Feature screening method in electric bus energy consumption prediction scene
By screening the key features in the electric bus energy consumption prediction model, a concise energy consumption prediction model is constructed, which solves the problem of high model complexity and improves the prediction efficiency and accuracy.
Patent Information
- Application Number
- CN202510721624.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2025-09-12
AI Technical Summary
In the existing technology, the energy consumption prediction model for electric buses uses all features that affect fuel consumption, resulting in high input dimensions, increased model complexity, and reduced training and prediction efficiency.
By obtaining the data set, determining the correlation coefficient of each feature to be screened, and sorting them in a preset order, generating a feature combination, and using the integrated learning model to screen out features that have a greater impact on energy consumption prediction and have a higher correlation, a concise energy consumption prediction model is constructed.
The input dimension and complexity of the energy consumption prediction model are reduced, the training and prediction efficiency of the model are improved, and the stability and accuracy of the prediction performance are ensured.
Smart Images
Figure CN120632403A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of feature screening, and in particular to a feature screening method in an electric bus energy consumption prediction scenario. Background Art
[0002] Electric buses are playing an increasingly critical role in modern urban public transportation systems. With the global emphasis on environmental protection and sustainable development, electric buses, with their zero-emission and low-noise advantages, are gradually replacing traditional fuel-powered buses and becoming the primary mode of transportation in many first-tier cities.
[0003] In the prior art, to predict the energy consumption of electric buses, ensemble learning models were directly used to predict the energy consumption of electric buses based on all the features that could affect fuel consumption. This resulted in high input feature dimensionality, increased model complexity, and reduced training and prediction efficiency. Therefore, accurately identifying the features that affect fuel consumption has become a technical challenge. Summary of the Invention
[0004] In view of this, the present invention provides a feature screening method in an electric bus energy consumption prediction scenario.
[0005] In the first aspect, the present invention provides a feature screening method in an electric bus energy consumption prediction scenario, the method comprising: obtaining a data set; wherein the data set comprises a plurality of features to be screened and a measured energy consumption rate; determining the correlation coefficient of each feature to be screened, and sorting the correlation coefficient of each feature to be screened in a preset order to obtain a feature combination; wherein determining the correlation coefficient of a target feature to be screened comprises: determining the correlation coefficient of the target feature to be screened according to the target feature to be screened and the measured energy consumption rate, the target feature to be screened being any one of the plurality of features to be screened; generating feature combinations corresponding to different features according to preset rules and feature combinations; The method comprises the following steps: determining the predicted energy consumption rate of the target sample in the target sample set according to the target sample set and an energy consumption prediction model trained by a training set corresponding to the target sample set, determining the predicted energy consumption rate of the target sample in the target sample set, wherein the target sample set is any one of the multiple sample sets; determining the average percentage error corresponding to each sample set according to the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, and taking the feature combination corresponding to the sample set that produces the minimum average percentage error as the target feature combination.
[0006] The feature screening method for electric bus energy consumption prediction provided in this embodiment obtains a data set, then determines the correlation coefficient of each feature to be screened, and sorts the correlation coefficients in a preset order to obtain a feature combination. Then, based on the feature combination according to preset rules, multiple sample sets corresponding to different feature combinations are generated, and the predicted energy consumption rate of the target sample in each sample set is determined. Based on the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, features with a greater impact on energy consumption prediction and a higher correlation can be screened from the numerous features to be screened, while those with a smaller impact or irrelevant features on the prediction results are removed, thereby reducing the feature dimension input into the energy consumption prediction model. Furthermore, only these key features are used to train the energy consumption prediction model, so that the energy consumption prediction model only needs to learn the mapping relationship between these fewer and more relevant features and energy consumption. Compared to using all features, the input dimension of the energy consumption prediction model is reduced, the model structure can be more concise, and the number of parameters is correspondingly reduced, thereby reducing the complexity of the model.
[0007] In one possible implementation, the correlation coefficient of the target feature to be screened is determined based on the target feature to be screened and the measured energy consumption rate, including: determining a first distance matrix corresponding to the target feature to be screened based on the target feature to be screened; determining the average value of the row where each eigenvalue is located and the average value of the column where each eigenvalue is located in the first distance matrix based on the first distance matrix, and determining the centralized eigenvalue corresponding to the target eigenvalue based on the average value of the row where the target eigenvalue is located, the average value of the column where the target eigenvalue is located, the average value of all eigenvalues in the first distance matrix and the target eigenvalue, and constructing a first target matrix based on the centralized eigenvalue; wherein the target eigenvalue is any one of all eigenvalues; root According to the measured energy consumption rate, determine the second distance matrix corresponding to the measured energy consumption rate; according to the second distance matrix, determine the average value of each measured energy consumption rate in the row and the average value of each measured energy consumption rate in the column in the second distance matrix, and according to the average value of the target measured energy consumption rate in the row, the average value of the target measured energy consumption rate in the column, the average value of all measured energy consumption rates in the second distance matrix and the target measured energy consumption rate, determine the centralized measured energy consumption rate corresponding to the target measured energy consumption rate, and construct a second target matrix based on the centralized measured energy consumption rate; wherein, the target measured energy consumption rate is any one of all the measured energy consumption rates; according to the first target matrix and the second target matrix, determine the correlation coefficient of the target feature to be screened.
