Battery early warning system based on migration adversarial network model and control method thereof

Abstract

The invention discloses a battery early warning system based on a migration adversarial network model and a control method thereof. The method comprises the steps of data acquisition, data preprocessing, construction of the migration adversarial network model, adversarial training, early warning decision mechanism and the like. Through the mode, according to the battery early warning system based on the migration adversarial network model and the control method of the battery early warning system, the health condition of a battery is monitored in real time and potential battery faults are predicted and diagnosed through an innovative battery management system real-time monitoring and early warning technical architecture, so that the reliability and safety of a battery system are ensured; therefore, the capacity of capturing complex dynamic changes and potential faults in the battery is improved, accurate control over the battery faults is achieved, the accuracy and real-time performance of early warning of the battery faults are improved, and the generalization capacity of the system is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management systems, and particularly to a battery warning system based on a transfer adversarial network model and a control method therefor. Background Art

[0002] With the transformation of the global energy structure and the rapid development of electric vehicles, the battery, as the core component of energy storage, plays an important role in fields such as electric vehicles and energy storage power stations. However, during the use of the battery, its performance may decline or even cause safety accidents due to reasons such as overcharging, over-discharging, and abnormal temperature. Battery failures may lead to serious safety accidents. Therefore, it is crucial to monitor the battery status in real time and give early warnings of failures. Its importance is mainly reflected in the following aspects: ① improving safety; ② extending the battery life; ③ reducing losses; ④ improving system reliability.

[0003] Existing battery warning systems mostly rely on traditional machine learning algorithms. However, traditional battery monitoring methods mainly rely on data such as voltage, current, and temperature collected by sensors, and then predict the battery status through simple threshold judgments. This makes the traditional methods have great limitations in capturing the complex dynamic changes and potential failures inside the battery when facing the non-linear and dynamic changes of the battery status, such as slow warning speed and low accuracy, and it is often difficult to give warnings in a timely and accurate manner. Summary of the Invention

[0004] To solve the above technical problems, a technical solution adopted by the present invention is:

[0005] Provide a control method for a battery warning system based on a transfer adversarial network model, characterized in that the steps include:

[0006] (1) Collect the working state data of the battery in real time and preprocess the working state data; wherein, the working state data includes voltage, current, temperature, and the number of charge and discharge cycles;

[0007] (2) Construct a transfer adversarial network model and perform adversarial training on the transfer adversarial network model using the historical working state data of the battery;

[0008] (3) Use the transfer adversarial network model to make a warning decision;

[0009] (3.1) Obtain the healthy state feature vector and the real-time monitoring feature vector generated by the transfer adversarial network model, and fuse the healthy state feature vector with the obtained real-time monitoring feature vector to form a comprehensive feature vector;

[0010] (3.2) State assessment: Based on a neural network, the fused comprehensive feature vector is used to perform real-time assessment of the battery health state, so as to obtain the current battery health degree and the current battery aging degree;

[0011] (3.2.1) The current battery health degree H is:

[0012] H = e(W2·ReLU(W1·F F +b1)+b2),

[0013] where H is a value between 0 and 1, representing the current health status of the battery; W1 is the weight matrix connecting the input layer and the first hidden layer in the neural network, W2 is the weight matrix connecting the first hidden layer and the output layer in the neural network; b1 is the bias vector of the hidden layer in the first neural network, b2 is the bias vector of the output layer in the neural network; ReLU() is the ReLU activation function; e is the activation function of the output layer; F F is the comprehensive feature vector;

[0014] (3.2.1) The current battery aging degree A is:

[0015] A = W2·ReLU(W1·F F +b1)+b2;

[0016] (3.3) Evaluate according to the current battery health degree, the current battery aging degree and the preset warning threshold;

[0017] (3.3.1) The preset health degree threshold of the battery is θ. If the value of the current battery health degree is less than the health degree threshold θ, it is determined that the battery has an abnormality or is about to enter an abnormal state, and a warning is triggered;

[0018] (3.3.2) The preset aging degree threshold of the battery is θ'. If the value of the current battery aging degree is higher than the aging degree threshold θ', it is determined that the battery needs maintenance or replacement, and a warning is triggered;

[0019] (3.4) Generate corresponding warning signals according to the evaluation results for warning reminder.

