm can accurately extract key features in complex data backgrounds, reducing training errors and model volatility.

[0019] 3. The autoencoder structure maintains key information during feature dimensionality reduction while reducing the dimension of the data, optimizing processing efficiency and computational load.

[0020] 4. Landau quantized support vector machine improves classification accuracy and stability through precise quantum state adjustment and data feature merging. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing 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 labor. Among them: Figure 1 A flowchart of a multi-sensor battery safety monitoring method based on machine learning in the present invention; Figure 2 This is a flow chart of feature extraction model training of the present invention.

[0022] Figure 3 This is a flow chart of feature dimension reduction model training of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme 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 this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0026] Embodiment 1 of the present invention provides a multi-sensor battery safety monitoring method based on machine learning, such as Figure 1 The flowchart of the multi-sensor battery safety monitoring method based on machine learning of the present invention is shown, and the method includes: Step 1: Collect data through sensors in the battery system.

[0027] In one embodiment of the present invention, in step 1, the sensors in the battery system include a temperature sensor, a humidity sensor, a smoke sensor, a carbon monoxide sensor, a hydrogen sensor, and a total organic volatile compound sensor. The collected data include temperature, humidity, smoke, carbon monoxide, hydrogen, and total organic volatile compounds.

[0028] Step 2: Store the collected data in a set format and annotate the collected data according to the preset annotation category.

[0029] Preferably but not restrictively, the collected data is stored in JSON format and manually labeled, and the labeled categories include: normal operation, overheating, overcharging and battery aging, a total of 4 categories.

[0030] Specifically, in this embodiment, by way of example but not limitation, one of the data is: { "atime": "2024-06-21T12:00:00Z", "atemperature": 35, "ahumidity": 45, "asmoke": 0.02, "aCO": 0.1, "aH2": 0.05, "aTVOC": 0.03, "adeformation": 0.002 } The fields and attribute values ​​of this data are: atemperature: 45°C (temperature) ahumidity: 50% (humidity) asmoke: 0.05 (smoke index) aCO: 0.2 (carbon monoxide concentration) aH2: 0.07 (hydrogen concentration) aTVOC: 0.04 (total volatile organic matter concentration) adeformation: 0.005 (degree of deformation) It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds.

[0031] Step 3: Use a generative adversarial network based on robust estimation to expand the data annotated in step 2 to form training samples.

[0032] It is understandable that in the technical solution of the present invention, the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The present invention adopts a generative adversarial network algorithm based on robust estimation for data expansion, and the generative adversarial network algorithm based on robust estimation includes two parts: a generator G and a discriminator D. The task of the generator is to generate new data that is as close to the real data as possible, and the task of the discriminator is to distinguish the difference between the generated data and the real data. In order to improve the quality of the generated data, the present invention adopts a strategy based on robust estimation in the loss function of the generator and the discriminator, which helps the algorithm to better handle outliers and noise data in the battery monitoring data.

[0033] Preferably but not restrictively, in step 3, the training process of the generative adversarial network algorithm based on robust estimation is as follows: Step 301: Initialize the network parameters of the generator G and the discriminator. Preferably, but not limited to, the initialization is random distribution subject to Gaussian distribution. The initialization of the generator and the discriminator is expressed by the following formulas (1) and (2): (1) (2) Where: are the parameters of the generator, is the standard deviation of the generator parameters; are the parameters of the discriminator, is the standard deviation of the discriminator parameters; Represents a Gaussian distribution.

[0034] Step 302: Perform iterative training, generator Receive random noise from the latent space , the random noise Following the standard normal distribution, the generator G maps the noise to the data space by learning, imitating the distribution of battery monitoring data, which is expressed as the following formulas (3) and (4): (3) (4) Where: For the generated data, is the input random noise vector, is the identity matrix.

[0035] Furthermore, the generated data is input into the discriminator together with the actual collected battery monitoring data. The discriminator evaluates the authenticity of the data and updates its parameters through back propagation. The training goal is to maximize the distinction between real data and generate data The ability is achieved by minimizing the following loss function (5) (6) (7) Where: The data generated for the generator; is the loss function; is the expectation of the real data, Expectations for generating data; is the distribution of real data, is the distribution of noise.

