Performance degradation data normalization method based on minimum distance optimization

Abstract

The invention relates to the technical field of battery life prediction, in particular to a performance degradation data normalization method based on minimum distance optimization. Comprising the following steps that S1, time sequence vectors of key features of a rechargeable battery are obtained, and the time sequence vectors comprise an original sampling sequence, an envelope sampling sequence, a discrete sampling sequence and a sampling repair sequence of the key features; s2, multi-dimensional feature vectors of different rechargeable batteries are constructed for the different rechargeable batteries, all dimensions of the multi-dimensional feature vectors are composed of different kinds of time sequence vectors, and the lengths of all the sequence vectors of the same battery are equal; s3, accumulating distances; s4, calculating an outlier length; and S5, analysis and application. The accuracy of battery life prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery life prediction, and in particular to a performance degradation data normalization method based on minimum distance optimization. Background Art

[0002] In the field of new energy vehicles and energy storage, lithium-ion batteries are core energy storage units, and the research on performance evaluation and life prediction technology is crucial. As the battery is used for a longer time, its performance will gradually deteriorate, manifested as increased internal resistance and reduced available capacity. Accurately evaluating the battery's state of health (SOH) and predicting its remaining useful life (RUL) are of great significance to ensure battery safety, stable operation, and reduce maintenance costs.

[0003] Battery performance evaluation technology: Battery performance evaluation technology mainly involves battery electrochemical technology, battery management system (BMS) and charging technology. Battery electrochemical technology focuses on the chemical reaction and electron flow inside the battery, which affects the energy density, charging speed, life and safety. BMS is responsible for monitoring the battery status, balancing the charge and temperature between battery cells, and ensuring that the battery operates within a safe range. Charging technology includes fast charging and wireless charging technology to improve the availability and convenience of electric vehicles.

[0004] Battery life prediction technology: The research progress of battery life prediction technology is mainly focused on machine learning methods, including support vector regression (SVR), autoregressive (AR) model, recurrent neural network (RNN) and feedforward neural network (FFNN). These methods analyze battery aging data, extract features and establish prediction models to predict the RUL of the battery. However, these methods may encounter the gradient vanishing problem when processing long sequence data, and have limited adaptability to battery curve characteristics at different degradation stages.

[0005] Multidimensional space line distance calculation technology: In battery performance evaluation and life prediction, multidimensional space line distance calculation technology is used to quantify the differences between battery feature vectors. For example, the Euclidean distance calculation method quantifies the difference by calculating the straight-line distance between two points. In addition, there are methods for calculating the distance from a point to a straight line in multidimensional space, which is particularly important when dealing with battery feature vectors, because the aging process of the battery can be regarded as a curve in multidimensional space.

[0006] Battery key feature definition technology: The key features of the battery include cumulative usage features, health status features, operating status features, and timing features. These features are crucial for battery performance evaluation and life prediction. By extracting and optimizing these features, the battery's health status can be evaluated more accurately and its remaining life can be predicted.

[0007] However, the existing battery performance evaluation and life prediction technologies have made some progress, but still face many challenges, such as the accuracy of feature extraction, model adaptability and prediction accuracy. Therefore, a technology is needed to minimize the multidimensional space distance between different battery feature vectors and improve the accuracy of battery performance evaluation and life prediction. Summary of the invention

[0008] The present invention provides a performance degradation data normalization method based on minimum distance optimization, which can improve the accuracy of battery life prediction.

[0009] In order to solve the above technical problems, the present application provides the following technical solutions: 1. A performance degradation data normalization method based on minimum distance optimization, characterized in that it includes the following steps:

[0010] S1: Obtaining a time series vector of a key feature of a rechargeable battery, wherein the time series vector includes an original sampling sequence, an envelope sampling sequence, a discrete sampling sequence, and a sampling repair sequence of the key feature;

[0011] S2: constructing a multidimensional feature vector for each rechargeable battery, wherein each dimension of the multidimensional feature vector is composed of different types of time series vectors, and the lengths of each sequence vector of the same battery are equal;

[0012] S3: for each rechargeable battery in the target battery set, calculating the multidimensional space line distance between the multidimensional feature vector of the rechargeable battery and the multidimensional feature vectors of other rechargeable batteries, and finally adding up the total distance to obtain the outlier length of the rechargeable battery; selecting the battery with the smallest outlier length as the reference battery in the target battery set;