[0020] In a preferred embodiment of the present invention, the step of preprocessing the working state data includes:

[0021] (1.1) Clean the collected working state data to remove outliers and noise;

[0022] (1.2) Perform normalization processing on the working state data; wherein, each type of working parameter is normalized separately.

[0023] In a preferred embodiment of the present invention, when constructing the migration adversarial network model, the generator is trained using the historical working state data of the battery, enabling the generator to learn and simulate the parameter distribution of the battery under different working states;

[0024] Among them, the training formula of the generator is:

[0025]

[0026] The training formula of the discriminator is:

[0027]

[0028] Among them, G is the generator, D is the discriminator, X is the real data, pdata(X) is the probability distribution of the real data, D(X) is the judgment result of the discriminator on the real data X, L G is the loss function of the generator, is the expected value, Z is the noise data, pz(Z) is the probability distribution of the noise data Z, G(Z) is the virtual data generated by the generator, and D(G(Z)) is the judgment result of the discriminator on the virtual data G(Z) generated by the generator.

[0029] In a preferred embodiment of the present invention, in step (3.1), the weighted summation method is used for fusion:

[0030] F F =α·F G +(1 - α)·F R ,

[0031] Among them, F F is the comprehensive feature vector, F G is the health state feature, F R is the real-time monitoring feature, α is the weight of the generated health state feature, and 1 - α is the weight of the real-time monitoring feature.

[0032] In a preferred embodiment of the present invention, in step (3.1), the generation steps of the health state feature vector include:

[0033] (3.1.1) Input fusion: The noise data Z and the historical working state data Xhist are fused to form an input vector: Input = [Z; X hist ;

[0034] (3.1.2) Multi-layer neural network processing: The input layer receives the fused input vector; the hidden layer performs a non-linear transformation, and the output of each layer is expressed as: H i = ReLU(W i ·H i-1 + b i ), where Wi and b i are the weight matrix and bias vector of the i-th layer respectively, and H i-1 is the output of the previous layer. Through the processing of multiple hidden layers, features that can reflect the battery health status are gradually extracted;

[0035] (3.1.3) Generation of the health status feature vector:

[0036] The health status feature vector generated by the output layer is:

[0037] FG = W out ·H last + b out ,

[0038] where W out and b out are the weight matrix and bias vector of the output layer respectively, and H last is the output of the last hidden layer;

[0039] (4) Generation and update of the health status feature vector: The generator periodically generates the health status feature vector according to a preset frequency, or dynamically updates the generated health status feature vector each time it receives the latest input data to reflect the real-time health status of the battery.

[0040] In a preferred embodiment of the present invention, in step (3.1), the comprehensive feature vector is not a single value and includes the mean value of voltage, the variance of current, and the change in temperature.

[0041] In a preferred embodiment of the present invention, the structure of the neural network includes an input layer, a hidden layer, and an output layer.

[0042] In a preferred embodiment of the present invention, in step (3.4), after a warning is triggered, corresponding warning and maintenance measure signals are sent according to different levels of the fault type and abnormal state.

[0043] In a preferred embodiment of the present invention, the warning and maintenance measure signals include sound signals and / or color indication models representing different levels of abnormal states.