[0008] The feature screening method provided in the electric bus energy consumption prediction scenario in this embodiment comprehensively considers the distribution of the target to-be-screened features and the measured energy consumption rate in the entire data set by constructing a distance matrix and calculating the average value of the row and column where each eigenvalue is located based on the distance matrix. It not only pays attention to the relative position of a single eigenvalue in its row and column, but also combines the average value of all eigenvalues to grasp the characteristics of the data from a global perspective, so that the calculated correlation coefficient can more accurately reflect the intrinsic connection between the target to-be-screened features and the measured energy consumption rate, avoiding the deviation that may be caused by analyzing from only a single perspective. In addition, the eigenvalues and the measured energy consumption rate are centered to obtain centralized eigenvalues and centralized measured energy consumption rates, eliminating the influence of the mean value of the data and making the data calculated with zero as the center. The centralized data can better highlight the relative changes in the data, reduce the impact of the overall level differences of the data on the calculation of the correlation coefficient, and improve the stability and reliability of the correlation coefficient.
[0009] In one possible implementation, the predicted energy consumption rate of the target samples in the target sample set is determined based on the target sample set and the energy consumption prediction model trained by the training set corresponding to the target sample set, including: dividing the training set and the validation set according to the target sample set; training the energy consumption prediction model based on ensemble learning according to the training set, and inputting the features of the validation set into the energy consumption prediction model to obtain the predicted energy consumption rate of each target sample in the sample set by the energy consumption prediction model.
[0010] The feature screening method provided in the electric bus energy consumption prediction scenario in this embodiment can train an energy consumption prediction model based on ensemble learning according to the training set by dividing the training set into a training set and a validation set, so as to accurately determine the predicted energy consumption rate of each target sample in the sample set by the energy consumption prediction model.
[0011] In one possible implementation, the average percentage error corresponding to each sample set is determined based on the measured energy consumption rate of each sample in each sample set and the predicted energy consumption rate of each sample in each sample set, including: determining the average percentage error of the validation set corresponding to the sample set based on the measured energy consumption rate of each sample in the validation set corresponding to the sample set and the predicted energy consumption rate of each sample in the validation set corresponding to the sample set; performing 10-fold cross-validation to obtain the average percentage error of the sample set.
[0012] The feature screening method provided in this embodiment for the electric bus energy consumption prediction scenario performs 10-fold cross validation, which can accurately obtain the average percentage error of the sample set and further accurately determine the target feature combination.
[0013] In a possible implementation, the types of the multiple features to be screened in the data set include at least one of the following: traffic condition type, environmental factor type, vehicle state type, and driver behavior type.
[0014] In one possible implementation, the method further includes: using an ensemble learning model to determine, based on the target feature combination, an importance coefficient corresponding to each feature in the target feature combination; determining a probability of the target feature based on the importance coefficient corresponding to the target feature and the importance coefficient corresponding to each feature in the target feature combination; determining a filtered feature based on the probability of each target feature, and constructing a first target feature combination based on other features in the target feature combination excluding the to-be-filtered feature; wherein the filtered feature is a feature to be filtered out in the target feature combination; determining, based on the first target feature combination, a predicted energy consumption rate corresponding to the first target feature combination; predicting the energy consumption rate based on the measured energy consumption rate, and determining an average percentage error corresponding to the first target feature combination; detecting whether the average percentage error corresponding to the first target feature combination is less than the average percentage error corresponding to the target feature combination; and if the average percentage error corresponding to the first target feature combination is less than the average percentage error corresponding to the target feature combination, repeatedly performing the steps of determining, based on the target feature combination, the importance coefficient corresponding to each feature in the target feature combination to detecting whether the average percentage error corresponding to the first target feature combination is less than the average percentage error corresponding to the target feature combination using the ensemble learning model until the average percentage error corresponding to the first target feature combination is greater than the average percentage error corresponding to the target feature combination, thereby obtaining a second target feature combination.
[0015] The feature screening method for electric bus energy consumption prediction, provided in this embodiment, utilizes an ensemble learning model to determine the importance coefficient for each feature in the target feature combination. This method leverages the strengths of multiple basic learners to provide a more comprehensive and accurate assessment of feature importance. Compared to a single model, it can capture more complex relationships between features and energy consumption in the data, reducing importance assessment errors caused by model bias. Furthermore, the probability of the target feature is determined based on the importance coefficient of each feature, and features are then filtered out based on this probability. This probability-based screening approach offers a degree of randomness and flexibility. Rather than simply eliminating features based on their importance coefficients, it considers probability, avoiding premature elimination of features that may be potentially important in specific circumstances. Furthermore, the second target feature combination, obtained after multiple iterative optimizations, minimizes the number of features while ensuring that the average percentage error of the predicted energy consumption rate does not increase. This feature combination effectively reduces model complexity while ensuring stable and reliable prediction performance, avoiding overfitting and underfitting issues.