[0013] S4: assigning respective normalization coefficients to respective dimensions of respective multidimensional feature vectors, and optimizing the values ​​of the normalization coefficients corresponding to respective dimensions of different rechargeable batteries; the optimization goal is to minimize the multidimensional space line distance between the respective dimensions of the optimized multidimensional feature vectors of each battery and the respective dimensions of the multidimensional feature vectors of the reference battery; the optimized multidimensional feature vectors are obtained by multiplying the optimized normalization coefficients with the respective dimensions of the multidimensional feature vectors;

[0014] S5: Taking the first N% intervals of each multidimensional feature vector of each rechargeable battery in the target battery set as input, and taking the optimized normalization coefficient as output, a normalization coefficient prediction model is constructed; wherein N is an integer representing the first N% intervals of the length of each multidimensional feature vector.

[0015] The basic principle and beneficial effects of this scheme: This scheme constructs multidimensional feature vectors of rechargeable batteries and optimizes the normalization coefficients of these vectors so that these feature vectors are as close as possible to the feature vector of a reference battery. The steps of this method include: Data acquisition: Collect time series vectors of key features of rechargeable batteries, which include original sampling sequences, envelope sampling sequences, discrete sampling sequences, and sampling repair sequences. Feature vector construction: Construct respective multidimensional feature vectors for different rechargeable batteries, and each dimension of these vectors is composed of different types of time series vectors. Reference battery selection: Calculate the multidimensional space line distance between the multidimensional feature vector of each battery and the multidimensional feature vectors of other batteries, and select the battery with the smallest outlier length as the reference battery. Normalization coefficient optimization: Assign normalization coefficients to each dimension of each multidimensional feature vector, and minimize the distance from the reference battery feature vector by optimizing these coefficients. Normalization coefficient prediction model construction: Use the first N% intervals of the multidimensional feature vector of each battery as input and the optimized normalization coefficients as output to construct a normalization coefficient prediction model.

[0016] This solution can improve the generalization ability of the model: By making the feature vectors of different batteries close to the feature vectors of the benchmark battery, the generalization ability of the model for different battery performance degradation can be improved. Enhanced prediction accuracy: The optimized normalization coefficient can more accurately reflect the trend of battery performance degradation, thereby improving the accuracy of the life prediction model. Speed ​​up model training: The normalized data is more evenly distributed, which helps to accelerate the convergence of the learning algorithm and reduce training time. Improve model stability: Normalization can reduce the complexity of numerical calculations, especially for distance-based algorithms, where features of different scales may have unequal effects on distance calculations. Normalization helps to improve the stability of the model. Adapt to specific algorithm requirements: Some machine learning algorithms, especially gradient-based optimization algorithms and regularization methods, require that data must be normalized to work correctly. This method meets the preprocessing needs of these algorithms. Better understand feature importance: After normalization, the weight of the feature can directly indicate its impact on the model, making it easier to interpret the model. Reduce numerical instability and improve accuracy: In many machine learning algorithms, very high or very low values ​​can cause numerical instability, such as gradient explosion or vanishing. Normalization helps avoid these problems.

[0017] Furthermore, the method further includes step S6: taking the first N% intervals of each multidimensional feature vector of each rechargeable battery in the target battery set as input and the remaining life as output, to construct a life prediction model.

[0018] Furthermore, the life prediction model is also used to predict the dynamic changes of each multidimensional eigenvector within a range outside the first N% interval; when optimizing the normalization coefficient of the eigenvector to minimize the distance from the benchmark eigenvector, one of the gradient descent method, Newton method, quasi-Newton method, genetic algorithm, particle swarm optimization, simulated annealing, conjugate gradient method, stochastic gradient descent, differential evolution, local search and Lagrange multiplier method is used for calculation.

[0019] Furthermore, the life prediction model and the normalization coefficient prediction model select one or more of a linear regression model, a support vector machine model, a double exponential model, an autoencoder, a time series model, a regression model, a neural network model, a system of equations model, a difference equation model, a big data model, a machine learning model, a support vector machine, a random process model, a Bayesian model, an algebraic equation model, an extreme learning network, a randomness model, a polynomial model, a recurrent neural network, a convolutional neural network, a continuous time model, a discrete time model, a Boltzmann machine, a differential equation model, a linear model, a nonlinear model, a Markov chain model, a deep learning network, a generative adversarial network, a long short-term memory network, and a deep belief network.