[0044] A battery warning system based on a transfer adversarial network model, which includes:

[0045] Data acquisition module: used to collect key working state data of the battery in real time;

[0046] Data preprocessing module: used to preprocess the collected working state data and normalize the data to ensure that the data is within a unified numerical range;

[0047] Transfer adversarial network module: The transfer adversarial network model includes a generator and a discriminator. The generator is used to generate virtual working state data samples that are close to the real battery state, and the discriminator is used to distinguish between real working state data samples and the virtual working state data samples generated by the generator;

[0048] Health state evaluation module: It is used to evaluate the current health state of the battery;

[0049] Early warning module: It is used to trigger an early warning mechanism according to the evaluation result and generate an early warning signal to prompt possible battery failures.

[0050] In a preferred embodiment of the present invention, it further includes a communication module and a power management module.

[0051] Communication module: It is used for data exchange with other system components;

[0052] Power management module: It is used to provide stable power supply for the whole system.

[0053] The beneficial effects of the present invention are as follows: Through the innovative real-time monitoring and early warning technology architecture of the battery management system, the health status of the battery is monitored in real time, potential battery failures are predicted and diagnosed, the reliability and safety of the battery system are ensured, thereby improving the ability to capture the complex dynamic changes and potential failures inside the battery, and then achieving precise control of battery failures, improving the accuracy and real-time performance of battery failure early warning, and enhancing the generalization ability of the system. Brief description of the drawings

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings, where:

[0055] Figure 1 It is the overall schematic diagram of the control method of a battery early warning system based on a transfer adversarial network model of the present invention;

[0056] Figure 2 It is the specific process schematic diagram of the control method of a battery early warning system based on a transfer adversarial network model of the present invention;

[0057] Figure 3 It is the flowchart of data preprocessing in the embodiment of the present invention;

[0058] Figure 4 It is the flowchart of constructing a transfer adversarial network in the embodiment of the present invention;

[0059] Figure 5 It is a flowchart for training a fault prediction model in an embodiment of the present invention. Detailed implementation manners

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0062] Please refer to Figures 1-5 , the embodiments of the present invention include:

[0063] A battery warning system based on a transfer adversarial network model, which includes:

[0064] Data acquisition module: used to collect key working state data of the battery in real time, such as current, voltage, temperature, etc.

[0065] A variety of data of the battery system is crucial for subsequent real-time monitoring and fault diagnosis. Therefore, in order to improve the accuracy of subsequent calculations, as many parameters as possible in the historical data of the battery system can be obtained.

[0066] Data preprocessing module: used to clean the collected battery data, remove noise and outliers, and normalize the data to ensure that the data is within the same numerical range for subsequent processing.

[0067] Transfer adversarial network module, including a generator and a discriminator. The generator is responsible for generating virtual working state data samples close to the real battery state, and the discriminator is responsible for distinguishing between real working state data samples and virtual working state data samples generated by the generator to improve the adaptability and generalization ability of the model to different battery types, which helps to improve the accuracy of real-time monitoring and fault warning information of the battery system.

[0068] Health status evaluation module: used to evaluate based on the health status features output by the transfer adversarial network to determine the current health status of the battery.

[0069] Warning module: used to trigger a warning mechanism according to the evaluation result and generate a warning signal to prompt possible battery failures.

[0070] Communication module: used for data exchange with other system components to ensure the real-time and reliability of data transmission.

[0071] Power management module: used to provide stable power supply for the entire system to ensure that the system can operate stably under high load.

[0072] A control method for a battery warning system based on a transfer adversarial network model, the steps of which include:

[0073] Step S1, data acquisition.

[0074] Collect the working state data of the battery in real time. Among them, the working state data includes voltage, current, temperature, and charge and discharge cycle times.

[0075] Step S2, preprocess the data.

[0076] Preprocess the collected working state data, and the specific steps of the preprocessing include

[0077] Step S201, clean the collected working state data to remove outliers and noise.

[0078] Step S202, perform normalization processing on the working state data to improve the data quality, provide accurate input for subsequent model training and warning decision-making, and thus improve the accuracy of the model.

[0079] Further preferably, each type of working parameter is normalized separately to better maintain the characteristics of each type of parameter.

