Data prediction method, device and equipment

Abstract

The invention relates to a data prediction method, device and equipment, and the method comprises the steps: carrying out the local direction centrality clustering processing of obtained related data of a thermal management domain in a historical time period, and obtaining a plurality of data clusters; variational mode decomposition is carried out on each data cluster, and P1 intrinsic mode functions are obtained; determining respective sample entropies of the P1 intrinsic mode functions, combining the intrinsic mode functions with sample entropy differences smaller than a first threshold to form new intrinsic mode functions, and obtaining P2 new intrinsic mode functions; determining at least one target influence factor which has the highest influence on the related data based on the plurality of influence factors of the historical time period; the influence factors comprise meteorological factors and vehicle related factors; and based on P2 new intrinsic mode functions corresponding to the plurality of data clusters and at least one target influence factor, performing data prediction in a preset future time period by using a first depth adaptive residual echo state network of a grey wolf optimization algorithm to obtain a first prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle thermal management, and in particular to a data prediction method, device and equipment. Background Art

[0002] As the electrification and intelligentization of automobiles accelerate, the complexity of thermal management systems continues to rise. In the process of iterating from traditional fuel vehicles to new energy vehicles, multiple subsystems such as power battery thermal management, electric drive temperature control, and cabin air quality management form a multi-dimensional coupling system. The parameters such as temperature, pressure, and PM2.5 that the system needs to monitor in real time are growing exponentially. The industry faces two challenges: first, the deployment cost of traditional physical sensors has skyrocketed, the number of sensors in a single high-end vehicle has increased exponentially, and the complexity of the wiring harness has led to an increase in the failure rate; second, under extreme working conditions (such as high-cold and humid environments), sensors are prone to data lag or failure, which restricts the real-time response capability of the thermal management system. Therefore, there is an urgent need to provide a data prediction method for the thermal management domain. Summary of the Invention

[0003] One of the purposes of the present invention is to provide a data prediction method to solve the problem that the hidden headlights in the prior art will affect the light transmission effect after long-term use, and may easily bring driving safety hazards when tested in low-temperature areas; the second purpose is to provide a data prediction device; the third purpose is to provide a vehicle-mounted device; the fourth purpose is to provide a computer-readable storage medium; and the fifth purpose is to provide a computer program product.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] An embodiment of the present invention provides a data prediction method, characterized in that the data prediction method includes:

[0006] The relevant data of the thermal management domain in the historical period are clustered by local directional centrality to obtain multiple data clusters;

[0007] Perform variational mode decomposition on each data cluster to obtain P1 intrinsic mode functions; P1 is an integer greater than or equal to 2;

[0008] Determine the sample entropy of each of the P1 intrinsic mode functions, combine the intrinsic mode functions whose sample entropy difference is less than a first threshold to form a new intrinsic mode function, and obtain P2 new intrinsic mode functions; P2 is an integer greater than or equal to 1;

[0009] Determining, based on the plurality of influencing factors of the historical period, at least one target influencing factor having the highest impact on relevant data of the thermal management domain of the historical period; wherein the plurality of influencing factors include meteorological factors and vehicle-related factors;

[0010] Based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target impact factor, a first deep adaptive residual echo state network of the gray wolf optimization algorithm is used to perform data prediction for a preset future time period to obtain a first prediction result.

[0011] Based on the above technical approach, a local directional centrality clustering algorithm is applied to accurately cluster data related to the thermal management domain over a historical period, generating multiple data clusters. Next, a variational mode decomposition technique is used to process the data sequence, adaptively decomposing it into several intrinsic mode function (IMF) components. This step effectively reduces the nonstationarity and complexity of the thermal management domain data, providing a more stable input for subsequent modeling. The sample entropy value of each IMF component is calculated to quantify its complexity. Then, based on an entropy similarity criterion, IMF components with similar sample entropy values ​​are merged and reconstructed into a new, more representative IMF, achieving classification and simplified representation of the original information. The new IMF and at least one target influencing factor with the highest impact on the thermal management domain data over the historical period are then input into the first deep adaptive residual echo state network of the Grey Wolf Optimization Algorithm to predict data for a predetermined future period, generating a first prediction result. This enables prediction of data related to the thermal management domain for the future period. Furthermore, the predicted data can be combined to proactively control the thermal management domain, thereby reducing failure rates and improving real-time responsiveness within the thermal management domain.

[0012] Furthermore, the method of determining at least one target influencing factor having the highest impact on the relevant data of the thermal management domain of the historical period based on the multiple influencing factors of the historical period includes: determining the first maximum information coefficient between the relevant data of the thermal management domain of the historical period and each of the multiple influencing factors of the historical period; and based on the first maximum information coefficient corresponding to each of the multiple influencing factors of the historical period, screening out the influencing factor corresponding to the maximum value as the target influencing factor having the highest impact on the relevant data of the thermal management domain of the historical period.

[0013] According to the above technical means, by screening out the single influencing factor of the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, that is, the target influencing factor, data prediction for the future period is subsequently performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0014] Furthermore, the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period is used to screen out the influencing factor corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period, including: taking the multiple influencing factors of the historical period as a whole to determine the second maximum information coefficient of the relevant data of the thermal management domain of the historical period; based on the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period and the second maximum information coefficient, screen out the influencing factor corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0015] According to the above technical means, by screening out a single influencing factor or a group of influencing factors in the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, data prediction for the future period is subsequently performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0016] Furthermore, the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period and the second maximum information coefficient are used to screen out the influencing factor corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period, including: obtaining multiple influencing factors of at least one lag period; wherein the at least one lag period is a lag period obtained by lagging the historical period by at least a period of time; determining the third maximum information coefficient between the relevant data of the thermal management domain of the historical period and each influencing factor among the multiple influencing factors of each lag period; taking the multiple influencing factors of each lag period as a whole, determining the fourth maximum information coefficient between the relevant data of the thermal management domain of the historical period and the multiple influencing factors of each lag period; based on the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period, the second maximum information coefficient, the third maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of each lag period, and the fourth maximum information coefficient corresponding to the multiple influencing factors of each lag period as a whole, screening out a single influencing factor or a group of influencing factors corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0017] According to the above technical means, by screening out a single influencing factor or a group of influencing factors in the same period or lag period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, data prediction for the future period is subsequently performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0018] Furthermore, the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period is used to screen out the influencing factor corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period, including: generating multiple groups of influencing factors based on the multiple influencing factors of the historical period; wherein the number of influencing factors included in each group of influencing factors is greater than or equal to 2; determining the fifth maximum information coefficient between the relevant data of the thermal management domain of the historical period and each group of influencing factors; based on the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period and the fifth maximum information coefficient corresponding to each group of influencing factors among the multiple groups of influencing factors, screening out a single influencing factor or a group of influencing factors corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0019] According to the above technical means, by screening out a single influencing factor or a group of influencing factors in the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, data prediction for the future period is subsequently performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0020] Furthermore, the data prediction method also includes: defining multiple hyperparameters of an initial second deep adaptive residual echo state network; using the gray wolf optimization algorithm to randomly generate multiple groups of hyperparameter combinations, and selecting the best hyperparameter combination from the multiple groups of hyperparameter combinations; wherein, the value of at least one hyperparameter in different groups of hyperparameter combinations is different; based on the best hyperparameter combination, multiple groups of first sample data and corresponding true values, training the output weights of the second deep adaptive residual echo state network to obtain the trained first deep adaptive residual echo state network of the gray wolf optimization algorithm.

[0021] Based on the above technical means, the Gray Wolf Optimization algorithm is used to automatically optimize the key hyperparameters of the Deep Adaptive Residual Echo State Network (DARESN), so that the structure and dynamic characteristics of DARESN are highly matched with the specific dataset, thereby improving the prediction accuracy.