[0020] Furthermore, in step S4, in the task of optimizing the values ​​of the normalization coefficients corresponding to the dimensions of different rechargeable batteries, it is necessary to specifically optimize the normalization coefficients of the battery feature vectors so that the feature vectors of different batteries are as close as possible to the feature vectors of the reference battery; this process can be divided into the following steps:

[0021] A1: Define the objective function: Suppose there are N batteries, each battery has M feature dimensions, and the feature vector of the benchmark battery is b = [b1, b2, ..., b M ], the characteristic vector of the i-th battery is x i =[x i1 ,x i2 ,...,x iM ], where b, x i The lengths of the one-dimensional vectors in each are the same; b and x i The lengths of the corresponding one-dimensional vectors are different; the normalization coefficient is w i =[w i1 ,w i2 ,...,w iM ], where w i1 and w i2 Each is a single value;

[0022] The optimization goal is to make the optimized multidimensional feature vector x′ of each battery i The multidimensional space line distance between the reference battery feature vector b is the smallest; the optimized multidimensional feature vector x′ i By transforming the original feature vector xi Multiply by the normalization factor w i Get x′ i =x i ⊙w i , where ⊙ represents the Hadamard product;

[0023] A2: Calculate the multidimensional space line distance between two multidimensional feature vectors and quantify the difference between the two feature vectors;

[0024] A3: Set constraints; the constraints are: all coefficients are constrained to be positive or the normalized coefficients are combined to be 1;

[0025] A4: Select an optimization algorithm.

[0026] Furthermore, the method for calculating the multidimensional space line distance between the two multidimensional feature vectors in step A2 is as follows: x is used to represent a certain multidimensional feature vector, and y is used to represent another multidimensional feature vector; in the process of calculating the multidimensional space line distance here, x and y are regarded as curves in the multidimensional space; when the lengths of x and y are inconsistent, a certain d-dimensional coordinate point x in x is taken i , calculate its distance to all d-dimensional coordinate points in y, and then find the minimum value among them, which is used as x i The dotted distance between x and y represents i The distance between the closest point in y; all coordinate points in x i Add the point-line distances between x and y to get the multidimensional space line distance between x and y.

[0027] Furthermore, the key characteristics of the rechargeable battery in step S1 include cumulative usage characteristics, health status characteristics, operating status characteristics, and timing characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Schematic diagram of the steps of a performance degradation data normalization method based on minimum distance optimization. DETAILED DESCRIPTION

[0029] The following is further described in detail through specific implementation methods:

[0030] Embodiment 1 is as attached Figure 1 As shown,

[0031] A performance degradation data normalization method based on minimum distance optimization includes the following steps:

[0032] S1: Select the time series vector of each key feature of the rechargeable battery, the time series vector includes at least one of the original sampling sequence of each key feature, the envelope sampling sequence of each key feature, the discrete sampling sequence of each key feature, and the sampling repair sequence of each key feature; the key features of the lithium battery include at least one of the cumulative usage feature, the health status feature, the working condition feature, and the timing feature; the envelope sampling sequence is a sampling sequence extracted from the original sampling sequence by envelope processing; the discrete sampling sequence is a sampling sequence discretely extracted from the original sampling sequence according to specific needs; the sampling repair sequence is an improved sampling sequence obtained by repairing or reconstructing the discrete sampling sequence or the envelope sampling sequence by signal processing. That is, this step is to determine the multidimensional feature vector: first, define the multidimensional feature vector for each rechargeable battery in the target battery set. This vector contains features of multiple dimensions, and each feature represents a certain attribute of the battery (such as voltage, capacity, temperature, etc.).

[0033] S2: Construct a multidimensional feature vector for each rechargeable battery. Each dimension of the multidimensional feature vector is composed of different types of time series vectors, and the lengths of the sequence vectors of the same battery are equal. This step is to calculate the multidimensional space line distance: for each rechargeable battery in the set, calculate the multidimensional space line distance between its multidimensional feature vector and the multidimensional feature vectors of other rechargeable batteries in the set. Commonly used distance metrics include Euclidean distance, Manhattan distance, etc. The specific choice depends on the characteristics of the data and the analysis requirements.