[0080] Further preferably, the calculation formula for the normalization processing is:

[0081] Among them, x is the original working state data, μ is the mean of the working state data, δ is the standard deviation of the working state data, and x' is the normalized data.

[0082] Step S3, construct a transfer adversarial network model.

[0083] When constructing the transfer adversarial network model, use the obtained historical working state data of the battery to train the generator, that is, train the generator with the working state data such as voltage, current, temperature, and charge and discharge cycle times together, so that the generator can learn and simulate the parameter distribution of the battery under different working states, in order to improve the accuracy of the generation of virtual working state data.

[0084] Among them, the training formula of the generator is:

[0085]

[0086] The training formula of the discriminator is:

[0087]

[0088] Among them, G is the generator, D is the discriminator, X is the real data, pdata(X) is the probability distribution of the real data, D(X) is the judgment result of the discriminator on the real data X, LG is the loss function of the generator, is the expected value, Z is the noise data, pz(Z) is the probability distribution of the noise data Z, G(Z) is the virtual data generated by the generator, and D(G(Z)) is the judgment result of the discriminator on the virtual data G(Z) generated by the generator.

[0089] Step S4, perform adversarial training on the transfer adversarial network model.

[0090] Through adversarial training, the discriminator can continuously improve its ability to distinguish between real samples and virtual samples. At the same time, the generator is also continuously optimized and improved to generate more realistic virtual samples.

[0091] Step S5, use the transfer adversarial network model for early warning decision-making.

[0092] Step S501, data fusion: Obtain the health status feature vector and real-time monitoring feature vector generated by the transfer adversarial network model, and fuse the health status feature vector with the obtained real-time monitoring feature vector to form a comprehensive feature vector.

[0093] The health status feature vector is not a single value. The health status feature vector is generated by the generator according to the input noise data and historical working status data to reflect the health status of the battery. The generator processes the input noise data and historical working status data through a series of neural network operations. Since the system needs to monitor the health status of the battery in real time, the health status feature vector needs to be generated on time. For example, it is generated once every certain period (several minutes or seconds).

[0094] The specific generation steps of the health status feature vector are as follows:

[0095] (1) Input fusion:

[0096] Fuse the noise data Z and the historical working status data Xhist to form an input vector:

[0097] Input = [Z; X hist .

[0098] (2) Multilayer neural network processing:

[0099] The input layer receives the fused input vector; the hidden layer performs a non-linear transformation, and the output of each layer is expressed as: H i = ReLU(W i ·H i-1 + b i )

[0100] where, W i and b iThey are the weight matrix and bias vector of the i-th layer, respectively, and H i-1 is the output of the previous layer; through the processing of multiple hidden layers, features that can reflect the battery health status are gradually extracted.

[0101] (3) Generation of the health status feature vector:

[0102] The output layer generates the health status feature vector FG, and the formula is:

[0103] FG = W out ·H last + b out

[0104] where W out and b out are the weight matrix and bias vector of the output layer, and H last is the output of the last hidden layer.

[0105] (4) Periodic generation and update:

[0106] The generator needs to generate the health status feature vector regularly, usually preset to generate once every few minutes or seconds; each time the health status feature vector is generated, the generator will dynamically update the output according to the latest input data (noise data and historical working status data) to reflect the real-time health status of the battery.

[0107] The real-time monitoring feature vector is the feature vector obtained by processing and analyzing the working status data, and it also needs to be generated according to the preset frequency. Among them, the method of processing and analyzing the working status data belongs to common knowledge and is a common technical means in the field of battery management systems. Here, one of the methods is used to illustrate specifically.