[0016] In the second aspect, the present invention provides a feature screening device in the energy consumption prediction scenario of an electric bus, the device comprising: an acquisition module for acquiring a data set; wherein the data set comprises a plurality of features to be screened and a measured energy consumption rate; a first determination module for determining the correlation coefficient of each feature to be screened, and sorting the correlation coefficients of each feature to be screened in a preset order to obtain a feature combination; wherein determining the correlation coefficient of the target feature to be screened comprises: determining the correlation coefficient of the target feature to be screened according to the target feature to be screened and the measured energy consumption rate, wherein the target feature to be screened is any one of the plurality of features to be screened; a second determination module for determining the correlation coefficient of each feature to be screened according to preset rules, feature combinations, and the like. combination, generating multiple sample sets corresponding to different feature combinations, and determining the predicted energy consumption rate of the target sample in each sample set; wherein, determining the predicted energy consumption rate of the target sample in the target sample set includes: determining the predicted energy consumption rate of the target sample in the target sample set according to the target sample set and the energy consumption prediction model trained by the training set corresponding to the target sample set, wherein the target sample set is any one of the multiple sample sets; a third determination module is used to determine the average percentage error corresponding to each sample set according to the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, and taking the feature combination corresponding to the sample set that produces the minimum average percentage error as the target feature combination.
[0017] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute the feature screening method in the electric bus energy consumption prediction scenario of the above-mentioned first aspect or any corresponding embodiment thereof.
[0018] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the feature screening method in the electric bus energy consumption prediction scenario of the above-mentioned first aspect or any corresponding embodiment thereof.
[0019] In a fifth aspect, the present invention provides a computer program product comprising computer instructions for enabling a computer to execute the feature screening method in an electric bus energy consumption prediction scenario according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0021] Figure 1 2 is a flow chart of a feature screening method in an electric bus energy consumption prediction scenario according to an embodiment of the present invention;
[0022] Figure 2 is a structural block diagram of a feature screening device in an electric bus energy consumption prediction scenario according to an embodiment of the present invention;
[0023] Figure 3 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0024] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.
[0025] According to an embodiment of the present invention, an embodiment of a feature screening method in an electric bus energy consumption prediction scenario is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0026] In this embodiment, a feature screening method for electric bus energy consumption prediction is provided, which can be used for computer equipment, such as computers, servers, etc. Figure 1 FIG. 1 is a flow chart of a feature screening method in an electric bus energy consumption prediction scenario according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:
[0027] Step S101 , obtaining a data set; wherein the data set includes a plurality of features to be screened and a measured energy consumption rate.
[0028] The data set may include a plurality of to-be-screened features that affect the energy consumption of the electric bus and a measured energy consumption rate of the electric bus, wherein the measured energy consumption rate indicates the energy consumption rate of the electric bus actually measured.
[0029] As an example, the types corresponding to the multiple features to be screened may include at least one of a traffic condition type, an environmental factor type, a vehicle state type, and a driver behavior type. Accordingly, the multiple features to be screened may include at least one of a traffic condition, an environmental factor, a vehicle state, and a driver behavior.
[0030] As an example, Table 1 shows various sub-items of each feature to be screened among multiple features to be screened.
[0031] Table 1
[0032]
[0033]
[0034] As shown in Table 1, a total of 25 features were extracted from four aspects: traffic conditions, environmental factors, vehicle status, and driver behavior. Among them, the definitions of departure time and air conditioning on status need to be explained in particular. The departure time is defined as the hour at the start of the trip, accurate only to the hour. For example, for a certain trip, if more than half of the time is between 8 and 9 o'clock, the departure time of the trip is set to 8. Regarding the air conditioning on status of the trip, it is defined as: if the air conditioning is on for more than half of the total trip time, the air conditioning on status is set to 1; otherwise, it is set to 0. In particular, the impact of battery health status (SOH) on the energy consumption of electric buses is also considered.
[0035] Step S102, determining the correlation coefficient of each feature to be screened, and sorting the correlation coefficient of each feature to be screened according to a preset order to obtain a feature combination; wherein, determining the correlation coefficient of the target feature to be screened includes: determining the correlation coefficient of the target feature to be screened according to the target feature to be screened and the measured energy consumption rate, and the target feature to be screened is any one of the multiple features to be screened.
[0036] The correlation coefficient may indicate a statistic that measures the association between two random variables.
[0037] The preset order may be a pre-set order, wherein the preset order may be an order from largest to smallest or an order from smallest to largest, which is not specifically limited here.
[0038] The target feature to be screened is any one of the multiple features to be screened. That is, when the correlation coefficient of the feature to be screened is determined, the feature to be screened is the target feature to be screened.
[0039] For each feature to be screened, a correlation coefficient of each feature to be screened can be determined. After determining the correlation coefficient of each feature to be screened, the correlation coefficients of each feature to be screened are sorted in a preset order to obtain a feature combination. For example, after determining the correlation coefficient of each feature to be screened, the correlation coefficients are sorted in descending order based on their magnitude.
[0040] As an example, the correlation coefficient of each feature to be screened can be determined using the distance correlation coefficient method, or the Pearson correlation coefficient method, or other methods. No specific limitations are given here, and it can be implemented by those skilled in the art.
[0041] Step S103, based on preset rules and feature combinations, generates multiple sample sets corresponding to different feature combinations, and determines the predicted energy consumption rate of the target sample in each sample set; wherein, determining the predicted energy consumption rate of the target sample in the target sample set includes: determining the predicted energy consumption rate of the target sample in the target sample set based on the target sample set and the energy consumption prediction model trained by the training set corresponding to the target sample set, and the target sample set is any one of the multiple sample sets.