[0022] Furthermore, the gray wolf optimization algorithm is used to randomly generate multiple groups of hyperparameter combinations, and the best hyperparameter combination is selected from the multiple groups of hyperparameter combinations, including: based on each group of hyperparameter combinations, multiple groups of second sample data and corresponding true values, training the output weights of the second deep adaptive residual echo state network to obtain a trained third deep adaptive residual echo state network; using multiple groups of third sample data and corresponding true values ​​as inputs of the third deep adaptive residual echo state network, calculating the error between the predicted value and the true value, and obtaining the fitness value of the corresponding group of hyperparameter combinations; based on the fitness values ​​corresponding to each of the multiple groups of hyperparameter combinations, sorting the multiple groups of hyperparameter combinations in descending order to obtain sorted multiple groups of hyperparameter combinations; updating the other hyperparameter combinations in the sorted multiple groups of hyperparameter combinations except the multiple groups of target hyperparameter combinations arranged in front according to a preset offset to obtain an updated hyperparameter combination; repeating the above steps until the iteration condition is met, and determining the best hyperparameter combination based on the latest sorted multiple groups of hyperparameter combinations.

[0023] Furthermore, based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target influencing factor, the first deep adaptive residual echo state network of the gray wolf optimization algorithm is used to perform data prediction for a preset future time period to obtain a first prediction result, including: normalizing the P2 new intrinsic mode functions corresponding to each of the multiple data clusters to obtain a first normalized result; normalizing the at least one target influencing factor to obtain a second normalized result; using the first normalized result and the second normalized result as inputs of the first deep adaptive residual echo state network of the gray wolf optimization algorithm to perform data prediction for a preset future time period to obtain a second prediction result; and denormalizing the second prediction result to obtain the first prediction result.

[0024] An embodiment of the present invention provides a data prediction device, characterized in that the data prediction device includes:

[0025] A clustering unit is used to perform local directional centrality clustering processing on the relevant data of the thermal management domain of the acquired historical period to obtain multiple data clusters; a decomposition unit is used to perform variational mode decomposition on each data cluster to obtain P1 intrinsic mode functions; P1 is an integer greater than or equal to 2; a merging unit is used to determine the sample entropy of each of the P1 intrinsic mode functions, merge the intrinsic mode functions whose sample entropy difference is less than a first threshold to form a new intrinsic mode function, and obtain P2 new intrinsic mode functions; P2 is an integer greater than or equal to 1; a determination unit is used to determine at least one target influencing factor that has the highest impact on the relevant data of the thermal management domain of the historical period based on multiple influencing factors of the historical period; wherein the multiple influencing factors include meteorological factors and vehicle-related factors; a prediction unit is used to use the first deep adaptive residual echo state network of the grey wolf optimization algorithm to perform data prediction for a preset future period based on the P2 new intrinsic mode functions and the at least one target influencing factor corresponding to each of the multiple data clusters to obtain a first prediction result.

[0026] An embodiment of the present invention provides an in-vehicle device, comprising: a processor and a memory configured to store a computer program that can be run on the processor, wherein the processor is configured to execute the steps of the aforementioned method when running the computer program.

[0027] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned method are implemented.

[0028] An embodiment of the present invention provides a computer program product, including a computer program or instructions, which implements the steps of the above method when executed by a processor.

[0029] Beneficial effects of the present invention:

[0030] (1) The present invention uses a local directional centrality clustering algorithm to accurately cluster the relevant data of the thermal management domain in the historical period to obtain multiple data clusters. Then, the variational mode decomposition technology is used to process the data sequence and adaptively decompose it into several intrinsic mode function components. This step effectively reduces the non-stationarity and complexity of the relevant data of the thermal management domain, providing a more stable input for subsequent modeling. The sample entropy value of each intrinsic mode function component is calculated to quantify its complexity. Then, based on the entropy similarity criterion, the intrinsic mode function components with similar sample entropy values ​​are merged and reconstructed into a new, more representative intrinsic mode function, thereby achieving the classification and simplified representation of the original information. The new intrinsic mode function and at least one target influence factor with the highest impact on the relevant data of the thermal management domain in the historical period are then input into the first deep adaptive residual echo state network of the gray wolf optimization algorithm to predict the data of the preset future period, and obtain the first prediction result. In this way, the prediction of the relevant data of the thermal management domain in the future period is achieved, and the predicted data can be further combined to control the thermal management domain in advance, thereby achieving the purpose of reducing the failure rate and improving the real-time response capability of the thermal management domain.

[0031] (2) The present invention screens out a single influencing factor of the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, and then predicts the data of the future period based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 1 ;

[0033] Figure 2 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 2 ;

[0034] Figure 3 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 3 ;

[0035] Figure 4 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 4 ;

[0036] Figure 5 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 5 ;

[0037] Figure 6 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 6 ;

[0038] Figure 7 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 7 ;

[0039] Figure 8 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 8 ;

[0040] Figure 9 A schematic diagram of the structure of a data prediction device according to an embodiment of the present invention;

[0041] Figure 10 Schematic diagram of the structure of the vehicle-mounted equipment in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable a more detailed understanding of the features and technical contents of the embodiments of the present invention, the implementation of the embodiments of the present invention is described in detail below with reference to the accompanying drawings. The accompanying drawings are for reference only and are not intended to limit the embodiments of the present invention.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention pertains. The terms used herein are for the purpose of describing the present embodiment only and are not intended to limit the present invention.

[0044] In the following description, references to “some embodiments,” “this embodiment,” “this embodiment,” and examples, etc., describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict.

[0045] If similar descriptions of "first / second" appear in the application documents, the following explanation is added. In the following description, the terms "first\second\third" are merely used to distinguish similar objects and do not represent a specific order of the objects. It can be understood that "first\second\third" can be interchanged with the specific order or sequence where permitted, so that the embodiment described herein can be implemented in an order other than that illustrated or described herein.

[0046] In this embodiment, the term "and / or" is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, object A and / or object B may represent three situations: object A exists alone, object A and object B exist at the same time, and object B exists alone.

[0047] The embodiment of the present invention provides a data prediction method. Figure 1 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 1 ,like Figure 1As shown, the data prediction method is applied to vehicle-mounted equipment and includes the following steps:

[0048] S101: Performing local directional centrality clustering processing on the relevant data of the thermal management domain in the historical period to obtain multiple data clusters.

[0049] In this embodiment of the present invention, the thermal management domain includes the power battery thermal management module, the electric drive temperature control module, and the cabin thermal management module. Accordingly, relevant data in the thermal management domain includes battery temperature, motor temperature, cabin temperature, cabin air quality, and cabin humidity.

[0050] In an embodiment of the present invention, the historical period may be a period consisting of X hours before the current time, for example, X may be 1, 2, 3, or 4. Typically, data related to the thermal management domain is collected according to a preset collection period, for example, the preset collection period may be 5 seconds, 10 seconds, 30 seconds, or 1 minute, without specific limitation.

[0051] It should be noted that the Clustering by Measuring Local Direction Centrality (CDC) algorithm distinguishes internal points from boundary points by measuring the uniformity of the K-nearest neighbor distribution of each point.

[0052] In the embodiment of the present invention, a local directional centrality clustering algorithm (CDC) is applied to accurately cluster relevant data of the thermal management domain in a historical period to obtain multiple data clusters.

[0053] Based on this, in an embodiment of the present invention, multiple points are generated based on each data in the relevant data of the thermal management domain of the historical period, wherein the horizontal and vertical coordinate values ​​of each point are represented by the same data. The K nearest neighbor points of each point in the multiple points are determined. Each point is connected to the K nearest neighbor points to form K angles. Based on the K angles corresponding to each point, the angular variance of each point, namely the local direction centrality measurement (DCM), is calculated. The local direction centrality measurement is used to characterize the uniformity of the K nearest neighbor distribution of each point. The calculation formula (1) of the local direction centrality measurement DCM is expressed as:

[0054]

[0055] Among them, K represents the number of nearest neighbor points set; α i Represents the angle value; i represents the number of angles.