[0108] The generation process of the real-time monitoring feature vector is:

[0109] (1) Data collection: Data preprocessing (the same as the content in steps S1 and S2);

[0110] (2) Feature extraction: Statistical analysis method is used to extract features that can reflect the current state of the battery from the preprocessed data, such as the mean value of voltage, the variance of current, and the change of temperature. The calculation formulas are as follows:

[0111] Mean value of voltage: where V i is the i-th voltage data point, and n is the number of data points;

[0112] Variance of current: where I i is the i-th current data point, and I mean is the mean value of current;

[0113] Change in temperature: where T current is the current temperature, and T previous is the temperature at the previous time point.

[0114] (3) Feature vector generation: Combine the extracted features into a vector, which is the real-time monitoring feature vector.

[0115] The fusion of data is triggered in real time according to the generation of real-time monitoring features. Whenever a new real-time monitoring feature vector is generated, it will be fused with the healthy state feature vector generated by the generator, that is, it is fused once every certain / preset time (a few minutes or seconds). Among them, the fusion is carried out in the model.

[0116] The comprehensive feature vector is not a single value, and it contains information in multiple dimensions, such as the mean value of voltage, the variance of current, the change in temperature, etc.

[0117] Further preferably, the weighted summation method is used for fusion, and the formula is:

[0118] F F = α·F G +(1 - α)·F R .

[0119] where F F is the comprehensive feature vector, F G is the healthy state feature vector, F R is the real-time monitoring feature vector, α is the weight of the healthy state feature vector, and 1 - α is the weight of the real-time monitoring feature vector. The weights are unified to simplify the complexity of the model.

[0120] The formula for determining the weight is as follows:

[0121] α = argmin α Loss(F F , Y true )

[0122] where Loss is the loss function, which is used to measure the difference between the comprehensive feature vector F F and the true health state label Y true , and argmin means to find the value of α that minimizes the loss function. Among them, the true health state label is preset and only needs to be directly obtained during the calculation.

[0123] Step S502, State evaluation: Use a neural network to perform real-time evaluation of the battery health state based on the fused comprehensive feature vector.

[0124] Real-time evaluation is carried out immediately after each new set of working status data is collected, that is, the evaluation frequency can be the same as the frequency of working status data collection.

[0125] Further preferably, the structure of the neural network includes an input layer, a hidden layer, and an output layer.

[0126] Input layer: The number of nodes is the same as the dimension of the comprehensive feature vector.

[0127] Hidden layer: There can be multiple hidden layers, and each layer uses the ReLU activation function.

[0128] Output layer: The number of nodes can be determined according to the number of evaluation metrics, and an appropriate activation function (such as the sigmoid activation function) is used.

[0129] Further preferably, the battery health status includes battery health and battery aging degree.

[0130] The formula for calculating the current battery health is:

[0131] H = e(W2·ReLU(W1·F F +b1)+b2),

[0132] where H is a value between 0 and 1, representing the current health status of the battery; W1 is the weight matrix connecting the input layer and the first hidden layer in the neural network, W2 is the weight matrix connecting the first hidden layer and the output layer in the neural network; b1 is the bias vector of the first hidden layer in the neural network, b2 is the bias vector of the output layer in the neural network; ReLU() is the ReLU activation function; e is the activation function of the output layer; F F is the comprehensive feature vector.

[0133] The formula for calculating the current battery aging degree is:

[0134] A = W2·ReLU(W1·F F +b1)+b2

[0135] where W1 is the weight matrix connecting the input layer and the first hidden layer, W2 is the weight matrix connecting the first hidden layer and the output layer; b1 is the bias vector of the first hidden layer, b2 is the bias vector of the output layer; ReLU() is the ReLU activation function.

[0136] A can be a continuous value, representing the aging degree of the battery; the numerical value of A obtained each time is an exact value. Since battery aging is a continuous process, the numerical values of A obtained after multiple calculations can be continuous values, and these continuous values can reflect the aging state of the battery at different time points.

[0137] Step S503: Set the warning threshold and evaluate according to the real-time warning threshold.

[0138] (a) Preset the health threshold of the battery as θ. If the current value of the battery health is less than the health threshold θ, it is determined that the battery is abnormal or about to enter an abnormal state.