[0042] The preset rule may be a pre-set rule. The preset rule may indicate a method for generating multiple sample sets corresponding to different feature combinations based on feature combinations. For example, the top 25, top 20, top 17, and top 15 features in the feature combination may be generated as multiple sample sets, respectively.
[0043] The target sample set is any one of the multiple sample sets. When the sample set is used to determine the predicted energy consumption rate of the target sample in the sample set, the sample set is the target sample set.
[0044] The energy consumption prediction model may be a model trained using a training set corresponding to the target sample set. The energy consumption prediction model may be an XGBoost model. After determining the energy consumption prediction model, the predicted energy consumption rate of the target sample in the target sample set may be determined.
[0045] In step S104, based on the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, the average percentage error corresponding to each sample set is determined, and the feature combination corresponding to the sample set that produces the minimum average percentage error is used as the target feature combination.
[0046] The target feature combination can be a combination that accurately determines the energy consumption rate. After determining the measured energy consumption rate for each sample in the sample set, the average percentage error corresponding to each sample set is further determined. This is achieved using a ten-fold cross-validation approach. The feature combination corresponding to the sample set that produces the smallest average percentage error is then selected as the target feature combination.
[0047] The feature screening method for electric bus energy consumption prediction provided in this embodiment obtains a data set, then determines the correlation coefficient of each feature to be screened, and sorts the correlation coefficients in a preset order to obtain a feature combination. Then, based on the feature combination according to preset rules, multiple sample sets corresponding to different feature combinations are generated, and the predicted energy consumption rate of the target sample in each sample set is determined. Based on the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, features with a greater impact on energy consumption prediction and a higher correlation can be screened from the numerous features to be screened, while those with a smaller impact or irrelevant features on the prediction results are removed, thereby reducing the feature dimension input into the energy consumption prediction model. Furthermore, only these key features are used to train the energy consumption prediction model, so that the energy consumption prediction model only needs to learn the mapping relationship between these fewer and more relevant features and energy consumption. Compared to using all features, the input dimension of the energy consumption prediction model is reduced, the model structure can be more concise, and the number of parameters is correspondingly reduced, thereby reducing the complexity of the model.
[0048] In one possible implementation, determining the correlation coefficient of the target feature to be screened based on the target feature to be screened and the measured energy consumption rate in step S102 includes:
[0049] Step S1021: determining a first distance matrix corresponding to the target feature to be screened according to the target feature to be screened.
[0050] The target feature to be screened may indicate the characteristics of different driving stages of the electric bus. For example, the target feature to be screened may be average speed, where the average speed may be [20, 24, 28]. The average speed is denoted as X, i.e., X = [20, 24, 28].
[0051] According to the target feature to be screened, the process of determining the first distance matrix corresponding to the target feature to be screened can be determined using the following formula:
[0052] A ij =|X i -X j |; Among them, A ij is the i-th row and j-th element, X i Refers to the i-th element of the variable, X jRefers to the jth element of the variable. Taking the average speed as an example, the average speed can be [20, 24, 28], where X i It represents the average speed of the i-th data in n data. So the matrix of average speed can be obtained as:
[0053]
[0054] Step S1022: Determine, based on the first distance matrix, the average value of the row where each eigenvalue is located, the average value of the column where each eigenvalue is located, and determine, based on the average value of the row where the target eigenvalue is located, the average value of the column where the target eigenvalue is located, the average value of all eigenvalues in the first distance matrix, and the target eigenvalue, the centralized eigenvalue corresponding to the target eigenvalue, and construct a first target matrix based on the centralized eigenvalue; wherein, the target eigenvalue is any one of all the eigenvalues.
[0055] The first target matrix may indicate a matrix obtained by centering each eigenvalue in the first distance matrix, wherein for each eigenvalue in the first target matrix, the eigenvalue after the centering needs to be determined.
[0056] As an example, the centralized eigenvalue corresponding to the target eigenvalue is determined based on the average value of the row where the target eigenvalue is located, the average value of the column where the target eigenvalue is located, the average value of all eigenvalues in the first distance matrix, and the target eigenvalue, which can be achieved using an integrated learning model.
[0057] As an example, according to the average value of the row where the target eigenvalue is located, the average value of the column where the target eigenvalue is located, the average value of all eigenvalues in the first distance matrix, and the target eigenvalue, the centralized eigenvalue corresponding to the target eigenvalue is determined, and the first target matrix is constructed according to the centralized eigenvalue corresponding to the target eigenvalue. The following formula can be used for calculation:
[0058] in, is the mean value of row i, is the mean of the jth column, is the global average value of the matrix, and A' is the first target matrix composed of the centralized eigenvalues corresponding to the target eigenvalues.
[0059] For example, the average speed can be [20, 24, 28]. Then the matrix obtained after centering is:
[0060]
[0061] Step S1023: Determine a second distance matrix corresponding to the measured energy consumption rate according to the measured energy consumption rate.