[0034] S3: For each rechargeable battery in the target battery set, calculate the multidimensional space line distance between the multidimensional feature vector of the rechargeable battery and the multidimensional feature vectors of other rechargeable batteries, and finally add up the total distance to obtain the outlier length of the rechargeable battery; select the battery with the smallest outlier length as the reference battery in the target battery set; this step is to accumulate the distance: add the distance between the rechargeable battery and all other rechargeable batteries in the set to obtain a total distance value. This total distance reflects the overall similarity or difference between the battery and other batteries.

[0035] The distance types that can be adopted in the multidimensional space line distance calculation process include any one of Manhattan distance, Euclidean distance, Hamming distance, standardized Euclidean distance, Mahalanobis distance, Chebyshev distance, and Minkowski distance.

[0036] S4: Assign each dimension of each multidimensional feature vector its own normalization coefficient, and optimize the value of the normalization coefficient corresponding to each dimension of different rechargeable batteries; the optimization goal is to minimize the multidimensional space line distance between each dimension of the optimized multidimensional feature vector of each battery and each dimension of the multidimensional feature vector of the reference battery; the optimized multidimensional feature vector is obtained by multiplying the optimized normalization coefficient with each dimension of the multidimensional feature vector; in the task of optimizing the value of the normalization coefficient corresponding to each dimension of different rechargeable batteries, it is necessary to specifically optimize the normalization coefficient of the battery feature vector so that the feature vectors of different batteries are as close as possible to the feature vectors of the reference battery. This process can be divided into the following steps:

[0037] A1: Define the objective function.

[0038] First, we need to measure the distance between the optimized multidimensional feature vector and the multidimensional feature vector of the reference battery. Assume that there are N batteries, each with M feature dimensions, and the feature vector of the reference battery is b = [b1, b2, ..., b M ], the characteristic vector of the i-th battery is x i =[x i1 ,x i2 ,…,x iM ]. The lengths of the one-dimensional vectors in b are the same, for example, the lengths of b1 and b2 are the same; x i The lengths of the one-dimensional vectors in are also the same, for example, x i1 and x i2 The lengths of b1, b2 and x are also the same; however, i1 、x i2 The lengths of are not the same. The normalization coefficient is w i =[w i1 ,w i2 ,…,w iM ], where w i1 and w i2 Since the lengths of the one-dimensional vectors in b are the same, b can be regarded as a curve in M-dimensional space, where the first coordinate point is composed of the one-dimensional vectors b1, b2, …, b M The first number is composed of the second coordinate point, and the second coordinate point is composed of the one-dimensional vectors b1, b2, ..., b M By analogy, x i It can also be viewed as a curve in M-dimensional space.

[0039] The optimization goal is to make the optimized multidimensional feature vector x′ of each battery i The multidimensional space line distance between the reference battery feature vector b is the smallest. The optimized multidimensional feature vector x′ i By transforming the original feature vector xi Multiply by the normalization factor w i Get x′ i =x i ⊙w i ⊙ represents the Hadamard product, which is the element-by-element multiplication of two matrices of the same dimension, in order to achieve feature fusion.

[0040] A2: Choose a suitable metric to calculate the two multidimensional feature vectors x′ i The multidimensional space line distance between and b is then used to select a suitable metric to quantify the difference between the two feature vectors.

[0041] A3: It is to set constraints. Usually, the normalized coefficients wi should satisfy certain constraints, such as all coefficients should be positive, or the sum of the normalized coefficients should be 1, etc. These constraints can be achieved by setting additional constraints.

[0042] A4: Select an optimization algorithm. Select an appropriate optimization algorithm to minimize the objective function. Common optimization algorithms include gradient descent, genetic algorithm, particle swarm optimization, etc. When selecting a specific algorithm, consider the scale and complexity of the problem according to the actual situation to select an appropriate method for calculation.

[0043] For each rechargeable battery in the target battery set, the mathematical distance between the multidimensional feature vector of the rechargeable battery and the multidimensional feature vectors of other rechargeable batteries is calculated and the total distance is obtained by adding them up, which is used as the outlier length of the rechargeable battery; the battery with the smallest outlier length is selected as the reference battery in the target battery set;

[0044] The following method can be used to specifically obtain the multidimensional space line distance between two multidimensional feature vectors.

