CN-121980268-A - New energy truck carbon emission model optimization method and system based on neural network

CN121980268ACN 121980268 ACN121980268 ACN 121980268ACN-121980268-A

Abstract

The application relates to the technical field of model training optimization, and discloses a new energy truck carbon emission model optimization method and system based on a neural network, wherein the method comprises the steps of collecting a history record of carbon emission influence characteristics and carbon emission amount of a new energy truck; the method comprises the steps of respectively carrying out response trend analysis on each carbon emission influence characteristic, identifying reversible influence characteristics in the carbon emission influence characteristics, constructing a carbon emission prediction model, defining structural parameters, exploring a group of reference structural parameters for the carbon emission prediction model based on an EPO algorithm, and carrying out iterative training, verification and optimization on the carbon emission prediction model based on the reference structural parameters. The application realizes the improvement of the carbon emission prediction model of the new energy truck in input characteristic compression, structural parameter optimization and abnormal state self-adaption, and enhances the prediction accuracy of carbon emission.

Inventors

JIANG MINGHUI
LU JIANXIN
XU JIAN
ZHANG YUXI

Assignees

江苏零浩网络科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The new energy truck carbon emission model optimization method based on the neural network is characterized by comprising the following steps of: Collecting a history record of carbon emission influence characteristics and carbon emission amount of the new energy truck; Respectively carrying out response trend analysis on each carbon emission influence characteristic, and identifying reversible influence characteristics in the carbon emission influence characteristics; The method comprises the steps of constructing a carbon emission prediction model and defining structural parameters, and exploring a group of reference structural parameters for the carbon emission prediction model based on an EPO algorithm, wherein the method specifically comprises the following steps: Setting a plurality of groups of structural parameters, and carrying out characteristic compression on each reversible influence characteristic; Processing carbon emission influence characteristics based on a carbon emission prediction model, calculating the priority corresponding to each group of structural parameters, and performing iterative optimization on each group of structural parameters through an EPO algorithm; Obtaining a structural parameter corresponding to the highest priority as a reference structural parameter; and performing iterative training, verification and tuning on the carbon emission prediction model based on the reference structural parameters.
2. The method for optimizing the carbon emission model of the new energy truck based on the neural network, which is characterized in that the carbon emission influencing characteristics comprise power system characteristics, driving behavior characteristics, task working condition characteristics and environmental characteristics; the method for response trend analysis of any carbon emission influencing feature is as follows: fitting a regression prediction model of the carbon emission quantity for the carbon emission influence characteristic, wherein the input of the regression prediction model is the carbon emission influence characteristic at any moment, and the input is a predicted value of the carbon emission quantity at the corresponding moment; Calculating a prediction residual error at each moment based on the regression prediction model, and constructing a first-order differential sequence of the prediction residual error, wherein the first-order differential sequence comprises residual error variation of each moment; calculating a residual variation index and a residual fluctuation index based on the first-order differential sequence, wherein the residual variation index is the mean value of the residual variation of each moment, and the residual fluctuation index is the variance of the residual variation of each moment; Setting a residual variation threshold and a residual fluctuation threshold, and if the residual variation index is smaller than the residual variation threshold and the residual fluctuation index is smaller than the residual fluctuation threshold, the corresponding carbon emission influence characteristic is a reversible influence characteristic.
3. The method for optimizing the carbon emission model of the new energy truck based on the neural network of claim 2, wherein the method for analyzing the response trend of any carbon emission influence characteristic further comprises the following steps: Setting a sliding time window with the length of T, wherein T is a positive integer, and performing sliding interception on a first-order differential sequence of the prediction residual error through the sliding time window; calculating residual trend indexes obtained by any two continuous sliding interception, taking absolute values to obtain convergence indexes corresponding to the two sliding interception, setting a convergence threshold, and if the continuous M convergence indexes are smaller than the convergence threshold, the corresponding carbon emission influence characteristics are reversible influence characteristics, and M is a positive integer.
4. The method for optimizing the carbon emission model of the new energy truck based on the neural network, which is characterized in that the input of the carbon emission prediction model is a time sequence of each carbon emission influence characteristic, and the output is a predicted value of carbon emission; The method for compressing the characteristics of each reversible influence characteristic is as follows: the method comprises the steps of constructing a training set and a verification set of a carbon emission prediction model, wherein the training set and the verification set comprise training data with different amounts, and any piece of training data comprises a group of input and corresponding output reference values of the carbon emission prediction model; Training an encoder and a corresponding decoder for the reversibly affected features based on the training set; and performing dimension reduction coding on the time sequence of each reversible influence characteristic in each piece of training data through the coder to obtain a dimension reduction characteristic sequence of each reversible influence characteristic.