[0062] The measured energy consumption rate can indicate the energy consumption rate of the electric bus during movement. According to the measured energy consumption rate, the second distance matrix corresponding to the measured energy consumption rate can be determined using the following formula:
[0063] B ij =|Y i -Y j |; Among them, B ij is the i-th row and j-th element of the energy consumption rate, Y i Refers to the i-th element of the energy consumption rate, Y j Refers to the jth element of the energy consumption rate. For example, the energy consumption rate can be [0.59, 0.62, 0.67], where the second distance matrix that can be calculated is:
[0064]
[0065] Step S1024, based on the second distance matrix, determine the average value of each measured energy consumption rate in the row and the average value of each measured energy consumption rate in the column in the second distance matrix, and determine the centralized measured energy consumption rate corresponding to the target measured energy consumption rate based on the average value of the row in which the target measured energy consumption rate is located, the average value of the column in which the target measured energy consumption rate is located, the average value of all measured energy consumption rates in the second distance matrix, and the target measured energy consumption rate, and construct a second target matrix based on the centralized measured energy consumption rate; wherein, the target measured energy consumption rate is any one of all the measured energy consumption rates.
[0066] The second target matrix may indicate a matrix obtained by centering each eigenvalue in the second distance matrix, wherein for each eigenvalue in the second target matrix, the eigenvalue after the centering needs to be determined.
[0067] As an example, the centralized measured energy consumption rate corresponding to the target measured energy consumption rate can be determined based on the average value of the row where the target measured energy consumption rate is located, the average value of the column where the target measured energy consumption rate is located, the average value of all measured energy consumption rates in the second distance matrix, and the target measured energy consumption rate. This can be achieved using an integrated learning model.
[0068] As an example, the centralized measured energy consumption rate corresponding to the target measured energy consumption rate is determined based on the average value of the row where the target measured energy consumption rate is located, the average value of the column where the target measured energy consumption rate is located, the average value of all measured energy consumption rates in the second distance matrix, and the target measured energy consumption rate. The following formula can be used for calculation:
[0069] Among them, B' is the second target matrix composed of the centralized measured energy consumption rates corresponding to multiple target measured energy consumption rates, is the mean value of row i, is the mean of the jth column, is the global mean of the matrix.
[0070] Taking the above second distance matrix as an example, the second target matrix can be:
[0071]
[0072] Step S1025 : determining the correlation coefficient of the target feature to be screened based on the first target matrix and the second target matrix.
[0073] According to the first target matrix and the second target matrix, the correlation coefficient of the target feature to be screened can be determined using the following formula:
[0074] Where dCov(A',B') is the covariance, ||A'|| 2 and ||B'|| 2 is the square of the Frobenius norm. The value of the correlation coefficient is between 0 and 1, where 0 indicates no correlation and 1 indicates perfect correlation.
[0075] Take the average speed and measured energy consumption rate as an example:
[0076]
[0077] Please refer to Table 2, which shows the correlation coefficients of some features.
[0078] Table 2
[0079]
[0080] The feature screening method provided in the electric bus energy consumption prediction scenario in this embodiment comprehensively considers the distribution of the target to-be-screened features and the measured energy consumption rate in the entire data set by constructing a distance matrix and calculating the average value of the row and column where each eigenvalue is located based on the distance matrix. It not only pays attention to the relative position of a single eigenvalue in its row and column, but also combines the average value of all eigenvalues to grasp the characteristics of the data from a global perspective, so that the calculated correlation coefficient can more accurately reflect the intrinsic connection between the target to-be-screened features and the measured energy consumption rate, avoiding the deviation that may be caused by analyzing from only a single perspective. In addition, the eigenvalues and the measured energy consumption rate are centered to obtain centralized eigenvalues and centralized measured energy consumption rates, eliminating the influence of the mean value of the data and making the data calculated with zero as the center. The centralized data can better highlight the relative changes in the data, reduce the impact of the overall level differences of the data on the calculation of the correlation coefficient, and improve the stability and reliability of the correlation coefficient.
[0081] In one possible implementation, in step S103, determining the predicted energy consumption rate of the target sample in the target sample set according to the target sample set and the energy consumption prediction model trained by the training set corresponding to the target sample set includes:
[0082] Step S1031: Divide the target sample set into a training set and a validation set.
[0083] The target sample set can be divided into 10 subsets, one of which can be selected as a validation set, and the remaining 9 subsets can be combined as training sets.
[0084] Step S1032: train an energy consumption prediction model based on ensemble learning according to the training set, and input the features of the validation set into the energy consumption prediction model to obtain the predicted energy consumption rate of each target sample in the sample set by the energy consumption prediction model.
[0085] After dividing the target sample set into a training set and a validation set, the energy consumption prediction model based on ensemble learning is trained by the training set corresponding to the target sample set, and the features of the validation set are input into the energy consumption prediction model to obtain the predicted energy consumption rate of each target sample in the sample set by the energy consumption prediction model.
[0086] The feature screening method provided in the electric bus energy consumption prediction scenario in this embodiment can train an energy consumption prediction model based on ensemble learning according to the training set by dividing the training set into a training set and a validation set, so as to accurately determine the predicted energy consumption rate of each target sample in the sample set by the energy consumption prediction model.
[0087] In one possible implementation, in step S103, the average percentage error corresponding to each sample set is determined based on the measured energy consumption rate of each sample in each sample set and the predicted energy consumption rate of each sample in each sample set, including:
[0088] Step S1033 , determining the average percentage error of the validation set corresponding to the sample set based on the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set.
[0089] Step S1034: perform 10-fold cross validation to obtain the average percentage error of the sample set.