[0045] Here, x represents a multidimensional feature vector, and y represents another multidimensional feature vector; the process of calculating the multidimensional space line distance here regards x and y as curves in the multidimensional space, so the dimensions of the two curves are the same, but the lengths may be different. This is the selection optimization algorithm.

[0046] Then, for a d-dimensional coordinate point x in x i , calculate its distance to all d-dimensional coordinate points in y, and then find the minimum value among them, which is used as x i The dotted distance between x and y represents i The distance between the closest point in y. iBy adding the point-line distance between x and y, we can get the multidimensional space line distance between x and y. This method of obtaining the multidimensional space line distance can deal with the situation where the lengths of x and y are inconsistent. By calculating the distance between each multidimensional coordinate point and all points in the target set, the relative distance is comprehensively considered, thereby effectively measuring the degree of outlier of each point. In the specific calculation process, in order to reduce the impact of noise in the data, it is also possible to find each x i The distance to the second closest point in y is taken as x i The point-to-line distance between and y.

[0047] When calculating the distance between two coordinate points, you can use any of the Euclidean distance formula, Manhattan distance formula, Chebyshev distance formula, Mahalanobis distance formula, Minkowski distance formula, Hamming distance formula, etc. However, once the distance formula is determined, it must remain unchanged because the adopted distance formula directly affects the distance between the two coordinate points.

[0048] Set all x i The second smallest distance added together can be used as the line distance in multidimensional space.

[0049] After that, for each d-dimensional coordinate point x i , find the minimum value of its distance from all multidimensional coordinate points in y, this minimum value represents x i The distance between the closest point in y. i By adding the minimum distance of x and y, we can get the multi-dimensional space line distance between x and y.

[0050] Step S4 is used to calculate the outlier length: the total distance accumulated is regarded as the outlier length of the rechargeable battery. The larger the outlier length, the greater the difference between the battery and other batteries, that is, it deviates more from other batteries in the feature space.

[0051] S5: Take the first N% intervals of each multidimensional feature vector of each rechargeable battery in the target battery set as input, take the optimized normalization coefficient as output, and build a normalization coefficient prediction model. Wherein, N is an integer, representing the first N% intervals of the length of each multidimensional feature vector. This step is used for analysis and application: Based on the calculated outlier length, the characteristics of the rechargeable battery can be further analyzed to identify abnormal or special batteries, or for classification and clustering analysis. This process helps to understand the characteristic distribution and similarity of each battery in the rechargeable battery set, and can be used in application scenarios such as quality control and anomaly detection.

[0052] S6: Taking the first N% interval of each multidimensional feature vector of each rechargeable battery in the target battery set as input and the remaining life as output, a life prediction model is constructed. The life prediction model can also predict the dynamic changes of each multidimensional feature vector within the range outside the first N% interval.

[0053] When optimizing the normalization coefficient of the eigenvector to minimize the distance from the benchmark eigenvector, methods such as gradient descent, Newton's method, quasi-Newton's method, genetic algorithm, particle swarm optimization, simulated annealing, conjugate gradient method, stochastic gradient descent, differential evolution, local search and Lagrange multiplier method can be used. Among them, the gradient descent method uses the gradient information of the objective function to gradually adjust the normalization coefficient to find the minimum value, which is suitable for the case where the objective function is smooth and differentiable. The Newton method uses the first and second order derivatives (Hessian matrix) of the objective function to update the normalization coefficient more accurately, which is suitable for the case where the objective function has good conditions. The quasi-Newton method (such as BFGS) is a method that approximates the Newton method and does not require the calculation of the Hessian matrix. It is suitable for the case where it is difficult to obtain an accurate second order derivative of the objective function. The genetic algorithm optimizes the normalization coefficient by simulating natural selection and genetic operations (such as crossover and mutation), which is suitable for the case where the objective function is complex and difficult to optimize through gradient information. Particle swarm optimization (PSO) optimizes the normalization coefficient by simulating the foraging behavior of bird flocks. It is suitable for large-scale or nonlinear problems and has strong global search capabilities. Simulated annealing optimizes the normalization coefficient by simulating the physical annealing process. It is suitable for complex problems with multiple local minima and can effectively jump out of the local optimal solution. In addition, the conjugate gradient method is an iterative method for large-scale optimization problems. It efficiently solves by searching in the conjugate direction. Stochastic gradient descent (SGD) updates the normalization coefficient by using only a part of the samples in each iteration. It is suitable for large data sets and online learning scenarios. Differential evolution is a population-based random optimization algorithm that optimizes the normalization coefficient through mutation and recombination operations. It is suitable for complex optimization problems. The local search method gradually improves the solution by searching in the neighborhood of the current solution. It is suitable for optimization tasks that require fine-tuning. The Lagrange multiplier method handles optimization problems with constraints by introducing Lagrange multipliers. It is suitable for situations where specific constraints need to be met.