[0036] Furthermore, based on the feedback from the discriminator, the generator updates its parameters to generate data that is more difficult for the discriminator to recognize. The goal is to make the discriminator It is impossible to distinguish between real data and generated data, which is achieved by minimizing the following loss function, expressed as follows: (8) (9) (10) Where: is the uncertainty measurement loss term; is the uncertainty metric loss coefficient, preferably but not limited to, Set to 3; is the uncertainty assessment function, then Data generated for the generator Conduct uncertainty assessments; is the matrix trace operation function.

[0037] Step 303: In each iteration, the quality of the generated data in the current training batch is evaluated, and the learning rates of the generator and the discriminator are dynamically adjusted. The adjustment method is expressed by the following formulas (11) and (12): (11) (12) Where: and are the new and old learning rates respectively; is the learning rate adjustment factor, preferably but not limited to, Set to 0.3; and are the previous and current loss function values ​​respectively; is the loss function change estimator.

[0038] Step 304: Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. Preferably, but not limiting, the preset stop iteration condition is reaching a preset maximum number of iterations, for example but not limited to, the preset maximum number of iterations is set to 1000 times.

[0039] Step 305: Use the similarity metric to evaluate the quality of the generated data and adjust the training strategy accordingly, as shown in the following formulas (13) and (14): (13) (14) Where: is the quality assessment function, which is realized by calculating the similarity between the generated data and the data in the real data set, that is, formula (14); To generate data; A function to adjust the training strategy based on the quality assessment; Represents the similarity measurement function, and the measurement method is implemented based on the Euclidean distance calculation method; Indicates the first samples, is the total number of samples.

[0040] When the evaluation value of the quality evaluation function is greater than a preset threshold, the data is retained; otherwise, the generated data is discarded.

[0041] After the data expansion model training is completed, the trained data expansion model is used to increase the number of samples. Preferably, but not restrictively, if the original collected samples are 800, and the data expansion model expands and generates 200 samples, then the expanded data set contains 1000 samples.

[0042] Step 4: Extract features from the training samples obtained in step 3 through the neural network of the adaptive wave optimization algorithm.

[0043] Preferably, but not restrictively, the expanded data is input into the feature extraction model for training the feature extraction model. The present invention adopts a neural network feature extraction algorithm based on adaptive fluctuation optimization, which is inspired by the fluctuation phenomenon. The fluctuation presents the characteristics of periodic change and energy transfer, and can transmit and transform energy forms in different media. The adaptive fluctuation optimization algorithm iteratively updates the weights and biases in the network by simulating the propagation and energy distribution characteristics of the fluctuation. Each parameter update is not only affected by the current gradient, but also by the cumulative influence of the fluctuation of the previous parameter update, which is similar to the fluctuation influence caused by the superposition of wave sources, making the parameter update process smoother and more adaptive, effectively avoiding the oscillation and local minimum problems in the traditional gradient descent method.

[0044] Preferably, but not limiting, in step 4, Figure 2The figure shows the feature extraction model training flow chart. The training process of the neural network feature extraction algorithm based on adaptive fluctuation optimization includes: Step 401: Define the neural network structure, including: the number of hidden layers, the number of neurons in each hidden layer, and the activation function.

[0045] Preferably but not restrictively, the number of hidden layers of the neural network is 2, the first hidden layer includes 100 neurons, the second hidden layer includes 50 neurons, and the activation function is the ReLU activation function.

[0046] Further, Represents the parameters of the neural network, including weights and bias . Set the initial parameters of the neural network , including the weights of the neural network and bias , and initialize the wave source Location ,strength and the first phase , set the total number of wave sources to , then the initialization is expressed by the following formulas (15) and (16): (15) (16) Where: are the initial parameters of the neural network, They are the initial weights and initial biases of the neural network respectively; For the wave source, by position ,strength and the first phase composition; Usually chosen as the initial position of the parameter, and Randomly set its initial value.

[0047] Step 402: Calculate the loss through forward propagation , using the parameter On input Perform forward propagation on it and calculate the output and the target value In contrast, the loss is calculated as follows: (17) (18) Where: Represents a network structure, and the output layer of the network structure uses a preset Softmax classifier to obtain a predicted label; is the cross entropy loss.