[0056] Furthermore, based on the local directional centrality measure DCM of each point, it is compared with a pre-set DCM threshold, and the points corresponding to the DCM greater than the DCM threshold are defined as boundary points, and the points corresponding to the DCM less than or equal to the DCM threshold are defined as internal points.

[0057] Furthermore, the connections of the constrained internal points are generated, that is, all the internal points are clustered to divide them into multiple clusters. The specific method is: calculate the boundary points and each internal point p i The closest distance r i , calculate the distance d between the interior points ij , if r i +r j ≥d ij , then the internal point p i With p j Same cluster, otherwise different clusters; where r i It is expressed by formula (2):

[0058]

[0059] Among them, r j It is different from the internal point p i An interior point p j The shortest distance to the boundary point; n refers to the number of all internal points; m refers to the number of clustered internal points.

[0060] The basis for internal point cluster division is expressed by formula (3):

[0061] d(p i ,p j )≤r i +r j (3)

[0062] The clustering results were analyzed using the Davies-Bouldin Index (DBI), and the calculation formula (4) is as follows:

[0063]

[0064] in, is the average distance from the data within the class to the cluster centroid, representing the degree of dispersion of each time series in cluster i, w i Represents the cluster center of each cluster; for each cluster, DBI first finds the average distance between the internal classes and The sum of the two cluster centers is divided by the maximum value; then these maximum values ​​are added together to get the average value; the CDC clustering algorithm has the lowest DBI index and the best clustering effect.

[0065] S102: Perform variational mode decomposition on each data cluster to obtain P1 intrinsic mode functions; P1 is an integer greater than or equal to 2.

[0066] In an embodiment of the present invention, variational mode decomposition (VMD) is an adaptive, non-recursive signal processing method for decomposing a complex signal into multiple intrinsic mode functions (IMFs) with specific frequency characteristics.

[0067] In this embodiment of the present invention, the data in each data cluster is sorted from smallest to largest to obtain a corresponding data sequence. Furthermore, the data sequence corresponding to each data cluster is processed using variational mode decomposition (VMD) technology, adaptively decomposing it into several intrinsic mode function (IMF) components. This step effectively reduces the non-stationarity and complexity of the original data, i.e., the relevant data in the thermal management domain during the historical period, providing a more stable input for subsequent modeling.

[0068] Based on this, the specific steps are as follows:

[0069] (1) Initialization in, Identify the kth modal component and center frequency respectively, is the Lagrangian operator, and the number 1 in the upper left corner indicates the first iteration;

[0070] (2) For each subsequence, namely the intrinsic mode function IMF, the equations (5) and (6) are continuously updated to obtain and

[0071]

[0072] in, is the Wiener filter of the current residual component, is the frequency center of the corresponding modal component, ω is the frequency value; Represents the original sequence f(t), i.e. the clustered data, and w k Fourier transform of , α is the quadratic penalty factor;

[0073] (3) For all ω ≥ 0, update It is expressed by formula (7):

[0074]

[0075] Where τ represents the noise tolerance and K represents the total number of modes;

[0076] (4) Determine whether the iteration termination condition is met, which is expressed by formula (8):

[0077]

[0078] Where ε represents the similarity function; if the termination condition is not met, steps (2) and (3) are repeated. If the condition is met, the iteration is terminated, and the decomposed K subsequences, i.e., P1 intrinsic mode functions (IMFs), are obtained.

[0079] S103: Determine the sample entropy of each of P1 intrinsic mode functions, merge the eigenmode functions whose sample entropy difference is less than a first threshold to form a new eigenmode function, and obtain P2 new eigenmode functions; P2 is an integer greater than or equal to 1.

[0080] In this embodiment of the present invention, the sample entropy (SE) value of each intrinsic mode function (IMF) is calculated to quantify its complexity. Then, based on the entropy similarity criterion, IMF components with similar SE values ​​are combined and reconstructed into a new, more representative sequence (RIMF), namely a new intrinsic mode function, to achieve classification and simplified representation of the original information.

[0081] The steps to determine sample entropy can be:

[0082] For a time series consisting of N points {x(n)}=x(1),x(2),…x(N), the sample entropy is calculated as follows:

[0083] (1) A set of vector sequences with dimension m is formed according to the sequence number, Xm(1),...,Xm(N-m+1);

[0084] Among them, Xm(i)={x(i),x(i+1),...,x(i+m-1)}, (1≤i≤N-m+1);

[0085] (2) The distance d[Xm(i),Xm(j)] between vectors Xm(i) and Xm(j) is defined as the absolute value of the maximum difference between the corresponding elements of the two vectors, which is expressed by formula (9):

[0086]

[0087] (3) Given a threshold r, record the number of j where d[Xm(i),Xm(j)] < r, denoted as B i , for 1≤i≤Nm, put B i The ratio to N-m+1 is expressed as follows using formula (10):

[0088]

[0089] (4) For all The average value is obtained, which is expressed by formula (11):

[0090]

[0091] (5) Increase the dimension to m+1 and calculate X m+1 (i) With X m+1 (j) The number of distances less than or equal to r, denoted as A i , It can be expressed as follows using formula (12):

[0092]

[0093] (6) Definition A m (r), expressed by formula (13):

[0094]

[0095] Among them, B m (r) and A m (r) are the matching probabilities of sequence pairs m and m+1 respectively. The sample entropy is expressed as follows using formula (14):

[0096]

[0097] When N takes a finite value, the estimated value of sample entropy is expressed by formula (15):

[0098]

[0099] S104: Determine at least one target influencing factor having the highest impact on relevant data of the thermal management domain during the historical period based on multiple influencing factors during the historical period; wherein the multiple influencing factors include meteorological factors and vehicle-related factors.

[0100] In the embodiment of the present invention, the influencing factor refers to a factor that affects relevant data of the thermal management domain.

[0101] Among them, multiple influencing factors include meteorological factors and vehicle-related factors.

[0102] For example, if the relevant data of the thermal management domain in the historical period is temperature data, the meteorological factors may include ambient temperature, light intensity, ambient humidity, wind speed, wind direction and particulate matter concentration, and the vehicle-related factors may include vehicle power type, battery chemical material and capacity, vehicle speed, drive power / torque / current, air conditioning operating mode, set temperature and air volume and compressor speed.

[0103] For example, if the relevant data of the thermal management domain in the historical period is humidity data, the meteorological factors may include ambient temperature, precipitation, dew point temperature, geographic location and time, and the vehicle-related factors may include ventilation system status and vehicle window temperature.

[0104] In an embodiment of the present invention, a single target impact factor having the highest impact on the relevant data of the thermal management domain during a historical period can be determined based on multiple impact factors. Alternatively, a combination of impact factors having the highest impact on the relevant data of the thermal management domain during a historical period can be determined based on multiple impact factors, where the combination of impact factors includes multiple target impact factors.

[0105] In the embodiment of the present invention, the corresponding impact factors for relevant data of different types of thermal management domains may be the same or different.

[0106] S105: Based on P2 new intrinsic mode functions and at least one target impact factor corresponding to each of the multiple data clusters, a first deep adaptive residual echo state network of the gray wolf optimization algorithm is used to predict data for a preset future time period to obtain a first prediction result.

[0107] In an embodiment of the present invention, the hyperparameter combination of the first Deep Adaptive Residual Echo State Network (DARESN) is the optimal hyperparameter combination, which is obtained using the Grey Wolf Optimizer (GWO) algorithm. Here, the first Deep Adaptive Residual Echo State Network is a trained network.