[0139] (b) Preset the aging degree threshold of the battery as θ'. If the current value of the battery aging degree is higher than the aging degree threshold θ', it is prompted that the battery needs maintenance or replacement.

[0140] The warning threshold can be adjusted according to the requirements and experience in actual applications to balance the accuracy of warnings and the false alarm rate.

[0141] Step S504: Generate a warning signal. When the current state parameters of the battery exceed the normal range or the evaluation result is lower than the warning threshold, the warning module will trigger a warning. After the warning is triggered, the system will generate a fault type and send corresponding warning and maintenance measure information / signals (including sounds and color indications for different levels of abnormal states) according to the different levels of the abnormal state, so as to take timely measures to prevent the occurrence or further deterioration of battery failures.

[0142] The beneficial effects of the battery warning system and its control method based on the transfer adversarial network model of the present invention are as follows: Through the innovative real-time monitoring and warning technology architecture of the battery management system, the health status of the battery is monitored in real time, potential battery failures are predicted and diagnosed, the reliability and safety of the battery system are ensured, thereby improving the ability to capture complex dynamic changes and potential failures inside the battery, and then achieving precise control of battery failures, improving the accuracy and real-time performance of battery failure warnings, and enhancing the generalization ability of the system.

[0143] The above are only embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structural or equivalent process transformations made using the content of the specification of the present invention, or directly or indirectly applied in other related technical fields, are similarly included in the patent protection scope of the present invention.

Claims

1. A control method for a battery early warning system based on a migration adversarial network model, characterized in that the steps include: (1) real-time collection of battery operating status data and preprocessing of the operating status data; wherein the operating status data includes voltage, current, temperature, and number of charge and discharge cycles; (2) Construct a transfer adversarial network model and use the historical working status data of the battery to perform adversarial training on the transfer adversarial network model; (3) Using the transfer adversarial network model to make early warning decisions; (3.1) Obtain the health status feature vector and the real-time monitoring feature vector generated by the transfer adversarial network model, and fuse the health status feature vector with the acquired real-time monitoring feature vector to form a comprehensive feature vector; (3.2) Status assessment: Based on the neural network, the fused comprehensive feature vector is used to perform real-time assessment of the battery health status to obtain the current battery health and current battery aging degree; (3.2.1) The current battery health H is: H=e(W2·ReLU(W1·F F +b1)+b2), Where H is a value between 0 and 1, indicating the current health of the battery; W1 is the weight matrix connecting the input layer to the first hidden layer in the neural network, and W2 is the weight matrix connecting the first hidden layer to the output layer in the neural network; b1 is the bias vector of the hidden layer in the first neural network, and b2 is the bias vector of the output layer in the neural network; ReLU() is the ReLU activation function; e is the activation function of the output layer; F F is the comprehensive feature vector; (3.2.1) The current battery aging degree A is: <h2 style=";text-align:left;direction:ltr">A = W2 ReLU(W1 F<h2 style=";text-align:left;direction:ltr"> F <h2 style=";text-align:left;direction:ltr"> +b1)+b2; (3.3) Evaluate based on the current battery health, current battery aging and preset warning thresholds; (3.3.1) The health threshold of the battery is preset to θ. If the current value of the battery health is less than the health threshold θ, it is judged that the battery is abnormal or is about to enter an abnormal state, and an early warning is triggered; (3.3.2) The preset battery aging threshold is θ'. If the current battery aging value is higher than the aging threshold θ', it is determined that the battery needs maintenance or replacement, and an early warning is triggered; (3.4) Generate corresponding warning signals based on the evaluation results to provide early warning reminders.

2. According to claim 1, a control method for a battery early warning system based on a transfer adversarial network model is characterized in that: The step of preprocessing the working status data comprises: (1.1) Clean the collected working status data to remove outliers and noise; (1.2) The working status data is normalized; wherein each type of working parameter is normalized separately.

