Search

CN-122008782-A - New energy automobile thermal management system based on AI technology and intelligent control method thereof

CN122008782ACN 122008782 ACN122008782 ACN 122008782ACN-122008782-A

Abstract

The invention relates to the field of new energy automobile control, in particular to a new energy automobile thermal management system based on an AI technology and an intelligent control method thereof, wherein the method comprises the steps of constructing a multi-field coupling feature set based on multi-source heterogeneous original signals in the running process of the new energy automobile; based on a kernel partial least square algorithm as a prediction framework, introducing a local linear mechanism to perform dimension reduction processing, obtaining a final thermal management key parameter predicted value through an embedded fuzzy reasoning strategy, and combining the sequential simulation of a thermal management system simulation model to obtain an optimal control strategy set. According to the invention, the optimal neighbor samples are screened through Euclidean distance measurement feature sample similarity, local embedding weights are locked, the low-dimensional space dimension is dynamically adapted by combining nuclear density analysis, and then iterative optimization is carried out through a gradient descent method, so that invalid redundant information in multi-field coupling features is removed, the calculation complexity of a subsequent prediction model is reduced, and a local nonlinear associated structure playing a key role in thermal management state change is reserved to the maximum extent.

Inventors

  • FENG LEI

Assignees

  • 常州聚元祥新技术有限公司

Dates

Publication Date
20260512
Application Date
20260109

Claims (9)