[0090] After determining the predicted energy consumption rate for each sample in the sample set, the average percentage error of the validation set can be determined based on the predicted energy consumption rate of each sample in each sample set and the measured energy consumption rate of each sample in the sample set. A 10-fold cross-validation is performed so that different validation sets can determine the predicted energy consumption rate of each sample in the validation set, thereby obtaining the average percentage error of the sample set.
[0091] The feature screening method provided in this embodiment for the electric bus energy consumption prediction scenario performs 10-fold cross validation, which can accurately obtain the average percentage error of the sample set and further accurately determine the target feature combination.
[0092] In one possible implementation, the method further includes:
[0093] Step S201: using an integrated learning model, according to the target feature combination, respectively determine the importance coefficient corresponding to each feature in the target feature combination.
[0094] The ensemble learning model can include models with feature importances (feature_importances_) or coefficients (coef_) (e.g., Xgboost, LightGBM, etc.). After determining the ensemble learning model, the importance coefficients corresponding to each feature in the target feature combination can be determined separately based on the target feature combination. The target feature combination can be the input of the ensemble learning model, and the importance coefficients corresponding to each feature can be the output of the ensemble learning model.
[0095] Step S202 : determining the probability of the target feature according to the importance coefficient corresponding to the target feature in the target feature combination and the importance coefficient corresponding to each feature.
[0096] After determining the importance coefficient of the target feature in the target feature combination, the probability of the target feature can be determined using the following formula:
[0097] Among them, P i is the probability of the i-th feature, F i is the importance coefficient of the i-th feature.
[0098] Step S203, determining the filtered features according to the probability of each target feature, and constructing a first target feature combination according to other features in the target feature combination except the feature to be filtered out; wherein the filtered features are the features to be filtered out in the target feature combination.
[0099] By comparing the probability of each target feature, the target feature with the smallest probability is determined as the filtered feature from the probabilities of each target feature, and then the other features except the filtered feature are determined as the first target feature combination.
[0100] Step S204: determining a predicted energy consumption rate corresponding to the first target feature combination according to the first target feature combination.
[0101] After determining the first target feature combination, the predicted energy consumption rate corresponding to the first target feature combination is determined based on the first target feature combination. Please refer to the above step S103 for details, and no further details will be given here.
[0102] Step S205: Based on the measured energy consumption rate, the energy consumption rate is predicted and the average percentage error corresponding to the first target feature combination is determined. Please refer to the above step S104 for details and will not be described in detail here.
[0103] Step S206, detecting whether the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, and if the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, repeatedly executing the steps of using the ensemble learning model to determine the importance coefficient corresponding to each feature in the target feature combination according to the target feature combination to detecting whether the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, until the average percentage error corresponding to the first target feature combination is greater than the average percentage error corresponding to the target feature combination, thereby obtaining the second target feature combination.
[0104] If the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, the characterization further requires repeating steps S201 to S205 until the average percentage error corresponding to the first target feature combination is larger than the average percentage error corresponding to the target feature combination (i.e., the model prediction accuracy corresponding to the first target feature combination deteriorates), thereby obtaining a second target feature combination. The second target feature combination indicates a combination for predicting energy consumption.
[0105] The second target feature set can include 11 features: air conditioning status, average speed, driving distance, holidays, number of stops, congestion index, average acceleration, average deceleration, outside temperature, inside temperature, and direction. Compared to the original 25 features, the number of features after screening was reduced by 56%, while the model's prediction performance was improved by 21%. This demonstrates that this method significantly reduces computing resource consumption while improving prediction accuracy, and has important practical application value.
[0106] As an example, the present invention uses the feature importance parameters obtained according to the XGBoost model (based on the trained model, the importance parameters of each feature can be obtained according to importances=model.feature_importances_) as an indicator for evaluating importance. Subsequently, an exponential probability is assigned to the features according to these importance indicators (the higher the feature importance, the smaller the probability of being removed), and the features are gradually removed according to the probability until each feature in the remaining feature set contributes significantly to the model performance. In the present invention, a specific algorithm stopping condition is set as follows: if continuing to eliminate any feature causes the model prediction effect to drop significantly, the feature elimination process is stopped immediately. Through this recursive elimination mechanism, Probability based-RFE can effectively select the feature subset that contributes most to the model performance. Finally, after Probability based-RFE processing, the remaining 11 features constitute the optimal feature combination in the electric bus energy consumption prediction scenario. The following is the pseudo code of Probability based-RFE:
[0107] Input parameters:
[0108] model: a model with feature importance (feature_importances_) or coefficients (coef_) (e.g., ensemble learning models like Xgboost and LightGBM);
[0109] beta: A parameter that controls the steepness of the probability distribution (the default value is 1.0), which can be set by the user according to needs.
[0110] initialization:
[0111] Store the given model and beta parameters as attributes of the class;
[0112] Create a Boolean array "support" with the same length as the number of features, and initialize it to "True", indicating that all features are initially selected;
[0113] Create an integer array "ranking" of length equal to the number of features.
[0114] Feature screening process:
[0115] Read a dataset containing all features and label values (a dataset containing 17 features and an energy consumption rate label);
[0116] Repeat the following steps until the model's prediction error (such as mean absolute percentage error (MAPE)) increases after removing any remaining features.