3. The control method of a battery early warning system based on a transfer adversarial network model according to claim 1 is characterized in that: When constructing the transfer adversarial network model, the generator is trained using the historical working status data of the battery, so that the generator can learn and simulate the parameter distribution of the battery under different working conditions; Among them, the training formula of the generator is: The training formula of the discriminator is: Among them, G is the generator, D is the discriminator, X is the real data, pdata(X) is the probability distribution of the real data, D(X) is the judgment result of the discriminator on the real data X, and L G is the loss function of the generator, is the expected value, Z is the noise data, pz(Z) is the probability distribution of the noise data Z, G(Z) is the virtual data generated by the generator, and D(G(Z)) is the judgment result of the discriminator on the virtual data G(Z) generated by the generator.

4. The control method of a battery early warning system based on a transfer adversarial network model according to claim 1 is characterized in that: In step (3.1), the steps of generating the health status feature vector include: (3.1.1) Input fusion: The noise data Z and the historical working status data Xhist are fused to form an input vector: Input = [Z; X hist ]; (3.1.2) Multi-layer neural network processing: The input layer receives the fused input vector; the hidden layer performs nonlinear transformation, and the output of each layer is expressed as: H i =ReLU(W i ·H i-1 +b i ), where W i and b i are the weight matrix and bias vector of the i-th layer, H i-1 It is the output of the previous layer. Through the processing of multiple hidden layers, the features that can reflect the health status of the battery are gradually extracted; (3.1.3) Generation of health status feature vector: The health status feature vector generated by the output layer is: FG=W out H last +b out , Among them, W out and b out is the weight matrix and bias vector of the output layer, H last is the output of the last hidden layer; (4) Generation and update of health status feature vector: The generator periodically generates the health status feature vector according to a preset frequency, or dynamically updates the generated health status feature vector each time the latest input data is received to reflect the real-time health status of the battery.

5. The control method of a battery early warning system based on a transfer adversarial network model according to claim 1 is characterized in that: In step (3.1), the weighted summation method is used for fusion: F F =α·F G +(1-α)·F R , Among them, F F is the comprehensive eigenvector, F G is the health status characteristic, F R is the real-time monitoring feature, α is the weight for generating health status features, and 1-α is the weight for real-time monitoring features.

6. The control method of a battery early warning system based on a transfer adversarial network model according to claim 1 is characterized in that: In step (3.1), The comprehensive characteristic vector is not a single value, it includes the mean of voltage, variance of current, and temperature change.

7. The control method of a battery early warning system based on a transfer adversarial network model according to claim 1 is characterized in that: The structure of the neural network includes an input layer, a hidden layer and an output layer.

8. The control method of a battery early warning system based on a transfer adversarial network model according to claim 1 is characterized in that: In step (3.4), after the warning is triggered, corresponding warning and maintenance measures signals are issued according to the fault type and different levels of abnormal conditions; the warning and maintenance measures signals include sound signals and / or color indication models representing different levels of abnormal conditions.

9. A battery early warning system based on a migration adversarial network model, characterized in that: include: Data acquisition module: used to collect key working status data of the battery in real time; Data preprocessing module: used to preprocess the collected working status data and normalize the data to ensure that the data is within a unified numerical range; Migration adversarial network module: The migration adversarial network model includes a generator and a discriminator, wherein the generator is used to generate a virtual working state data sample close to the real battery state, and the discriminator is used to distinguish the real working state data sample from the virtual working state data sample generated by the generator; Health status assessment module: used to assess the current health status of the battery; Early warning module: used to trigger the early warning mechanism according to the evaluation results and generate early warning signals to indicate possible battery failure.

10. A battery early warning system based on a transfer adversarial network model according to claim 9, characterized in that: It also includes a communication module and a power management module. Communication module: used for data exchange with other system components; Power management module: used to provide stable power supply for the entire system.

Images