  1. 1. An intelligent control method of a new energy automobile thermal management system based on AI technology is characterized by comprising the following steps: S101, constructing a coupling incidence matrix among multi-source signals based on multi-source heterogeneous original signals in the running process of a new energy automobile, obtaining a key signal subset, and extracting multi-dimensional features of the key signal subset based on a time domain, a frequency domain and a time-frequency domain to obtain a multi-field coupling feature set; S102, taking a kernel-bias least square algorithm as a prediction framework, introducing a local linear mechanism to perform dimension reduction on the obtained multi-field coupling feature set as input to obtain a low-dimensional feature set with a local nonlinear correlation structure reserved, inputting the low-dimensional feature set into the prediction framework to obtain a preliminary thermal management key parameter predicted value, dynamically correcting the preliminary thermal management key parameter predicted value through an embedded fuzzy reasoning strategy to obtain an error correction amount, and obtaining a final thermal management key parameter predicted value according to the preliminary thermal management key parameter predicted value and the error correction amount; s103, based on the final predicted value of the thermal management key parameter, combining with the one-by-one simulation of the simulation model of the thermal management system, taking the obtained final predicted value of the thermal management key parameter as the ascending value of each execution part of the new energy automobile, and obtaining an optimal control strategy set according to the constraint of the execution parts; S104, analyzing the obtained optimal control strategy set into specific control instructions of each execution component through the central controller, and carrying out real-time communication of the specific control instructions with the local execution controller through the CAN bus to complete thermal management control.
  2. 2. The intelligent control method of the new energy automobile thermal management system based on the AI technology as claimed in claim 1, wherein after the multi-source heterogeneous original signals are obtained, S101 uses the driving mileage pulse signals of the new energy automobile as time references, extracts timestamp information of all the multi-source heterogeneous original signals, complements missing sampling points of the signals through a linear interpolation method, unifies all the multi-source heterogeneous original signals into the same sampling frequency, and completes time axis alignment of the input multi-source heterogeneous original signals; Dividing any two groups of aligned multi-source heterogeneous original signals into data segments with equal length respectively, calculating conceptual density distribution of each data segment through a kernel density estimation method, solving mutual information entropy values through integral operation, finishing calculation of mutual information entropy values of any two groups of aligned multi-source heterogeneous original signals, constructing an association intensity matrix based on the mutual information entropy values of all any two groups of aligned multi-source heterogeneous original signals, wherein element values of the association intensity matrix correspond to the mutual information entropy values of the two groups of aligned multi-source heterogeneous original signals; carrying out Granger causal examination on any two groups of aligned multi-source heterogeneous original signals with mutual information entropy values higher than the mutual information entropy mean value in the correlation intensity matrix to generate a causal relation matrix; And fusing the correlation intensity matrix and the causal relation matrix by adopting element-level product operation to form a coupling correlation matrix among the multi-source signals, wherein matrix element values of the coupling correlation matrix are products of corresponding element values of the correlation intensity matrix and direction coefficients of corresponding elements of the causal relation matrix, and signals influencing the coupling states of a thermal field, an electric field and a flow field are screened out based on the coupling correlation matrix to obtain a key signal subset.
  3. 3. The intelligent control method of the new energy automobile thermal management system based on the AI technology as claimed in claim 1, wherein after the key signal subset is obtained, the S101 extracts statistical features in the key signal subset in a time domain, extracts frequency domain features including feature frequency, frequency amplitude and frequency band energy through fast Fourier transform, extracts time-frequency domain features including wavelet coefficient energy and entropy value through wavelet packet transform, and performs feature dimension correction and normalization processing to obtain a multi-field coupling feature set.
  4. 4. The intelligent control method of the new energy automobile thermal management system based on AI technology according to claim 1, wherein S102 measures the similarity between each feature sample and all other feature samples in the multi-field coupling feature set based on euclidean distance, screens out at least ten candidate samples most similar to the feature samples based on a similarity sorting result, determines the number of optimal neighbor samples, namely the optimal number of candidate samples, through cross-validation, and determines the number of optimal neighbor samples and the corresponding candidate samples corresponding to the multi-field coupling feature set; constructing a local linear reconstruction matrix based on the number of the optimal neighbor samples and the total number of the characteristic samples, wherein the dimension of the local linear reconstruction matrix is set to be the product of the number of the optimal neighbor samples and the total number of the characteristic samples in the multi-field coupling characteristic set, and each matrix element in the local linear reconstruction matrix corresponds to the linear reconstruction coefficient of the candidate sample to the characteristic sample; When a local linear reconstruction matrix is constructed, firstly, setting initial values of all elements in the local linear reconstruction matrix to be zero values, setting matrix elements corresponding to the characteristic samples and the candidate samples to be non-zero values to be solved, fixing the rest elements to be zero values, taking a reconstruction error of linear combination of the characteristic samples and the candidate samples as a target, constructing an error loss function, wherein the reconstruction error is a two-norm of a difference value of a characteristic vector of the characteristic samples and a characteristic vector of the candidate samples, adding constraint conditions that the sum of weights of the reconstruction coefficients corresponding to all the candidate samples is 1, and finally obtaining the linear reconstruction coefficient of each candidate sample to the characteristic samples through a least square method, and taking the linear reconstruction coefficient as a local embedding weight.
  5. 5. The intelligent control method of a new energy automobile thermal management system based on AI technology of claim 4, wherein after obtaining the local embedded weights, the weights of each feature sample and the corresponding candidate sample are matched to form a feature sample-candidate sample weight set, and the weight set is stored and locked; The method comprises the steps of analyzing original dimension distribution of all feature samples in a multi-field coupling feature set through nuclear density, counting probability density integral values of overlapping areas of all feature sample distribution, obtaining redundancy overlapping degree among feature samples, wherein the redundancy overlapping degree is more than or equal to 60%, taking one third dimension of the feature samples, the redundancy overlapping degree is less than 40%, taking one half dimension of the feature samples, and completing setting of a low-dimensional space; And iteratively searching a corresponding low-dimensional characteristic sample for each characteristic sample based on a gradient descent method, wherein the method comprises the steps of taking a principal component analysis result of the characteristic sample as an initial low-dimensional coordinate, taking the deviation minimization of a low-dimensional reconstruction weight of the characteristic sample in a low-dimensional space and a fixed local embedding weight as a target, gradually adjusting the initial low-dimensional coordinate until the deviation minimization, finally enabling all the characteristic samples in a multi-field coupling characteristic set to be matched with the corresponding low-dimensional characteristic sample one by one, obtaining a low-dimensional characteristic set with a local nonlinear correlation structure, and completing the introduction of a local linear mechanism and the dimension reduction processing of the multi-field coupling characteristic set.