[0108] In an embodiment of the present invention, P2 new intrinsic mode functions and at least one target impact factor corresponding to each of the multiple data clusters are encoded into a vector, which is used as the input of a first deep adaptive residual echo state network, and the first prediction result is output.

[0109] It should be noted that the structure of the first deep adaptive residual echo state network includes an input layer, multiple reservoirs, multiple adaptive residual layers, an attention mechanism layer, and an output layer. A gating mechanism is introduced into the reservoirs to enhance temporal modeling capabilities and memory management. Residual layers are provided between reservoirs or across layers, and the attention mechanism layer is connected to the output layer. The first deep adaptive residual echo state network stacks multiple reservoirs. Deep structures can learn more complex and higher-level temporal features. Residual connections are added between multiple reservoirs or across layers. This structure can alleviate the vanishing / exploding gradient problem and facilitate deep training. It allows information to flow directly across layers, enabling the network to simultaneously learn shallow and deep features, and improves the network's ability to model long-term dependencies. The attention mechanism layer can improve feature selectivity and information utilization efficiency.

[0110] In an embodiment of the present invention, a local directional centrality clustering algorithm is applied to accurately cluster data related to the thermal management domain over a historical period, generating multiple data clusters. Next, a variational mode decomposition technique is used to process the data sequence, adaptively decomposing it into several intrinsic mode function (IMF) components. This step effectively reduces the nonstationarity and complexity of the thermal management domain data, providing a more stable input for subsequent modeling. The sample entropy value of each IMF component is calculated to quantify its complexity. Then, based on an entropy similarity criterion, IMF components with similar sample entropy values ​​are merged and reconstructed into a new, more representative IMF, achieving classification and simplified representation of the original information. The new IMF and at least one target influencing factor with the highest impact on the thermal management domain data over the historical period are then input into the first deep adaptive residual echo state network of the Grey Wolf Optimization Algorithm to predict data for a predetermined future period, generating a first prediction result. This enables prediction of data related to the thermal management domain for the future period. Furthermore, the predicted data can be combined to proactively control the thermal management domain, thereby reducing the failure rate and improving the real-time response capability of the thermal management domain.

[0111] In some embodiments of the present invention, determining, based on the multiple influencing factors of the historical period, at least one target influencing factor having the highest impact on the relevant data of the thermal management domain of the historical period comprises the following steps:

[0112] S201: Determine a first maximum information coefficient between relevant data of a thermal management domain in a historical period and each of a plurality of influencing factors in the historical period.

[0113] It should be noted that some existing studies rely primarily on historical values ​​of thermal management domain data when predicting future values, often ignoring the impact of influencing factors (including meteorological and vehicle-related factors). These methods have significant drawbacks, resulting in insufficient extrapolation capabilities. Therefore, it is necessary to further explore the correlation between data and influencing factors to improve prediction accuracy. However, during the modeling process, the introduction of input variables with low correlation to the load will cause additional interference to the model. These redundant variables not only fail to improve prediction capabilities, but also increase the model training burden and reduce prediction performance. Therefore, a key issue in using neural network models to predict thermal management domain data is how to select key variables that significantly influence the data from the numerous influencing factors. The maximum mutual information coefficient (MIC) is a method for measuring the correlation between variables. Its core idea is to quantify their relationship by finding the maximum mutual information between variables. Compared to the traditional correlation coefficient, the MIC is not limited to linear relationships, can identify a wider range of correlation patterns, and is robust to outliers and nonlinear relationships. Furthermore, the MIC performs well when processing high-dimensional data and can be used to explore the correlation between multiple variables.

[0114] Among them, the maximum mutual information value I in MIC * [D(a,b)] is expressed using formulas (16) and (17):

[0115]

[0116] I * [D(a,b)]=max I(D| G ) (17)

[0117] Where P(X,Y) is the joint probability distribution function of X and Y, P(X) and P(Y) are the marginal probability distribution functions of X and Y, and X and Y represent the relevant data and influencing factors of the thermal management domain in the historical period, respectively.

[0118] The maximum mutual information value is normalized so that it is in the interval [0, 1], and the maximum information coefficient MIC is obtained and expressed as follows using formulas (18) and (19):

[0119]

[0120] MIC(D)=max ab＜B(n) {M(D) a,b} (19)

[0121] Where a and b are the number of grids divided in the x and y directions; n is the number of samples; B(n)It is a function of the number of samples, which is used to limit the size of the grid division area. B( n ) =n 0.6 The MIC value represents the maximum mutual information value between two variables. If the mutual information value between the variables is larger, the correlation between them is stronger, and vice versa.

[0122] Based on this, the relevant data of the thermal management domain in the historical period is encoded into a vector in chronological order. Also, the individual influencing factors in the historical period are encoded into a vector in chronological order. These two vectors are then substituted into the above formulas (16) to (19) to obtain the first maximum information coefficient (MIC) of the relevant data of the thermal management domain in the historical period and each influencing factor.

[0123] S202: Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, the influencing factor corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

[0124] In an embodiment of the present invention, multiple first maximum information coefficients MIC are sorted in order from large to small or from small to large to screen out the maximum value, and the impact factor corresponding to the maximum value is used as the target impact factor with the highest impact on the relevant data of the thermal management domain in the historical period.

[0125] In an embodiment of the present invention, a single influencing factor of the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, is screened out, and then data prediction for the future period is performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0126] In some embodiments of the present invention, the step of selecting the influence factor corresponding to the maximum value based on the multiple first maximum information coefficients corresponding to the multiple influence factors in the historical period as the target influence factor having the highest influence on the relevant data of the thermal management domain in the historical period includes the following steps:

[0127] S301: Taking multiple influencing factors of the historical period as a whole, determining the second maximum information coefficient of the data related to the thermal management domain of the historical period.

[0128] In this embodiment of the present invention, multiple influencing factors for a historical period are sorted according to a preset sorting method and chronological order, and encoded into a vector. Data related to the thermal management domain for the historical period is also encoded into a vector in chronological order. These two vectors are then substituted as two variables into the MIC calculation formula to obtain the second maximum information coefficient.

[0129] For example, if multiple influencing factors include influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4, and the preset sorting method is influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4, and the historical period includes four time points t1, t2, t3 and t4, then the encoded vector can be [influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 of t1; influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 of t2; influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 of t3; influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 of t4].

[0130] S302: Based on the first maximum information coefficient and the second maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, the influencing factor corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

[0131] In this embodiment of the present invention, multiple first maximum information coefficients MIC and second maximum information coefficients are sorted in descending or ascending order to select the maximum value. The impact factor corresponding to the maximum value is used as the target impact factor with the highest impact on the relevant data of the thermal management domain during the historical period. Here, when the maximum value is the first maximum information coefficient, the target impact factor is a single impact factor. When the maximum value is the second maximum information coefficient, the target impact factor is all impact factors, i.e., multiple impact factors.

[0132] In an embodiment of the present invention, a single influencing factor or a group of influencing factors of the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, is screened out, and then data prediction for the future period is performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0133] In some embodiments of the present invention, the step of selecting the influence factor corresponding to the maximum value based on the first maximum information coefficient corresponding to each influence factor among the multiple influence factors in the historical period and the second maximum information coefficient as the target influence factor having the highest impact on the relevant data of the thermal management domain in the historical period includes the following steps:

[0134] S401: Acquire multiple impact factors of at least one hysteresis period; wherein the at least one hysteresis period is a hysteresis period obtained by lagging a historical period by at least a period of time.

[0135] For example, the historical period is from 7:00 to 10:00, and the at least one lag period may be 30 minutes or 60 minutes, then the corresponding at least one lag period is from 7:30 to 10:30, and from 8:00 to 11:00.