[0048] Step 403: Calculate the gradient of each parameter based on the loss function , expressed as the following formulas (19)-(23): (19) (20) (twenty one) (twenty two) (twenty three) Where: is the gradient of each parameter; Weight The gradient of , using the chain method to calculate the gradient, assuming that the network output After the Softmax function, is the output of Softmax, It is the linear output before the Softmax layer.

[0049] Step 404: Generate a wave source at the corresponding parameter position according to the direction and magnitude of the gradient. The strength is updated to , expressed as the following formula (24): (twenty four) Where: is the learning rate, preferably but not limited to, Set to 0.01; is the attenuation coefficient; is the amplification factor.

[0050] Step 405: The fluctuations of all wave sources are superimposed in the parameter space. The update amount of each parameter is determined by the superposition result of the fluctuations of all wave sources at that point. The parameter update method is expressed by the following formulas (25)-(27): (25) (26) (27) Where: are the parameters of the neural network, including weights and bias , For the The iteration Neural network parameters, For the The iteration Neural network parameters; Indicates the source of the wave location, Respectively represent the wave source The strength and initial phase of To set the total number of wave sources, is the wave number; is the angular frequency, which indicates the speed and direction of wave propagation; is the ReLU activation function; For the The fluctuation threshold of each wave source is dynamically adjusted with the layer sensitivity index. is the initial fluctuation threshold, is the adjustment factor; preferably, is set to 0.1, Set to 3, For the The expected value of the fluctuation threshold of each wave source; For the The layer sensitivity index of a layer is calculated based on the norm of the gradient output by the layer to measure the contribution of the layer to the overall network error. It is The norm of the layer parameter gradients, is the number of network layers.

[0051] Step 406: Dynamically adjust the attenuation coefficient of each wave source according to the loss reduction of this iteration and the amplification factor , which can be expressed as the following formulas (28) and (29): (28) (29) Where: are the attenuation coefficient and amplification coefficient of the previous iteration respectively; are respectively the attenuation coefficient adjustment factor and the amplification coefficient adjustment factor. Preferably, but not restrictively, the parameters and Set to 0.1 and 0.05 respectively; is the change in loss.

[0052] Step 407: Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. Preferably, but not limiting, the preset stop iteration condition is reaching a preset maximum number of iterations, for example but not limited to, the preset maximum number of iterations is set to 1000 times.

[0053] Step 5: The autoencoder neural network based on manifold mapping learning reduces the dimensionality of the extracted features to form a low-dimensional feature space.

[0054] Preferably, but not restrictively, the data after feature extraction is input into a feature dimensionality reduction model for training the feature dimensionality reduction model, and an autoencoder neural network based on manifold mapping learning is used for feature dimensionality reduction. The autoencoder neural network not only learns the intrinsic structure of the data in the encoding stage, but also attempts to reconstruct the input through the decoding process to ensure that the features after dimensionality reduction can still retain key information. The autoencoder neural network model structure consists of an encoder and a decoder. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder reconstructs the original input data from this low-dimensional space. The present invention uses Riemann factors based on manifold learning to optimize the representation of the feature space. By utilizing the geometric properties on the Riemann manifold, the local structure of the data can be better understood and more accurate data reconstruction can be achieved.

[0055] Preferably, but not limiting, such as Figure 3 The figure shows the training flow chart of the feature dimensionality reduction model. The training process of the autoencoder neural network algorithm based on manifold mapping learning includes: Step 501: Initialize network parameters. The present invention uses Gaussian distribution to initialize the weights of the autoencoder. and bias The parameters are expressed as follows: (30) Where: Indicates The weight of the layer; represents the standard deviation; preferably, , is the number of nodes in the input layer.

[0056] Step 502: During the encoding process, the input data is mapped to a low-dimensional feature space through the encoder; during the decoding process, the decoder attempts to reconstruct the original data. Specifically, through the autoencoder, the data Mapped to a low-dimensional feature space The method is expressed as follows: (31) (32) (33) Where: It is a low-dimensional feature space; As the decoder attempts to reconstruct the original input, is the activation function; and are the weights and biases of the encoder, respectively. and are the decoder weights and biases.

[0057] Preferably, but not restrictively, the weights of the encoder It is pre-set by principal component analysis, that is, formula (33), is from the covariance matrix Before getting The matrix consists of eigenvectors, is the corresponding diagonal matrix of eigenvalues.