  6. 6. The intelligent control method of the new energy automobile thermal management system based on the AI technology as claimed in claim 4, wherein S102 inputs a low-dimensional feature set into a kernel-bias least square algorithm, selects a radial basis function as a mapping carrier from low-dimensional features to high-dimensional feature space, takes the prediction error of the kernel-bias least square algorithm as a target, sets a search range of the radial basis function, adjusts weights and step sizes through a self-adaptive weight particle swarm optimization algorithm, iteratively updates the value of the radial basis function until the prediction error of the kernel-bias least square algorithm is minimized and/or the maximum iteration number is reached, finally obtains the kernel parameters of the optimal radial basis function, calculates the low-dimensional feature vector of the low-dimensional feature set and the optimal radial basis function, and completes nonlinear mapping from the low-dimensional features to the high-dimensional feature space; Arranging all high-dimensional feature vectors obtained after mapping from low-dimensional features to high-dimensional features in rows to construct a high-dimensional feature matrix, wherein each element in the high-dimensional feature matrix corresponds to a single high-dimensional feature value of a single sample, extracting a thermal management key parameter true value corresponding to each high-dimensional feature vector to form a thermal management key parameter vector, minimizing fitting errors of the high-dimensional feature vector and the thermal management key parameter true value into a loss function, and quantifying the fitting errors to obtain a product result of the high-dimensional feature matrix and a linear reconstruction matrix and a two-norm difference value of the thermal management key parameter vector; And solving a loss function by adopting a least square method to obtain a linear coefficient matrix with the minimum fitting error as a linear prediction model, and performing matrix multiplication operation on the high-dimensional feature matrix and the constructed linear coefficient matrix to obtain a preliminary thermal management key parameter predicted value corresponding to each feature sample.
  7. 7. The intelligent control method of the new energy automobile thermal management system based on the AI technology as set forth in claim 1, wherein the S102 performs time axis alignment on the preliminary thermal management key parameter predicted value, the low-dimensional feature set and the thermal management key parameter actual value based on the fuzzy reasoning strategy, and completes normalization processing, and obtains an input data set by detecting and rejecting abnormal values through local abnormal factors; determining the optimal clustering quantity by adopting a fuzzy mean clustering algorithm through cross verification, inputting the preprocessed input data set into the fuzzy mean clustering algorithm, calculating the membership degree of each sample to each clustering center, generating fuzzy rules, wherein each rule corresponds to one clustering category, and finally forming a fuzzy rule base covering the full input scene; Dividing the preprocessed input data set into a training subset and a verification subset, iteratively optimizing membership function parameters used for calculation in the membership process from each sample in the training subset to each clustering center by adopting a gradient descent method, stopping optimization when the training subset reaches the maximum iteration number, obtaining an optimal parameter combination, and solidifying the optimal parameter combination into a fuzzy rule base; The preliminary thermal management key parameter predicted value and the low-dimensional characteristic of the low-dimensional characteristic set are matched with corresponding rules according to a fuzzy rule base, a fuzzy reasoning result is converted into membership degree, an error correction amount is obtained according to the rules corresponding to the membership degree, the preliminary thermal management key parameter predicted value is adjusted based on the error correction amount, specifically, the sum of the preliminary thermal management key parameter predicted value and the error correction amount is used as a corrected predicted value, and finally the thermal management key parameter predicted value and the predicted value set of the thermal management key parameter are obtained.
  8. 8. The intelligent control method of the new energy automobile thermal management system based on the AI technology as claimed in claim 1, wherein the S103 assigns the execution components of the thermal management system with decision priority based on the analytic hierarchy process, eliminates the thermal management key parameter predicted values exceeding the hardware performance constraint and the safety constraint based on the hardware performance constraint and the safety constraint of the execution components, and performs one-by-one simulation on each execution component in combination with the thermal management system simulation model, performs control marking on the execution components exceeding the thermal management key parameter predicted values, the hardware performance constraint and the safety constraint based on the hardware performance constraint of the execution components, and sequentially turns off and/or turns off the execution components of the control marking according to the assigned decision priority.
  9. 9. The utility model provides a new energy automobile thermal management system based on AI technique, is applied to the intelligent control method of new energy automobile thermal management system based on AI technique of any one of claims 1-8, characterized in that, the system includes: the system comprises a data acquisition module processing module, a user, a multi-field coupling feature set and a multi-source data acquisition module processing module, wherein the user acquires multi-source heterogeneous original signals in the running process of the new energy automobile, constructs a coupling incidence matrix among the multi-source signals according to the acquired multi-source heterogeneous original signals, and extracts features to obtain the multi-field coupling feature set; The feature dimension reduction and parameter prediction module is used for introducing a local linear mechanism, screening an optimal neighbor sample of a feature sample based on Euclidean distance, constructing a local linear reconstruction matrix, solving a linear reconstruction coefficient to serve as a local embedding weight, obtaining a low-dimension feature set with a local nonlinear association structure based on iterative optimization of a gradient descent method, inputting the low-dimension feature set into a kernel-bias least square prediction framework, selecting a radial basis function as a mapping carrier, determining an optimal kernel parameter through a self-adaptive weight particle swarm optimization algorithm, completing nonlinear mapping from the low-dimension feature to a high-dimension feature space, constructing a linear prediction model, outputting a preliminary thermal management key parameter predicted value, and outputting a final thermal management key parameter predicted value based on a fuzzy reasoning strategy. The control strategy optimization module is used for giving a scene decision priority to each execution component of the thermal management system based on an analytic hierarchy process, taking the final predicted value of the thermal management key parameter as a target reference value of each execution component, carrying out one-by-one simulation on each execution component by combining a thermal management system simulation model, carrying out control marking on the execution components with simulation results exceeding the predicted value, hardware performance constraint and safety constraint, and sequentially executing shutdown or regulating down operation according to the decision priority to finally form an optimal control strategy set.