[0054] When constructing the life prediction model and the normalization coefficient prediction model, more complex neural network models can be used, as well as simpler empirical mathematical models, including linear regression models, support vector machine models, double exponential models, autoencoders, time series models, regression models, neural network models, system of equations models, difference equation models, big data models, machine learning models, support vector machines, random process models, Bayesian models, algebraic equation models, extreme learning networks, randomness models, polynomial models, recurrent neural networks, convolutional neural networks, continuous time models, discrete time models, Boltzmann machines, differential equation models, linear models, nonlinear models, Markov chain models, deep learning networks, generative adversarial networks, long short-term memory networks, deep belief networks, etc.

[0055] Among them, the linear regression model establishes a linear relationship between the target variable and one or more independent variables, and is used for simple prediction tasks and trend analysis. The support vector machine model classifies data or performs regression analysis by finding the optimal hyperplane, and is suitable for processing high-dimensional data and classification problems with obvious boundaries. The polynomial model models data through polynomial regression and is suitable for processing data with nonlinear relationships, such as quadratic or cubic polynomial regression models. The double exponential model is used to describe the double exponential growth or decay trend in the data, and is suitable for data with complex dynamic changes, such as applications in financial markets. Time series models such as the autoregressive integrated moving average model are used to process trends and seasonal changes in time series data. Neural network models include deep learning networks such as convolutional neural networks and recursive neural networks, which are suitable for processing complex nonlinear relationships and sequence data. Big data models use large amounts of data for modeling, and commonly use integrated learning methods such as random forests or gradient boosting trees to improve prediction accuracy by training large-scale data sets. Machine learning models include various supervised learning and unsupervised learning algorithms, such as decision trees, K nearest neighbors, naive Bayes, etc., which are modeled and predicted through data-driven methods. Bayesian models perform probabilistic inference and prediction based on Bayesian theorem, such as the Naive Bayes classifier, and are suitable for processing data with uncertainty. These model methods can be selected and combined according to the characteristics of the data and the complexity of the problem to achieve more accurate and effective predictions.

[0056] The above are only embodiments of the present invention. The invention is not limited to the fields involved in this implementation case. The common sense such as the known specific structures and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field to which the invention belongs before the application date or the priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to explain the content of the claims.

Claims

1. A performance degradation data normalization method based on minimum distance optimization, characterized in that: The following steps are involved: S1: Obtaining a time series vector of a key feature of a rechargeable battery, wherein the time series vector includes an original sampling sequence, an envelope sampling sequence, a discrete sampling sequence, and a sampling repair sequence of the key feature; S2: constructing a multidimensional feature vector for each rechargeable battery, wherein each dimension of the multidimensional feature vector is composed of different types of time series vectors, and the lengths of each sequence vector of the same battery are equal; S3: for each rechargeable battery in the target battery set, calculating the multidimensional space line distance between the multidimensional feature vector of the rechargeable battery and the multidimensional feature vectors of other rechargeable batteries, and finally adding up the total distance to obtain the outlier length of the rechargeable battery; selecting the battery with the smallest outlier length as the reference battery in the target battery set; S4: assigning respective normalization coefficients to respective dimensions of respective multidimensional feature vectors, and optimizing the values ​​of the normalization coefficients corresponding to respective dimensions of different rechargeable batteries; the optimization goal is to minimize the multidimensional space line distance between the respective dimensions of the optimized multidimensional feature vectors of each battery and the respective dimensions of the multidimensional feature vectors of the reference battery; the optimized multidimensional feature vectors are obtained by multiplying the optimized normalization coefficients with the respective dimensions of the multidimensional feature vectors; S5: Taking the first N% intervals of each multidimensional feature vector of each rechargeable battery in the target battery set as input, and taking the optimized normalization coefficient as output, a normalization coefficient prediction model is constructed; wherein N is an integer representing the first N% intervals of the length of each multidimensional feature vector.