[0136] S402: Determine the third maximum information coefficient between the relevant data of the thermal management domain in the historical period and each of the multiple influencing factors in each hysteresis period.

[0137] In this embodiment of the present invention, the relevant data of the thermal management domain for the historical period is encoded into a vector in chronological order. The individual impact factors for each lag period are also encoded into a vector in chronological order. These two vectors are then substituted as two variables into the MIC calculation formula to obtain the corresponding third maximum information coefficient.

[0138] S403: Taking the multiple influencing factors of each hysteresis period as a whole, determining the fourth maximum information coefficient of the relevant data of the thermal management domain in the historical period.

[0139] In this embodiment of the present invention, multiple influencing factors for the lag period are sorted according to a preset sorting method and chronological order, and encoded into a vector. Data related to the thermal management domain for the historical period is also encoded into a vector in chronological order. These two vectors are then substituted as two variables into the MIC calculation formula to obtain the fourth maximum information coefficient.

[0140] Exemplarily, if multiple influencing factors include influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4, the preset sorting method is influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4, the historical period includes 4 time points t1, t2, t3 and t4, and the lag period on the historical period is t0, then the lag period includes 4 time points t1+t0, t2+t0, t3+t0 and t4+t0, and the encoded vector can be [influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 at t1+t0; influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 at t2+t0; influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 at t3+t0; influencing factor 1, influencing factor 2, influencing factor 3 and influencing factor 4 at t4+t0].

[0141] For example, if the relevant data of the thermal management domain is temperature data, and the historical period includes four time points t1, t2, t3 and t4, the encoded vector can be [temperature data of t1; temperature data of t2; temperature data of t3; temperature data of t4].

[0142] S404: Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, and the second maximum information coefficient, and the third maximum information coefficient corresponding to each of the multiple influencing factors in each lag period, and the fourth maximum information coefficient corresponding to the multiple influencing factors in each lag period as a whole, a single influencing factor or a group of influencing factors corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

[0143] In an embodiment of the present invention, the first maximum information coefficient MIC corresponding to each of the multiple impact factors in the historical period, the second maximum information coefficient corresponding to the multiple impact factors of the historical period as a whole, the third maximum information coefficient corresponding to each of the multiple impact factors of each hysteresis period, and the fourth maximum information coefficient corresponding to the multiple impact factors of each hysteresis period as a whole are sorted in order from large to small or from small to large to screen out the maximum value, and the impact factor corresponding to the maximum value is used as the target impact factor with the highest impact on the relevant data of the thermal management domain of the historical period. Here, when the maximum value is the first maximum information coefficient, its target impact factor is a single impact factor in the historical period. When the maximum value is the second maximum information coefficient, its target impact factor is the overall impact factors of the historical period. When the maximum value is the third maximum information coefficient of a hysteresis period, its target impact factor is a single impact factor of the hysteresis period. When the maximum value is the fourth maximum information coefficient of a hysteresis period, its target impact factor is the overall impact factors of the hysteresis period.

[0144] In an embodiment of the present invention, a single influencing factor or a group of influencing factors of the same period or lag period that has the highest impact on the relevant data of the thermal management domain of the historical period, namely the target influencing factor, is screened out, and then data prediction for the future period is performed based on the relevant data of the thermal management domain of the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0145] In some embodiments of the present invention, the step of selecting, based on the first maximum information coefficient corresponding to each of the multiple influencing factors, the influencing factor corresponding to the maximum value as the target influencing factor having the highest impact on the relevant data of the thermal management domain in the historical period, includes the following steps:

[0146] S501: Generate multiple groups of impact factors based on multiple impact factors in a historical period; wherein the number of impact factors included in each group of impact factors is greater than or equal to 2.

[0147] For example, if the multiple impact factors of the historical period include impact factor 1, impact factor 2, impact factor 3 and impact factor 4, then the multiple groups of impact factors include: the first group, namely impact factor 1 and impact factor 2; the second group, namely impact factor 1 and impact factor 3; the third group, namely impact factor 1 and impact factor 4; the fourth group, namely impact factor 1, impact factor 2 and impact factor 3; the fifth group, namely impact factor 1, impact factor 2 and impact factor 4; the sixth group, namely impact factor 2, impact factor 3 and impact factor 4; and the seventh group, namely impact factor 1, impact factor 2, impact factor 3 and impact factor 4.

[0148] S502: Determine the second maximum information coefficient between the relevant data of the thermal management domain in the historical period and each group of influencing factors.

[0149] In this embodiment of the present invention, the relevant data of the thermal management domain for a historical period is encoded into a vector in chronological order. Each group of influencing factors for the historical period is sorted according to a preset sorting method for multiple influencing factors and in chronological order, and encoded into a vector. These two vectors are then substituted as two variables into the MIC calculation formula to obtain the second maximum information coefficient.

[0150] For example, if the preset sorting order of multiple impact factors is impact factor 1, impact factor 2, impact factor 3, and impact factor 4, and the historical period includes four time points t1, t2, t3, and t4, then the vector encoded into the second group of impact factors may be [impact factor 1, impact factor 3 at t1; impact factor 1, impact factor 3 at t2; impact factor 1, impact factor 3 at t3; impact factor 1, impact factor 3 at t4]. The vector encoded into the third group of impact factors may be [impact factor 1, impact factor 4 at t1; impact factor 1, impact factor 4 at t2; impact factor 1, impact factor 4 at t3; impact factor 1, impact factor 4 at t4].

[0151] S503: Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, and the fifth maximum information coefficient corresponding to each group of influencing factors in the multiple groups of influencing factors, a single influencing factor or a group of influencing factors corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

[0152] In this embodiment of the present invention, the first maximum information coefficient MIC corresponding to each of the multiple influencing factors in the historical period and the fifth maximum information coefficient corresponding to each group of influencing factors in the historical period are sorted in descending or descending order to select the maximum value. The influencing factor corresponding to the maximum value is used as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period. Here, when the maximum value is the first maximum information coefficient, its target influencing factor is a single influencing factor in the historical period. When the maximum value is the fifth maximum information coefficient, its target influencing factor is a group of influencing factors in the historical period.

[0153] In an embodiment of the present invention, a single influencing factor or a group of influencing factors of the same period that has the highest impact on the relevant data of the thermal management domain in the historical period, namely the target influencing factor, is screened out, and then data prediction for the future period is performed based on the relevant data of the thermal management domain in the historical period and the target influencing factor, thereby improving the accuracy of the first prediction result.

[0154] In some embodiments of the present invention, the data prediction method further comprises the following steps:

[0155] S601: Define multiple hyperparameters of an initial second deep adaptive residual echo state network.

[0156] In the embodiment of the present invention, multiple hyperparameters include the size of the reservoir of each layer, spectral radius, sparsity, leakage rate, number of layers, residual connection weights, etc.

[0157] S602: Using the Gray Wolf Optimization Algorithm, randomly generate multiple sets of hyperparameter combinations, and select the best hyperparameter combination from the multiple sets of hyperparameter combinations; wherein the value of at least one hyperparameter in different sets of hyperparameter combinations is different.

[0158] In an embodiment of the present invention, the gray wolf optimization algorithm is used to randomly generate multiple sets of hyperparameter combinations, i.e., multiple wolves, determine the level of each wolf, and ultimately use the hyperparameter combination corresponding to the leader-level wolf as the optimal hyperparameter combination.

[0159] It should be noted that the gray wolf optimization algorithm is a group intelligence optimization algorithm that imitates the hunting (encirclement, pursuit, and attack) behavior of gray wolf groups. Wolf packs are divided into four levels: α (optimal solution), β (suboptimal solution), δ (third-optimal solution), and ω (ordinary wolf / follower). By simulating the process of collaborative hunting by wolf packs, the optimal solution, that is, the best hyperparameter combination, is searched in the solution space.