[0058] Step 503: In the encoded low-dimensional space, use the Riemann factor To achieve manifold learning, the Riemann factor constraint encoding features The method is expressed as follows: (34) Where: is the preset manifold center, is a parameter that controls the compactness of the manifold, preferably but not limitingly, Set as the mean vector of the low-dimensional space, Set to 0.1; is the L2 norm.

[0059] Step 504: The loss function not only considers the reconstruction error, but also the Riemann manifold constraint. The reconstruction error ensures that the data can be effectively restored. The manifold constraint strengthens the geometric structure of the feature space. The loss function is expressed as follows: (35) (36) Where: is the regularization coefficient, preferably but not limited to, Set to 0.3; is the reconstruction error; yes In the The value of the dimension, is the preset manifold center at The value of the dimension.

[0060] Step 505: Based on the loss function, adjust the network parameters through the back propagation algorithm to optimize the performance of the encoder and decoder. As an example, the gradient descent method is used to update the parameters as shown in the following formulas (37)-(41): (37) (38) (39) (40) (41) Where: is the updated weight parameter, is the weight parameter before updating; is the learning rate, represents the number of iterations, preferably but not limited to, Set to 0.05; is the derivative of the Sigmoid activation function, For vector is a diagonal matrix with diagonal elements, for The transpose of .

[0061] It can be understood that formula (37) represents the gradient descent method to update the parameters, and formula (38) gives the weights The calculation method of It is calculated by the error back propagation method, that is, formula (39); Through the decoder weights The transposed representation of is formula (40); since yes and The result obtained by the activation function after the linear combination of It is expressed in the form of formula (41).

[0062] Step 506: Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. Preferably, but not limiting, the preset stop iteration condition is reaching a preset maximum number of iterations, for example but not limited to, the preset maximum number of iterations is set to 1000 times.

[0063] Step 6: Classify the reduced-dimensional data using the support vector machine classification algorithm based on Landau quantization.

[0064] Preferably, but not restrictively, the reduced-dimensional data is input into a classifier to train the classifier model. The present invention adopts a support vector machine classification algorithm based on Landau quantization. The traditional support vector machine maximizes the margin between different categories of data by finding a hyperplane. The present invention optimizes the selection of support vectors and the definition of classification hyperplanes through a quantum state conversion strategy, and further optimizes the classification process by adopting the concepts of fission and fusion. During the training process, the model dynamically adjusts the support vectors through a fission mechanism, that is, splits in the data space to form a more detailed decision boundary. At the same time, the fusion mechanism allows the model to merge similar features in multiple data set partitions to enhance the model's generalization ability for new data. This two-way dynamic adjustment mechanism provides the model with higher flexibility and accuracy.

[0065] Preferably, but not restrictively, the training process of the support vector machine classification algorithm based on Landau quantization is as follows: Step 601: Initialize the quantum state of the training data. Each sample is represented by a quantum bit. The quantum bit has multiple possible states, corresponding to different classification labels. Specifically, let the initial The sample data vector is ,and Indicates The sample features, converted into quantum states It is expressed as the following formula (42): (42) Where: is the complex amplitude, satisfying ; Features The corresponding quantum ground state.

[0066] Step 602: Construct an initial quantized hyperplane using the Landau quantization principle, and use the probability amplitude of the quantum state to evaluate its relative position with the hyperplane to determine the initial support vector, which is expressed as the following formulas (43)-(45): (43) (44) (45) Where: is the quantum hyperplane, that is, formula (43), is the hyperplane vector; is the decision boundary between the quantum state and the hyperplane, that is, formula (44), is the inner product of the quantum state and the hyperplane, is the decision boundary function, is the inner product of the feature and the hyperplane; is the weight of the hyperplane vector, determined by principal component analysis, that is, formula (45), is the data covariance matrix eigenvalues, is the data covariance matrix feature values.

[0067] Step 603: In each iteration, check the distance and angle between each support vector and other data points, and perform fission operations according to cluster density and edge distance to refine the decision boundary, which is expressed as the following formulas (46) and (47): (46) (47) Where: is the quantum state after the support vector is adjusted, that is, formula (47) realizes the fission process by adjusting the quantum state of each support vector. Indicates quantum states; is a fission factor, preferably but not limited to, Set to 3; For the quantum state and the The inverse coefficient of the distance between quantum states is data points and The distance between data points is dynamically adjusted, i.e., formula (48); For data points and data points The square of the Euclidean distance between is a small constant, preferably but not limitingly, Set to 0.0001.