[0117] - Get the current remaining features;
[0118] -Based on the read dataset, the model is trained using the current remaining features;
[0119] -Calculate the importance of each feature (feature_importances_ or coef_);
[0120] -Calculate the probability P of each feature i (The subscript i represents the i-th feature, and the importance coefficient of the i-th feature is recorded as Fi):
[0121]
[0122] - Randomly select features to be deleted based on a probability distribution;
[0123] -Mark the deleted feature and set the corresponding element in the "support" array to "False";
[0124] -Update the feature ranking array "ranking" based on feature importance.
[0125] The feature screening method for electric bus energy consumption prediction, provided in this embodiment, utilizes an ensemble learning model to determine the importance coefficient for each feature in the target feature combination. This method leverages the strengths of multiple basic learners to provide a more comprehensive and accurate assessment of feature importance. Compared to a single model, it can capture more complex relationships between features and energy consumption in the data, reducing importance assessment errors caused by model bias. Furthermore, the probability of the target feature is determined based on the importance coefficient of each feature, and features are then filtered out based on this probability. This probability-based screening approach offers a degree of randomness and flexibility. Rather than simply eliminating features based on their importance coefficients, it considers probability, avoiding premature elimination of features that may be potentially important in specific circumstances. Furthermore, the second target feature combination, obtained after multiple iterative optimizations, minimizes the number of features while ensuring that the average percentage error of the predicted energy consumption rate does not increase. This feature combination effectively reduces model complexity while ensuring stable and reliable prediction performance, avoiding overfitting and underfitting issues.
[0126] This embodiment also provides a feature screening device for an electric bus energy consumption prediction scenario. This device is used to implement the above-mentioned embodiments and preferred embodiments. Details already described will not be repeated here. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.
[0127] This embodiment provides a feature screening device for electric bus energy consumption prediction scenario, such as Figure 2 As shown, it includes: an acquisition module 201 for acquiring a data set; wherein the data set includes multiple features to be screened and measured energy consumption rates; a first determination module 202 for determining the correlation coefficient of each feature to be screened, and sorting the correlation coefficients of each feature to be screened in a preset order to obtain a feature combination; wherein determining the correlation coefficient of the target feature to be screened includes: determining the correlation coefficient of the target feature to be screened according to the target feature to be screened and the measured energy consumption rate, wherein the target feature to be screened is any one of the multiple features to be screened; a second determination module 203 for generating multiple corresponding feature combinations according to preset rules and feature combinations. A sample set, and determining the predicted energy consumption rate of the target sample in each sample set; wherein, determining the predicted energy consumption rate of the target sample in the target sample set includes: determining the predicted energy consumption rate of the target sample in the target sample set according to the target sample set and the energy consumption prediction model trained by the training set corresponding to the target sample set, where the target sample set is any one of the multiple sample sets; a third determination module 204 is used to determine the average percentage error corresponding to each sample set according to the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, and taking the feature combination corresponding to the sample set that produces the minimum average percentage error as the target feature combination.
[0128] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.
[0129] The feature screening device in the electric bus energy consumption prediction scenario in this embodiment is presented in the form of a functional unit, where the functional unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.
[0130] The embodiment of the present invention also provides a computer device having the above Figure 2 The feature screening device in the electric bus energy consumption prediction scenario is shown.
[0131] See also Figure 3 , Figure 3 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 3As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 3 A processor 10 is taken as an example.
[0132] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.
[0133] The memory 20 stores instructions that can be executed by at least one processor 10, so as to enable at least one processor 10 to execute the method shown in the above embodiment.
[0134] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0135] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.
[0136] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 3 The bus connection is taken as an example.
[0137] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touch pad, an indicator stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display, and a plasma display. In some optional embodiments, the display device can be a touch screen.
[0138] The computer device further includes a communication interface for the computer device to communicate with other devices or a communication network.
[0139] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.
[0140] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.
[0141] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.
Claims
1. A feature screening method for electric bus energy consumption prediction scenario, characterized in that: The method comprises: Acquire a data set; wherein the data set includes a plurality of features to be screened and a measured energy consumption rate; Determining the correlation coefficient of each feature to be screened, and sorting the correlation coefficients of each feature to be screened according to a preset order to obtain a feature combination; wherein determining the correlation coefficient of the target feature to be screened includes: determining the correlation coefficient of the target feature to be screened according to the target feature to be screened and the measured energy consumption rate, wherein the target feature to be screened is any one of the multiple features to be screened; Generate multiple sample sets corresponding to different feature combinations according to preset rules and feature combinations, and determine a predicted energy consumption rate of a target sample in each sample set; wherein determining the predicted energy consumption rate of the target sample in the target sample set includes: determining the predicted energy consumption rate of the target sample in the target sample set according to the target sample set and an energy consumption prediction model trained by a training set corresponding to the target sample set, wherein the target sample set is any one of the multiple sample sets; According to the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, the average percentage error corresponding to each sample set is determined respectively, and the feature combination corresponding to the sample set that produces the minimum average percentage error is used as the target feature combination.