[0160] S603: Based on the optimal hyperparameter combination, multiple groups of first sample data and corresponding true values, train the output weights of the second deep adaptive residual echo state network to obtain the trained first deep adaptive residual echo state network of the gray wolf optimization algorithm.

[0161] In an embodiment of the present invention, the second deep adaptive residual echo state network is initialized using the optimal hyperparameter combination, multiple groups of first sample data and corresponding true values ​​are used as inputs of the initialized second deep adaptive residual echo state network, and the predicted value corresponding to each group of sample data is output. Based on the difference between the true value and the predicted value, the output weight of the second deep adaptive residual echo state network is adjusted until the training conditions are met, thereby obtaining the trained first deep adaptive residual echo state network of the gray wolf optimization algorithm.

[0162] In an embodiment of the present invention, the Gray Wolf Optimization Algorithm is used to automatically optimize the key hyperparameters of the Deep Adaptive Residual Echo State Network (DARESN), so that the structure and dynamic characteristics of DARESN are highly matched with a specific dataset, thereby improving the prediction accuracy.

[0163] In some embodiments of the present invention, the method of randomly generating multiple sets of hyperparameter combinations using the Gray Wolf Optimization Algorithm and selecting the best hyperparameter combination from the multiple sets of hyperparameter combinations includes:

[0164] Based on each set of hyperparameter combinations, multiple sets of second sample data and corresponding true values, training the output weights of the second deep adaptive residual echo state network to obtain a trained third deep adaptive residual echo state network;

[0165] Using multiple groups of third sample data and corresponding true values ​​as inputs of the third deep adaptive residual echo state network, calculating the errors between the predicted values ​​and the true values, and obtaining fitness values ​​of corresponding groups of hyperparameter combinations;

[0166] Based on the fitness values ​​corresponding to the multiple sets of hyperparameter combinations, the multiple sets of hyperparameter combinations are sorted in descending order to obtain a plurality of sorted hyperparameter combinations;

[0167] According to a preset offset, the other hyperparameter combinations in the sorted multiple sets of hyperparameter combinations except the multiple sets of target hyperparameter combinations arranged in front are updated to obtain updated hyperparameter combinations;

[0168] Repeat the above steps until the iteration condition is met, and then determine the optimal hyperparameter combination based on the latest sorted multiple hyperparameter combinations.

[0169] In an embodiment of the present invention, each set of hyperparameter combinations is used to initialize the second deep adaptive residual echo state network, multiple sets of second sample data and corresponding true values ​​are used as inputs of the initialized second deep adaptive residual echo state network, and the predicted values ​​corresponding to each set of second sample data are output. Based on the difference between the true value and the predicted value, the output weight of the second deep adaptive residual echo state network is adjusted until the training conditions are met, thereby obtaining a trained third deep adaptive residual echo state network.

[0170] In the embodiment of the present invention, for different hyperparameters, the corresponding preset offsets may be the same or different.

[0171] In an embodiment of the present invention, by continuously updating the arrangement positions of multiple sets of hyperparameter combinations until the iteration conditions are met, multiple sets of hyperparameter combinations with the latest sorting are obtained, and then the hyperparameter combination ranked first is selected as the optimal hyperparameter combination.

[0172] In some embodiments of the present invention, the method of performing data prediction for a preset future period using a first deep adaptive residual echo state network of a Grey Wolf Optimization Algorithm based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target impact factor to obtain a first prediction result includes the following steps:

[0173] S701: Normalize P2 new intrinsic mode functions corresponding to each of the multiple data clusters to obtain a first normalized result.

[0174] S702: Perform normalization processing on at least one target impact factor to obtain a second normalization result.

[0175] In the embodiment of the present invention, the new intrinsic mode function and target impact factor are normalized and mapped to the range [0, 1]. Specifically, the new intrinsic mode function and target impact factor are scaled in a specific interval and converted into dimensionless pure numerical values. The min-max normalization method is used to map the data to the range [0, 1]. The calculation formula is as follows:

[0176]

[0177] Among them, x max 、x min Represent the maximum and minimum values ​​respectively, and x' represents the mapped value.

[0178] S703: Using the first normalization result and the second normalization result as inputs of a first deep adaptive residual echo state network of a gray wolf optimization algorithm, performing data prediction for a preset future time period, and obtaining a second prediction result.

[0179] In an embodiment of the present invention, for the first normalized result and the second normalized result of a historical time period, if the historical time period includes four time points t1, t2, t3, and t4, the encoded vector can be [the first normalized result and the second normalized result of t1; the first normalized result and the second normalized result of t2; the first normalized result and the second normalized result of t3; the first normalized result and the second normalized result of t4]. This vector is then used as the input of the first deep adaptive residual echo state network of the gray wolf optimization algorithm to perform data prediction for a preset future time period, thereby obtaining the second prediction result. The specific contents of the first normalized result can be sorted according to a preset order, and the specific contents of the second normalized result can also be sorted according to a preset order.

[0180] In an embodiment of the present invention, for the first normalized result of the historical period and the second normalized result of the lag period, if the historical period includes four time points t1, t2, t3 and t4, and the lag period on the historical period is t0, then the lag period includes four time points t1+t0, t2+t0, t3+t0 and t4+t0, then the encoded vector can be [the first normalized result of t1, the second normalized result of t1+t0; the first normalized result of t2, the second normalized result of t2+t0; the first normalized result of t3, the second normalized result of t3+t0; the first normalized result of t4, the second normalized result of t4+t0], and then the vector is used as the input of the first deep adaptive residual echo state network of the grey wolf optimization algorithm to perform data prediction for a preset future period to obtain the second prediction result.

[0181] S704: Perform denormalization processing on the second prediction result to obtain the first prediction result.

[0182] Based on the above embodiment, the present invention specifically illustrates a data prediction method. Figure 8 The process diagram of the data prediction method in the embodiment of the present invention is as follows Figure 8 ,like Figure 8 As shown, the data prediction method includes the following steps:

[0183] S801: Obtain relevant data of the thermal management domain in a historical period.

[0184] S802: Perform CDC clustering processing on the relevant data of the thermal management domain in the historical period to obtain multiple data clusters.

[0185] S803: Perform VMD decomposition on each data cluster to obtain P1 IMFs; P1 is an integer greater than or equal to 2.

[0186] S804: Determine the sample entropy of each of P1 IMFs, merge the IMFs whose sample entropy difference is less than a first threshold to form a new IMF, and obtain P2 RIMFs; P2 is an integer greater than or equal to 1.

[0187] The specific implementation of S802 to S804 has been described in the previous embodiment and will not be repeated here.

[0188] S805: Combine at least one target impact factor.

[0189] Here, at least one target influencing factor having the highest impact on relevant data of the thermal management domain in the historical period is determined based on multiple influencing factors in the historical period; wherein the multiple influencing factors include meteorological factors and vehicle-related factors.

[0190] The specific method for determining the target impact factor has been specifically described in the previous embodiments and will not be repeated here.

[0191] S806: Normalization processing.

[0192] Specifically, normalization is performed on the P2 RIMFs and at least one target influencing factor of each data cluster to obtain a first normalization result and a second normalization result.

[0193] S807: GWO-DARESN model.

[0194] Specifically, the first normalization result and the second normalization result are used as inputs of the GWO-DARESN model to perform data prediction for a preset future period.

[0195] S808: Sequence reconstruction + denormalization.

[0196] Specifically, the second prediction result output by the GWO-DARESN model is sequence reconstructed and then denormalized to obtain the first prediction result.

[0197] S809: First prediction result.