[0068] Step 604: For data groups that exhibit similar characteristics, a fusion operation is performed to merge the representations of these data points to enhance the model's response to these characteristics. The quantum state merging operation of similar data points is expressed by the following formulas (48) and (49): (48) (49) Where: is the quantum state after the fusion operation; is a collection of quantum states similar to ; is the weight assigned according to similarity, which is calculated based on the similarity and the local density of the data points, that is, formula (49), It is a data point To Collection The average distance to other points in .

[0069] Step 605: Adjust the state of the qubit through quantum gate operation to optimize the accuracy of the classification result, which is expressed as the following formulas (50) and (51): (50) (51) Where: is the rotation angle of the quantum gate, which is used to adjust the state of the quantum bit and is calculated using formula (51). is a specific Hamiltonian, preferably but not restrictively, using a Pauli-Z gate to realize phase rotation in quantum computing, Controls the angle of rotation.

[0070] Step 606: Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. Preferably, but not limiting, the preset stop iteration condition is reaching a preset maximum number of iterations, for example but not limited to, the preset maximum number of iterations is set to 1000 times.

[0071] Step 7: Use the trained model to process new samples to monitor the battery safety status.

[0072] Preferably but not restrictively, using a trained model to process new samples to monitor the battery safety status includes: the input data of the sample is first sent to a trained feature extraction model to obtain a first feature vector; further, the first feature vector is input into a trained feature dimension reduction model to obtain a second feature vector; further, the second feature vector is input into a trained classifier model to obtain a classification category, which is the battery safety status; in this embodiment, the categories include: normal operation, overheating, overcharging, and battery aging, a total of 4 categories.

[0073] Embodiment 2 of the present invention provides a multi-sensor battery safety monitoring system based on machine learning, including: a data acquisition module, a data marking module, a data expansion module, a feature extraction module, a feature dimension reduction module and a data classification module; A data acquisition module, used to collect data through sensors in the battery system; The data labeling module is used to store the collected data in JSON format and manually label the data. The labeling categories include normal operation, overheating, overcharging and battery aging. A data expansion module is used to expand the data annotated in step 2 using a generative adversarial network based on robust estimation to form training samples; A feature extraction module, used for extracting features from the training samples obtained in step 3 through a neural network of an adaptive wave optimization algorithm; A feature dimension reduction module is used to reduce the dimension of the extracted features using an autoencoder neural network based on manifold mapping learning to form a low-dimensional feature space; The data classification module is used to classify the reduced-dimensional data through a support vector machine classification algorithm based on Landau quantization to determine the safety status of the battery.

[0074] Embodiment 3 of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the multi-sensor battery safety monitoring method based on machine learning according to Embodiment 1 is implemented.

[0075] Embodiment 4 of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the multi-sensor battery safety monitoring method based on machine learning according to Embodiment 1.

[0076] 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 multi-sensor battery safety monitoring method based on machine learning, characterized in that: include: Collect data through sensors in the battery system; The collected data is stored in a set format and annotated according to a preset annotation category; Use a generative adversarial network based on robust estimation to expand the labeled data to form training samples; The obtained training samples are subjected to feature extraction through a neural network of an adaptive fluctuation optimization algorithm, and a feature extraction model is obtained through training; The autoencoder neural network based on manifold mapping learning reduces the dimension of the extracted features to form a low-dimensional feature space, and trains to obtain a feature dimension reduction model; The support vector machine classification algorithm based on Landau quantization is used to classify the reduced-dimensional data and train the classifier model; The trained model is used to process new samples to monitor the battery safety status.

2. The multi-sensor battery safety monitoring method based on machine learning according to claim 1 is characterized in that: Sensors in the battery system include: temperature sensor, humidity sensor, smoke sensor, carbon monoxide sensor, hydrogen sensor and total organic volatile compound sensor.

3. The multi-sensor battery safety monitoring method based on machine learning according to claim 1 is characterized in that: The loss function of the generator of the generative adversarial network based on robust estimation is constructed by the following formula (1): (1) Where: is the loss function of the generator; Data generated for the generator The mathematical expectation of represents the uncertainty measurement loss coefficient; is the uncertainty assessment function, Data generated for the generator uncertainty assessment.