5. The method for optimizing the carbon emission model of the new energy truck based on the neural network of claim 4, wherein the method for calculating the priority of any group of structural parameters is as follows: Assigning a weight coefficient and a bias coefficient in the carbon emission prediction model based on the corresponding group of structural parameters; performing r-round iterative training on the carbon emission prediction model through a training set, wherein r is a positive integer; randomly selecting m pieces of training data from the verification set, respectively inputting the m pieces of training data into a carbon emission prediction model, calculating corresponding m pieces of prediction errors, and solving an average value to serve as a prediction performance index, wherein m is a positive integer; And assigning a priority based on the predicted performance index, wherein the priority is inversely related to the predicted performance index.
6. The method for optimizing the carbon emission model of the new energy truck based on the neural network of claim 5, wherein each group of structural parameters is iteratively optimized by an EPO algorithm, and the method specifically comprises the following steps: respectively coding each group of structural parameters to obtain feature vectors of each group of structural parameters, and respectively defining the feature vectors of each group of structural parameters as a position vector of an individual in the EPO algorithm primary population; in any round of iteration, the fitness of each individual is calculated based on the carbon emission prediction model, and an optimal individual is selected, wherein the fitness of any individual is the priority of the structural parameter corresponding to the position vector of the individual; Updating the position vector of each individual based on the position vector of the optimal individual, and entering the next iteration; In the iteration process of the EPO algorithm, identifying whether the iteration convergence trend is abnormal or not based on the fitness and the position vector of the individual; If the iteration convergence trend is abnormal, calculating the compression sensitivity of each reversible influence characteristic, and carrying out decoding reconstruction on the reversible influence characteristics based on the compression sensitivity.
7. The method for optimizing the carbon emission model of the new energy truck based on the neural network of claim 6, wherein the method for identifying whether the iteration convergence trend is abnormal based on the fitness and the position vector of the individual is characterized by comprising the following steps: Extracting data every K rounds of iteration, and extracting K total fitness of the optimal individual in each round of iteration as target fitness, wherein K is a positive integer; Calculating the difference between any two adjacent target fitness in the target fitness sequence to obtain K-1 fitness variable quantities; Setting an iteration optimization threshold value, and if the iteration optimization index is smaller than the iteration optimization threshold value, carrying out iteration convergence trend abnormity.
8. The method for optimizing the carbon emission model of the new energy truck based on the neural network of claim 7, wherein the step of identifying whether the iteration convergence trend is abnormal is further performed by: Extracting data once every K rounds of iteration, and extracting the position vector of each individual in each round of iteration; for any iteration, calculating Euclidean distance between the position vector of each individual and the position vector of the optimal individual, and solving an average value to obtain target optimization distances of corresponding iteration rounds; performing curve fitting on the optimized distance sequence, and extracting a fitting slope as an iteration convergence index; setting an iteration convergence threshold value, and if the iteration convergence index is larger than the iteration convergence threshold value, making the iteration convergence trend abnormal.
9. The method for optimizing the carbon emission model of the new energy truck based on the neural network of claim 8, wherein the method for calculating the compression sensitivity of any reversible influence characteristic is as follows: For any iteration of the EPO algorithm, respectively extracting error gradients of each structural parameter in r-round iterative training of a carbon emission prediction model based on the structural parameter corresponding to the optimal individual, and respectively calculating the average value of r error gradients of each structural parameter to be used as the first sensitivity of each structural parameter; calculating the average value of the second sensitivity of each structural parameter in the recent K iterations of the EPO algorithm to obtain a second sensitivity index of each structural parameter; Calculating the sum of the first sensitive indexes and the sum of the second sensitive indexes of all the associated structural parameters of the reversible influence characteristic respectively, and assigning values for the compression sensitivity of the reversible influence characteristic, wherein the sum of the first sensitive indexes and the sum of the second sensitive indexes of the associated structural parameters are positively correlated with the compression sensitivity; decoding and reconstructing reversible influence characteristics based on compression sensitivity, specifically comprising: And decoding and reconstructing the dimension-reduced feature sequence of the target feature in each piece of training data through the decoder to obtain a reconstructed feature sequence of the target feature.
10. The new energy truck carbon emission model optimization system based on the neural network is used for realizing the new energy truck carbon emission model optimization method based on the neural network as claimed in any one of claims 1-9, and is characterized by comprising a data acquisition module, a feature recognition module, a prediction model module, an optimization module and an adjustment module, wherein the data acquisition module is used for collecting the history record of carbon emission influence features and carbon emission quantity of the new energy truck; The characteristic recognition module is used for carrying out response trend analysis on each carbon emission influence characteristic and recognizing reversible influence characteristics; The prediction model module is used for constructing a carbon emission prediction model, defining structural parameters and carrying out feature compression or decoding reconstruction on each reversible influence feature; the optimization module explores reference structural parameters for the carbon emission prediction model based on an EPO algorithm, and the prediction model module carries out iterative training, verification and optimization on the carbon emission prediction model based on the reference structural parameters; The adjusting module is used for identifying whether the iteration convergence trend of the EPO algorithm is abnormal or not and triggering the prediction model module to reconstruct the decoding of the reversible influence features.