[0198] Based on the above embodiment, the present invention further provides a data prediction device, which is applied to vehicle-mounted equipment. Figure 9 FIG. 1 is a schematic diagram showing the structure of the data prediction device according to an embodiment of the present invention. Figure 9 As shown, the data prediction device 90 includes:

[0199] A clustering unit 901 is configured to perform local directional centrality clustering processing on the acquired data related to the thermal management domain in the historical period to obtain multiple data clusters;

[0200] Decomposition unit 902 is used to perform variational mode decomposition on each data cluster to obtain P1 intrinsic mode functions; P1 is an integer greater than or equal to 2;

[0201] A merging unit 903 is configured to determine the sample entropy of each of the P1 eigenmode functions, merge the eigenmode functions whose sample entropy difference is less than a first threshold to form a new eigenmode function, and obtain P2 new eigenmode functions; P2 is an integer greater than or equal to 1;

[0202] A determining unit 904 is configured to determine, based on the plurality of influencing factors in the historical period, at least one target influencing factor having the highest impact on the relevant data of the thermal management domain in the historical period; wherein the plurality of influencing factors include meteorological factors and vehicle-related factors;

[0203] The prediction unit 905 is used to predict data for a preset future time period using a first deep adaptive residual echo state network of the gray wolf optimization algorithm based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target impact factor to obtain a first prediction result.

[0204] In an embodiment of the present invention, a local directional centrality clustering algorithm is applied to accurately cluster data related to the thermal management domain over a historical period, generating multiple data clusters. Next, a variational mode decomposition technique is used to process the data sequence, adaptively decomposing it into several intrinsic mode function (IMF) components. This step effectively reduces the nonstationarity and complexity of the thermal management domain data, providing a more stable input for subsequent modeling. The sample entropy value of each IMF component is calculated to quantify its complexity. Then, based on an entropy similarity criterion, IMF components with similar sample entropy values ​​are merged and reconstructed into a new, more representative IMF, achieving classification and simplified representation of the original information. The new IMF and at least one target influencing factor with the highest impact on the thermal management domain data over the historical period are then input into the first deep adaptive residual echo state network of the Grey Wolf Optimization Algorithm to predict data for a predetermined future period, generating a first prediction result. This enables prediction of data related to the thermal management domain for the future period. Furthermore, the predicted data can be combined to proactively control the thermal management domain, thereby reducing the failure rate and improving the real-time response capability of the thermal management domain.

[0205] In some embodiments of the present invention, the determination unit 904 is specifically used to determine the first maximum information coefficient between the relevant data of the thermal management domain of the historical period and each of the multiple influencing factors of the historical period; based on the first maximum information coefficient corresponding to each of the multiple influencing factors of the historical period, the influencing factor corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0206] In some embodiments of the present invention, the determination unit 904 is specifically used to take the multiple influencing factors of the historical period as a whole to determine the second maximum information coefficient of the relevant data of the thermal management domain of the historical period; based on the first maximum information coefficient corresponding to each influencing factor among the multiple influencing factors of the historical period, and the second maximum information coefficient, the influence factor corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0207] In some embodiments of the present invention, the determination unit 904 is specifically used to obtain multiple influencing factors of at least one lag period; wherein the at least one lag period is a lag period obtained by lagging the historical period by at least a period of time; determine the third maximum information coefficient of the relevant data of the thermal management domain of the historical period and each influencing factor in the multiple influencing factors of each lag period; take the multiple influencing factors of each lag period as a whole, and determine the fourth maximum information coefficient of the relevant data of the thermal management domain of the historical period; based on the first maximum information coefficient corresponding to each influencing factor in the multiple influencing factors of the historical period, and the second maximum information coefficient, and the third maximum information coefficient corresponding to each influencing factor in the multiple influencing factors of each lag period, and the fourth maximum information coefficient corresponding to the multiple influencing factors of each lag period as a whole, screen out a single influencing factor or a group of influencing factors corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0208] In some embodiments of the present invention, the determination unit 904 is specifically used to generate multiple groups of influencing factors based on multiple influencing factors of the historical period; wherein the number of influencing factors included in each group of influencing factors is greater than or equal to 2; determine the fifth maximum information coefficient between the relevant data of the thermal management domain of the historical period and each group of influencing factors; based on the first maximum information coefficient corresponding to each influencing factor in the multiple influencing factors of the historical period, and the fifth maximum information coefficient corresponding to each group of influencing factors in the multiple groups of influencing factors, screen out a single influencing factor or a group of influencing factors corresponding to the maximum value as the target influencing factor with the highest impact on the relevant data of the thermal management domain of the historical period.

[0209] In some embodiments of the present invention, a training unit is further included for defining multiple hyperparameters of an initial second deep adaptive residual echo state network; using the gray wolf optimization algorithm, multiple groups of hyperparameter combinations are randomly generated, and the best hyperparameter combination is selected from the multiple groups of hyperparameter combinations; wherein, the value of at least one hyperparameter in different groups of hyperparameter combinations is different; based on the best hyperparameter combination, multiple groups of first sample data and corresponding true values, the output weights of the second deep adaptive residual echo state network are trained to obtain the trained first deep adaptive residual echo state network of the gray wolf optimization algorithm.

[0210] In some embodiments of the present invention, the training unit is further used to train the output weights of the second deep adaptive residual echo state network based on each group of hyperparameter combinations, multiple groups of second sample data and corresponding true values ​​to obtain a trained third deep adaptive residual echo state network; use multiple groups of third sample data and corresponding true values ​​as inputs of the third deep adaptive residual echo state network, calculate the error between the predicted value and the true value, and obtain the fitness value of the corresponding group of hyperparameter combinations; based on the fitness values ​​corresponding to each of the multiple groups of hyperparameter combinations, sort the multiple groups of hyperparameter combinations in order from large to small to obtain sorted multiple groups of hyperparameter combinations; update the other hyperparameter combinations in the sorted multiple groups of hyperparameter combinations except the multiple groups of target hyperparameter combinations arranged in front according to a preset offset to obtain an updated hyperparameter combination; repeat the above steps until the iteration condition is met, and determine the optimal hyperparameter combination based on the latest sorted multiple groups of hyperparameter combinations.

[0211] In some embodiments of the present invention, the prediction unit 905 is specifically used to normalize the P2 new intrinsic mode functions corresponding to each of the multiple data clusters to obtain a first normalized result; normalize the at least one target influencing factor to obtain a second normalized result; use the first normalized result and the second normalized result as inputs of the first deep adaptive residual echo state network of the gray wolf optimization algorithm to perform data prediction for a preset future time period to obtain a second prediction result; and denormalize the second prediction result to obtain the first prediction result.

[0212] The embodiment of the present invention further provides another vehicle-mounted device, Figure 10 FIG. 1 is a schematic diagram showing the structure of the controller in an embodiment of the present invention. Figure 10 As shown, the vehicle-mounted device 100 includes: a processor 1001 and a memory 1002 configured to store a computer program that can be run on the processor;

[0213] The processor 1001 is configured to execute the method steps in the aforementioned embodiment when running a computer program.

[0214] Of course, in actual application, Figure 10 As shown, the various components in the vehicle-mounted device 100 are coupled together via a bus system 1003. It is understood that the bus system 1003 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 1003 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 10 Various buses are labeled as bus system 1003.

[0215] In practical applications, the processor may be at least one of an application-specific integrated circuit (ASIC), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a controller, a microcontroller, and a microprocessor. It is understood that for different devices, the electronic device used to implement the functions of the processor may also be other, and the embodiments of the present invention do not specifically limit this.

[0216] The above-mentioned memory can be a volatile memory (volatile memory), such as a random-access memory (RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (ROM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor.

[0217] In an exemplary embodiment, the present invention further provides a computer-readable storage medium for storing a computer program.