Description

New energy automobile thermal management system based on AI technology and intelligent control method thereof Technical Field The invention relates to the field of new energy automobile control, in particular to a new energy automobile heat management system based on an AI technology and an intelligent control method thereof. Background The new energy automobile thermal management system is a key for guaranteeing stable operation of core components such as a power battery, a driving motor and the like, and the operation state of the new energy automobile thermal management system is directly related to the cruising ability, the service life and the safety performance of a vehicle. Multi-source heterogeneous signals such as power battery voltage/current, motor temperature, environmental parameters, vehicle running state and the like can be generated in the running process of the current new energy automobile, and the signals have the characteristics of inconsistent time sequence, complex association relationship and strong coupling, so that deep coupling association among signals is difficult to accurately excavate in the prior art, key influence signals are inaccurate in screening, and redundant information is easy to interfere. In the link of feature processing and parameter prediction, the traditional method faces multiple bottlenecks, namely on one hand, the multi-field coupling feature dimension is high, the conventional dimension reduction technology is easy to lose a local nonlinear correlation structure which plays a key role in the change of the thermal management state, the calculation load of a follow-up prediction model is increased, on the other hand, the single prediction algorithm is insufficient in the nonlinear change adaptation of the thermal management parameters, the fluctuation of the parameters is difficult to dynamically track, an effective error correction mechanism is lacked, the prediction precision is limited, the anti-interference capability is weak, the efficient and optimal operation of the thermal management system cannot be realized, and even the stability of the whole system is possibly influenced by the overload of local components. Disclosure of Invention Aiming at the technical problems existing in the prior art, the invention provides a new energy automobile heat management system based on an AI technology and an intelligent control method thereof. The technical scheme for solving the technical problems is as follows, an intelligent control method of a new energy automobile thermal management system based on an AI technology, the method comprises the following steps: S101, constructing a coupling incidence matrix among multi-source signals based on multi-source heterogeneous original signals in the running process of a new energy automobile, wherein the multi-source heterogeneous original signals comprise, but are not limited to, power battery single voltage/current/internal resistance time sequence signals, motor stator/rotor temperature time sequence signals, air conditioning system evaporation/condensation pressure signals, cooling liquid flow/temperature signals, external environment temperature/wind speed/sunlight intensity signals, vehicle running state parameters such as vehicle speed/acceleration/braking frequency signals and battery SOC/SOF/SOH state signals, acquiring a key signal subset which plays a leading role in the stability of a thermal management system, and acquiring a multi-field coupling feature set based on multi-dimensional features of the key signal subset in a time domain, a frequency domain and a time domain; S102, taking a kernel-bias least square algorithm as a prediction framework, introducing a local linear mechanism to perform dimension reduction on the obtained multi-field coupling feature set as input to obtain a low-dimensional feature set with a local nonlinear correlation structure reserved, inputting the low-dimensional feature set into the prediction framework to obtain a preliminary thermal management key parameter predicted value, dynamically correcting the preliminary thermal management key parameter predicted value through an embedded fuzzy reasoning strategy to obtain an error correction amount, and obtaining a final thermal management key parameter predicted value according to the preliminary thermal management key parameter predicted value and the error correction amount; s103, based on the final predicted value of the thermal management key parameter, combining with the one-by-one simulation of the simulation model of the thermal management system, taking the obtained final predicted value of the thermal management key parameter as the ascending value of each execution part of the new energy automobile, and obtaining an optimal control strategy set according to the constraint of the execution parts; S104, analyzing the obtained optimal control strategy set into specific control instructions of each execution component through