2. The feature screening method for electric bus energy consumption prediction scenario according to claim 1 is characterized in that: Determining the correlation coefficient of the target feature to be screened based on the target feature to be screened and the measured energy consumption rate includes: Determining a first distance matrix corresponding to the target feature to be screened according to the target feature to be screened; Determine, based on the first distance matrix, the average value of the row where each eigenvalue is located and the average value of the column where each eigenvalue is located in the first distance matrix, and determine, based on the average value of the row where the target eigenvalue is located, the average value of the column where the target eigenvalue is located, the average value of all eigenvalues in the first distance matrix, and the target eigenvalue, a centralized eigenvalue corresponding to the target eigenvalue, and construct a first target matrix based on the centralized eigenvalue; wherein the target eigenvalue is any one of all eigenvalues; Determining, according to the measured energy consumption rate, a second distance matrix corresponding to the measured energy consumption rate; Determine, based on the second distance matrix, the average value of each measured energy consumption rate in the row and the average value of each measured energy consumption rate in the column in the second distance matrix, and determine, based on the average value of the target measured energy consumption rate in the row and the average value of the target measured energy consumption rate in the column, the average value of all measured energy consumption rates in the second distance matrix, and the target measured energy consumption rate, a centralized measured energy consumption rate corresponding to the target measured energy consumption rate, and construct a second target matrix based on the centralized measured energy consumption rate; wherein the target measured energy consumption rate is any one of all the measured energy consumption rates; Determine the correlation coefficient of the target feature to be screened based on the first target matrix and the second target matrix.
3. The feature screening method for electric bus energy consumption prediction scenario according to claim 1 is characterized in that: The step of determining the predicted energy consumption rate of the target samples in the target sample set based on the target sample set and the energy consumption prediction model trained by the training set corresponding to the target sample set includes: According to the target sample set, divide the training set and validation set; According to the training set, an energy consumption prediction model based on ensemble learning is trained, and the features of the validation set are input into the energy consumption prediction model to obtain the predicted energy consumption rate of each target sample in the sample set by the energy consumption prediction model.
4. The feature screening method for electric bus energy consumption prediction scenario according to claim 3 is characterized in that: Based on the measured energy consumption rate of each sample in each sample set and the predicted energy consumption rate of each sample in each sample set, the average percentage error corresponding to each sample set is determined, including: Determine the average percentage error of the validation set corresponding to the sample set based on the measured energy consumption rate of each sample in the validation set corresponding to the sample set and the predicted energy consumption rate of each sample in the validation set corresponding to the sample set; Perform 10-fold cross validation to obtain the average percentage error of the sample set.
5. The feature screening method for electric bus energy consumption prediction scenario according to claim 1 is characterized in that: The types of the multiple features to be screened in the dataset include at least one of the following: Traffic condition type, environmental factor type, vehicle status type, and driver behavior type.
6. The feature screening method for electric bus energy consumption prediction scenario according to any one of claims 1 to 5, characterized in that: The method further comprises: Using the integrated learning model, according to the target feature combination, the importance coefficient corresponding to each feature in the target feature combination is determined; Determining the probability of the target feature according to the importance coefficient corresponding to the target feature in the target feature combination and the importance coefficient corresponding to each feature; Determine a filtered feature based on the probability of each target feature, and construct a first target feature combination based on other features in the target feature combination except the filtered feature; wherein the filtered feature is a feature to be filtered out in the target feature combination; Determining a predicted energy consumption rate corresponding to the first target feature combination according to the first target feature combination; Determining an average percentage error corresponding to the first target feature combination based on the measured energy consumption rate and the predicted energy consumption rate; Detect whether the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, and if the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, repeatedly perform the steps of using the ensemble learning model to determine the importance coefficient corresponding to each feature in the target feature combination according to the target feature combination to detect whether the average percentage error corresponding to the first target feature combination is smaller than the average percentage error corresponding to the target feature combination, until the average percentage error corresponding to the first target feature combination is greater than the average percentage error corresponding to the target feature combination, and obtain the second target feature combination.
7. A feature screening device for electric bus energy consumption prediction scenario, characterized in that: The device comprises: An acquisition module, configured to acquire a data set, wherein the data set includes a plurality of features to be screened and a measured energy consumption rate; A first determination module is configured to determine a correlation coefficient of each feature to be screened, and to sort the correlation coefficients of each feature to be screened according to a preset order to obtain a feature combination; wherein determining the correlation coefficient of a target feature to be screened includes: determining the correlation coefficient of the target feature to be screened based on the target feature to be screened and a measured energy consumption rate, wherein the target feature to be screened is any one of the multiple features to be screened; a second determination module, configured to generate, based on preset rules and feature combinations, a plurality of sample sets corresponding to different feature combinations, and determine a predicted energy consumption rate of a target sample in each sample set; wherein determining the predicted energy consumption rate of the target sample in the target sample set comprises: determining the predicted energy consumption rate of the target sample in the target sample set based on the target sample set and an energy consumption prediction model trained using a training set corresponding to the target sample set, wherein the target sample set is any one of the plurality of sample sets; The third determination module is used to determine the average percentage error corresponding to each sample set based on the measured energy consumption rate of each sample in the sample set and the predicted energy consumption rate of each sample in each sample set, and to use the feature combination corresponding to the sample set that produces the minimum average percentage error as the target feature combination.
8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the feature screening method in the electric bus energy consumption prediction scenario according to any one of claims 1 to 6 by executing the computer instructions.
9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the feature screening method in the electric bus energy consumption prediction scenario according to any one of claims 1 to 6.
10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the feature screening method in the electric bus energy consumption prediction scenario according to any one of claims 1 to 6.