Description

New energy truck carbon emission model optimization method and system based on neural network Technical Field The application relates to the technical field of model training optimization, in particular to a new energy truck carbon emission model optimization method and system based on a neural network. Background Although the new energy truck does not produce tail gas emission in the use process, most of the power sources are regional power grids, and high-carbon power generation modes mainly based on coal power still widely exist in a power grid structure, so that the carbon footprint of the new energy truck is not zero, and more indirect carbon emission is reflected. This phenomenon promotes enterprises and platforms to pay more attention to the problems of carbon emission monitoring, assessment and optimization caused by electricity consumption behaviors. Current research on new energy vehicles carbon emissions mainly focuses on matching electricity consumption estimation with regional carbon factors, but related modeling methods still face some problems. Most carbon emission estimation methods are based on typical working conditions or evaluation data, and an average unit power consumption or piecewise linear regression model is used for estimating the power consumption so as to calculate the carbon emission. Although the method is convenient for large-scale deployment, the time sequence fluctuation characteristics and individual differences in actual operation cannot be reflected, and the method lacks support for actual requirements such as dynamic scheduling, path optimization and the like. There are a large number of disturbance terms and redundant features of the vehicle operation data, such as sensor noise, atypical working condition input, short-time invalid data, etc., and if the disturbance terms and redundant features are directly input into the model, modeling effect is affected. The traditional data cleaning method is difficult to carry out structural processing on vehicle data with complex disturbance, cannot identify which features have reversibility in carbon emission modeling, and which features need to be compressed or removed under the disturbance, and lacks a fine feature processing mechanism. While attempts have been made to model carbon emissions using neural networks, standard BP neural networks, GRU or LSTM structures are typically used, lacking a process of structural parameter optimization that targets carbon emissions. The model structure is usually designed empirically or depends on manual parameter adjustment, so that the performance of the model structure is difficult to ensure to be optimal. For example, china patent with the publication number CN119476684B discloses a vehicle carbon emission safety calculation method and system based on the Internet of vehicles. According to the scheme, the carbon emission is calculated based on the embedded local carbon metering model through the vehicle-mounted carbon metering terminal arranged on each vehicle, the real-time performance and high accuracy of carbon emission monitoring are ensured, and strong safety guarantee is provided in the data transmission process through a traditional encryption algorithm and homomorphic encryption technology, so that the privacy and safety of carbon data are ensured. The carbon metering operation platform is based on the federal learning technology, and performs calculation and optimization of a global carbon emission model on the encrypted data uploaded by the plurality of vehicle-mounted terminals, so that the efficiency and the precision of large-scale data processing are effectively improved. In addition, the system supports various communication networks, can perform data interaction with an external system, provides support for carbon emission report generation and carbon asset application, and meets application requirements in different scenes. The patent application with the publication number of CN118607797A discloses a carbon emission evaluation method in the using stage of a power battery of a new energy automobile, a plurality of new energy automobiles are selected to evaluate the carbon emission performance of the power battery, the evaluation method comprises the steps of determining basic parameters of each new energy automobile and the corresponding power battery, calculating the total energy of the power battery, calculating the carbon emission of the power battery caused by energy loss in the using stage and the carbon emission caused by weight, and evaluating the carbon emission performance of each new energy automobile with the aim of high energy efficiency or light weight by integrating the obtained calculation results of the carbon emission and the carbon emission of the weight. The scheme realizes scientific, efficient and accurate accounting and evaluation of carbon emission in the using stage of the power battery. The technical scheme has the problems that the structural pr