[0218] Optionally, the computer-readable storage medium can be applied to any one of the methods in the embodiments of the present invention, and the computer program enables the computer to execute the corresponding processes implemented by the processor in each method in the embodiments of the present invention. For the sake of brevity, they are not repeated here.

[0219] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0220] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0221] In addition, the functional units in the embodiments of the present invention can all be integrated into one processing module, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units. It can be understood by those skilled in the art that all or part of the steps of the above-mentioned method embodiments can be completed by hardware related to program instructions, and the above-mentioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiments; and the above-mentioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks or optical disks.

[0222] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0223] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0224] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0225] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A data prediction method, characterized in that: The data prediction method comprises: The relevant data of the thermal management domain in the historical period are clustered by local directional centrality to obtain multiple data clusters; Perform variational mode decomposition on each data cluster to obtain P1 intrinsic mode functions; P1 is an integer greater than or equal to 2; Determine the sample entropy of each of the P1 intrinsic mode functions, combine the intrinsic mode functions whose sample entropy difference is less than a first threshold to form a new intrinsic mode function, and obtain P2 new intrinsic mode functions; P2 is an integer greater than or equal to 1; Determining, based on the plurality of influencing factors of the historical period, at least one target influencing factor having the highest impact on relevant data of the thermal management domain of the historical period; wherein the plurality of influencing factors include meteorological factors and vehicle-related factors; Based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target impact factor, a first deep adaptive residual echo state network of the gray wolf optimization algorithm is used to perform data prediction for a preset future time period to obtain a first prediction result.

2. The data prediction method according to claim 1, characterized in that The determining, based on the multiple influencing factors of the historical period, at least one target influencing factor having the highest impact on the relevant data of the thermal management domain of the historical period includes: Determine a first maximum information coefficient between relevant data of the thermal management domain in the historical period and each of a plurality of influencing factors in the historical period; Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, the influencing factor corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

3. The data prediction method according to claim 2, characterized in that: The first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period is selected, and the influencing factor corresponding to the maximum value is selected as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period, including: Taking the multiple influencing factors of the historical period as a whole, determining a second maximum information coefficient of data related to the thermal management domain of the historical period; Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period and the second maximum information coefficient, the influencing factor corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

4. The data prediction method according to claim 3, characterized in that: The first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period and the second maximum information coefficient are selected to select the influencing factor corresponding to the maximum value as the target influencing factor having the highest impact on the relevant data of the thermal management domain in the historical period, including: Obtaining a plurality of influencing factors of at least one hysteresis period; wherein the at least one hysteresis period is a hysteresis period obtained by lagging the historical period by at least a period of time; determining a third maximum information coefficient between the relevant data of the thermal management domain in the historical period and each of the multiple influencing factors in each hysteresis period; Taking the multiple influencing factors of each hysteresis period as a whole, determining a fourth maximum information coefficient of the data related to the thermal management domain of the historical period; Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, the second maximum information coefficient, the third maximum information coefficient corresponding to each of the multiple influencing factors in each lag period, and the fourth maximum information coefficient corresponding to the multiple influencing factors in each lag period as a whole, a single influencing factor or a group of influencing factors corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

5. The data prediction method according to claim 2, characterized in that: The first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period is selected, and the influencing factor corresponding to the maximum value is selected as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period, including: Based on the multiple impact factors of the historical period, generating multiple groups of impact factors; wherein the number of impact factors included in each group of impact factors is greater than or equal to 2; determining the fifth largest information coefficient between the relevant data of the thermal management domain during the historical period and each group of influencing factors; Based on the first maximum information coefficient corresponding to each of the multiple influencing factors in the historical period, and the fifth maximum information coefficient corresponding to each group of influencing factors in the multiple groups of influencing factors, a single influencing factor or a group of influencing factors corresponding to the maximum value is screened out as the target influencing factor with the highest impact on the relevant data of the thermal management domain in the historical period.

6. The data prediction method according to any one of claims 1 to 5, characterized in that: The data prediction method further includes: defining a plurality of hyperparameters of an initial second deep adaptive residual echo state network; Using the Gray Wolf Optimization Algorithm, randomly generating multiple sets of hyperparameter combinations, and selecting the best hyperparameter combination from the multiple sets of hyperparameter combinations; wherein the value of at least one hyperparameter in different sets of hyperparameter combinations is different; Based on the optimal hyperparameter combination, multiple groups of first sample data and corresponding true values, the output weights of the second deep adaptive residual echo state network are trained to obtain the trained first deep adaptive residual echo state network of the gray wolf optimization algorithm.

7. The data prediction method according to claim 6, characterized in that: The method of randomly generating multiple sets of hyperparameter combinations by using the Gray Wolf Optimization Algorithm and selecting the best hyperparameter combination from the multiple sets of hyperparameter combinations includes: Based on each set of hyperparameter combinations, multiple sets of second sample data and corresponding true values, training the output weights of the second deep adaptive residual echo state network to obtain a trained third deep adaptive residual echo state network; Using multiple groups of third sample data and corresponding true values ​​as inputs of the third deep adaptive residual echo state network, calculating the errors between the predicted values ​​and the true values, and obtaining fitness values ​​of corresponding groups of hyperparameter combinations; Based on the fitness values ​​corresponding to the multiple sets of hyperparameter combinations, the multiple sets of hyperparameter combinations are sorted in descending order to obtain a plurality of sorted hyperparameter combinations; According to a preset offset, the other hyperparameter combinations in the sorted multiple sets of hyperparameter combinations except the multiple sets of target hyperparameter combinations arranged in front are updated to obtain updated hyperparameter combinations; Repeat the above steps until the iteration condition is met, and then determine the optimal hyperparameter combination based on the latest sorted multiple hyperparameter combinations.

8. The data prediction method according to any one of claims 1 to 5, characterized in that: The method of performing data prediction for a preset future period using a first deep adaptive residual echo state network of a Grey Wolf optimization algorithm based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target impact factor to obtain a first prediction result includes: Normalizing the P2 new intrinsic mode functions corresponding to each of the plurality of data clusters to obtain a first normalized result; Normalizing the at least one target impact factor to obtain a second normalized result; Using the first normalized result and the second normalized result as inputs of a first deep adaptive residual echo state network of the gray wolf optimization algorithm, performing data prediction for a preset future time period, and obtaining a second prediction result; Perform denormalization processing on the second prediction result to obtain the first prediction result.

9. A data prediction device, characterized in that: The data prediction device comprises: A clustering unit is used to perform local directional centrality clustering processing on the relevant data of the thermal management domain obtained in the historical period to obtain multiple data clusters; The decomposition unit is used to perform variational mode decomposition on each data cluster to obtain P1 intrinsic mode functions; P1 is an integer greater than or equal to 2; a merging unit, configured to determine the sample entropy of each of the P1 intrinsic mode functions, merge the eigenmode functions whose sample entropy difference is less than a first threshold to form a new eigenmode function, and obtain P2 new eigenmode functions; P2 is an integer greater than or equal to 1; a determining unit, configured to determine, based on a plurality of influencing factors in the historical period, at least one target influencing factor having the highest impact on relevant data of the thermal management domain in the historical period; wherein the plurality of influencing factors include meteorological factors and vehicle-related factors; A prediction unit is used to predict data for a preset future time period using a first deep adaptive residual echo state network of a gray wolf optimization algorithm based on the P2 new intrinsic mode functions corresponding to each of the multiple data clusters and the at least one target impact factor, to obtain a first prediction result.

10. A vehicle-mounted device, characterized in that: The vehicle-mounted device includes: a processor and a memory configured to store a computer program that can be run on the processor, Wherein, the processor is configured to execute the steps of the data prediction method according to any one of claims 1 to 8 when running the computer program.

Images