4. The multi-sensor battery safety monitoring method based on machine learning according to claim 1 is characterized in that: The feature extraction of the obtained training samples by the neural network of the adaptive fluctuation optimization algorithm includes: By simulating the propagation and energy distribution characteristics of the wave, the weights and biases in the neural network are iteratively updated according to the following formula (2): (2) Where: are the parameters of the neural network; For the The iteration Neural network parameters; For the The iteration Neural network parameters; Indicates the source of the wave location, Respectively represent the wave source The intensity and initial phase of is the total number of set wave sources; is the angular frequency, which indicates the speed and direction of wave propagation; is the ReLU activation function; For the The fluctuation threshold of each wave source; For the The expected value of the fluctuation threshold of a wave source.

5. The multi-sensor battery safety monitoring method based on machine learning according to claim 1 is characterized in that: The dimensionality reduction of the extracted features using the autoencoder neural network based on manifold mapping learning includes: The loss function of the autoencoder neural network is constructed according to the following formula (3): (3) Where: is the preset manifold center; is a parameter that controls the compactness of the manifold; is the regularization coefficient; It is a low-dimensional feature space; is the L2 norm; is the feature data that needs to be reduced in dimension. is the reconstruction error.

6. The multi-sensor battery safety monitoring method based on machine learning according to claim 1 is characterized in that: The support vector machine classification algorithm based on Landau quantization is used to classify the reduced-dimensional data, including: Initialize the quantum state of the data after dimensionality reduction; Constructing a decision boundary between a quantum state and a hyperplane, wherein the quantized hyperplane is constructed according to the support vector mechanism principle of Landau quantization; Perform fission operations on support vectors based on cluster density and edge distance; For groups of data that exhibit similar characteristics, a fusion operation is performed.

7. The multi-sensor battery safety monitoring method based on machine learning according to claim 6 is characterized in that: Constructing the decision boundary between the quantum state and the hyperplane includes constructing the decision boundary according to the following formula (4): (4) Where: Indicates The sample Features, is the complex amplitude; Indicates the initial sample data vector; express The corresponding quantum state; Indicates the number of features; represents a quantum hyperplane; is the inner product of the quantum state and the hyperplane; is the decision boundary function; is the hyperplane vector; is the inner product of the feature and the hyperplane; is the weight of the hyperplane vector.

8. A multi-sensor battery safety monitoring method based on machine learning according to claim 6 or 7, characterized in that: The fission operation based on cluster density and edge distance includes adjusting the quantum state to achieve fission according to the following formula (5): (5) Where: Indicates quantum states; is the quantum state after the support vector is adjusted; is the fission factor; For the quantum state and the The inverse coefficient of the distance between the quantum states.

9. The multi-sensor battery safety monitoring method based on machine learning according to claim 8, characterized in that: For a data group exhibiting similar characteristics, performing a fusion operation includes performing a fusion operation for a data group exhibiting similar characteristics according to the following formula (6): (6) Where: is the quantum state after the fusion operation; is a collection of quantum states similar to ; is the weight assigned according to the similarity.

10. A multi-sensor battery safety monitoring system based on machine learning, running the multi-sensor battery safety monitoring method based on machine learning according to any one of claims 1 to 9, comprising: The data collection module, the data labeling module, the data expansion module, the feature extraction module, the feature dimension reduction module and the data classification module are characterized by: A data acquisition module, used to collect data through sensors in the battery system; The data labeling module is used to store the collected data in JSON format and manually label the data. The labeling categories include normal operation, overheating, overcharging and battery aging. The data expansion module is used to expand the labeled data using a generative adversarial network based on robust estimation to form training samples; A feature extraction module, used for extracting features from the obtained training samples through a neural network of an adaptive fluctuation optimization algorithm; A feature dimension reduction module is used to reduce the dimension of the extracted features using an autoencoder neural network based on manifold mapping learning to form a low-dimensional feature space; The data classification module is used to classify the reduced-dimensional data through a support vector machine classification algorithm based on Landau quantization to determine the safety status of the battery.

11. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the multi-sensor battery safety monitoring method based on machine learning according to any one of claims 1-9.

12. 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 multi-sensor battery safety monitoring method based on machine learning described in any one of claims 1-9 are implemented.

Images