2. The performance degradation data normalization method based on minimum distance optimization according to claim 1, characterized in that: The method further includes step S6: taking the first N% intervals of each multidimensional feature vector of each rechargeable battery in the target battery set as input and the remaining life as output, to construct a life prediction model.

3. The performance degradation data normalization method based on minimum distance optimization according to claim 2, characterized in that: The life prediction model is also used to predict the dynamic changes of each multidimensional eigenvector within a range outside the first N% interval; when optimizing the normalization coefficient of the eigenvector to minimize the distance from the benchmark eigenvector, one of the gradient descent method, Newton method, quasi-Newton method, genetic algorithm, particle swarm optimization, simulated annealing, conjugate gradient method, stochastic gradient descent, differential evolution, local search and Lagrange multiplier method is used for calculation.

4. The performance degradation data normalization method based on minimum distance optimization according to claim 3 is characterized in that: The life prediction model and the normalization coefficient prediction model are selected from one or more of a linear regression model, a support vector machine model, a double exponential model, an autoencoder, a time series model, a regression model, a neural network model, a system of equations model, a difference equation model, a big data model, a machine learning model, a support vector machine, a random process model, a Bayesian model, an algebraic equation model, an extreme learning network, a randomness model, a polynomial model, a recurrent neural network, a convolutional neural network, a continuous time model, a discrete time model, a Boltzmann machine, a differential equation model, a linear model, a nonlinear model, a Markov chain model, a deep learning network, a generative adversarial network, a long short-term memory network, and a deep belief network.

5. The performance degradation data normalization method based on minimum distance optimization according to claim 1, characterized in that: In step S4, in the task of optimizing the values ​​of the normalization coefficients corresponding to the dimensions of different rechargeable batteries, it is necessary to specifically optimize the normalization coefficients of the battery feature vectors so that the feature vectors of different batteries are as close as possible to the feature vectors of the reference battery; this process can be divided into the following steps: A1: Define the objective function: Suppose there are N batteries, each battery has M feature dimensions, and the feature vector of the benchmark battery is b = [b1, b2, ..., b M ], the characteristic vector of the i-th battery is x i =[x i1 ,x i2 ,…,x iM ], where b, x i The lengths of the one-dimensional vectors in each are the same; b and x i The lengths of the corresponding one-dimensional vectors are different; the normalization coefficient is w i =[w i1 ,w i2 ,…,w iM ], where w i1 and w i2 Each is a single value; The optimization goal is to make the optimized multidimensional feature vector x′ of each battery i The multidimensional space line distance between the reference battery feature vector b is the smallest; the optimized multidimensional feature vector x′ i By transforming the original feature vector x i Multiply by the normalization factor w i Get x′ i =x i ⊙w i , where ⊙ represents the Hadamard product; A2: Calculate the multidimensional space line distance between two multidimensional feature vectors and quantify the difference between the two feature vectors; A3: Set constraints; the constraints are: all coefficients are constrained to be positive or the normalized coefficients are combined to be 1; A4: Select an optimization algorithm.

6. The performance degradation data normalization method based on minimum distance optimization according to claim 5, characterized in that: The method for calculating the multidimensional space line distance between two multidimensional feature vectors in step A2 is as follows: x is used to represent a certain multidimensional feature vector, and y is used to represent another multidimensional feature vector; in the process of calculating the multidimensional space line distance here, x and y are regarded as curves in the multidimensional space; when the lengths of x and y are inconsistent, a certain d-dimensional coordinate point x in x is taken. i , calculate its distance to all d-dimensional coordinate points in y, and then find the minimum value among them, which is used as x i The dotted distance between x and y represents i The distance between the closest point in y; all coordinate points in x i Add the point-line distances between x and y to get the multidimensional space line distance between x and y.

7. The performance degradation data normalization method based on minimum distance optimization according to claim 1, characterized in that: The key characteristics of the rechargeable battery in step S1 include cumulative usage characteristics, health status characteristics, operating status characteristics, and timing characteristics.

Images