A new energy second-hand car charging behavior portrait and residual value correlation analysis method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU SUISHENG INFORMATION TECH CO LTD
- Filing Date
- 2026-06-09
- Publication Date
- 2026-08-04
AI Technical Summary
[0002]随着新能源汽车产业的快速发展，新能源二手车市场规模持续扩大，电池残值评估已成为影响二手车交易定价、金融贷款审批及电池回收利用的核心环节，然而现有技术在残值评估精度方面仍存在显著不足
[0017]1、从海量充电行为数据中自动筛选出对残值具有显著影响的核心因子，并利用权重知识图谱记录不同新能源二手车类型对应因子的差异化权重，实现了因车施权的精细化建模；其次构建的残值评估模型通过因子权重调制通道将知识图谱中的结构化先验知识深度嵌入特征提取过程，使模型在预测时既能利用个体充电行为画像的数据特征，又能借助类型化权重知识进行引导，实现了数据驱动与知识驱动的深度耦合；最终综合充电行为画像与权重知识图谱输出残值，显著提升了新能源二手车电池残值评估的精度、泛化能力和可解释性；
Smart Images
Figure CN122508522A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and in particular to a method for analyzing the correlation between charging behavior profiles and residual value of used new energy vehicles. Background Technology
[0002] With the rapid development of the new energy vehicle industry, the market size of new energy used cars continues to expand. Battery residual value assessment has become a core link affecting used car transaction pricing, financial loan approval and battery recycling. However, existing technologies still have significant shortcomings in terms of residual value assessment accuracy.
[0003] Currently, mainstream battery residual value assessment methods can be divided into three categories: The first category is physical model-based methods, which estimate the remaining life by establishing battery electrochemical degradation equations. Although the mechanism of this type of method is clear, it requires accurate internal battery parameters (such as electrolyte concentration, electrode porosity, etc.), which are difficult to obtain in the used car scenario, thus limiting its practical application. The second category is data-driven machine learning methods, such as random forests, support vector machines, and XGBoost, which directly use vehicle usage data to predict residual value. Although this type of method avoids complex physical modeling, it usually treats all features as equally important, and existing methods fail to distinguish between them, resulting in significant deviations in the assessment results. The third category is deep learning-based methods, which use networks such as LSTM and CNN to extract temporal features for residual value prediction. Although this has improved the feature extraction capability, it also lacks the ability to model differences in different usage environments and does not fully explore the deep semantic information contained in charging behavior data.
[0004] Therefore, there is a need to provide a method for profiling the charging behavior of new energy used vehicles and analyzing the correlation between the residual value and the battery value, in order to improve the accuracy of the residual value of new energy used vehicles. Summary of the Invention
[0005] This invention provides a method for profiling charging behavior and correlating residual value of used new energy vehicles, including:
[0006] Multiple sets of sample data are acquired, including historical charging data and historical residual values of sample new energy used vehicles. Based on these sample data sets, multiple residual value-related factors are determined. A weighted knowledge graph is constructed based on the sample data sets and these residual value-related factors, whereby the weighted knowledge graph records the weights of multiple residual value-related factors corresponding to different types of new energy used vehicles. A residual value assessment model is constructed based on the sample data sets, the residual value-related factors, and the weighted knowledge graph. Historical charging data of the new energy used vehicle to be assessed is acquired. Based on the multiple residual value-related factors and the historical charging data of the new energy used vehicle to be assessed, a charging behavior profile of the new energy used vehicle to be assessed is constructed. Using the residual value assessment model, the battery residual value of the new energy used vehicle to be assessed is determined based on the charging behavior profile and the weighted knowledge graph.
[0007] Furthermore, based on multiple sets of sample data, several residual value-related factors are determined, including: determining multiple charging behavior factors; for each charging behavior factor, determining the value of the charging behavior factor corresponding to each sample of new energy used vehicles based on the historical charging data of multiple sample new energy used vehicles; and determining multiple residual value-related factors based on the value of each charging behavior factor corresponding to each sample of new energy used vehicles through an adaptive wolf pack algorithm.
[0008] Furthermore, using an adaptive wolf pack algorithm, multiple residual value-related factors are determined based on the value of each sample of new energy used vehicles corresponding to each charging behavior factor. These factors include: for each charging behavior factor, calculating the residual value influence coefficient of the charging behavior factor based on the value of each sample of new energy used vehicles corresponding to the charging behavior factor and the historical residual value of each sample of new energy used vehicles; for any two charging behavior factors, calculating the mutual influence coefficient between the two charging behavior factors based on the values of each sample of new energy used vehicles corresponding to the two charging behavior factors; selecting effective charging behavior factors from multiple charging behavior factors using the residual value influence coefficient of each charging behavior factor and the mutual influence coefficient between any two charging behavior factors; and determining multiple residual value-related factors based on multiple effective charging behavior factors using the adaptive wolf pack algorithm.
[0009] Furthermore, using the adaptive wolf pack algorithm, multiple residual-related factors are determined based on multiple effective charging behavior factors, including: S11, initializing multiple wolf packs based on the residual influence coefficients of multiple effective charging behavior factors and the mutual influence coefficients of any two effective charging behavior factors; S12, constructing a fitness function; S13, for each wolf pack, performing multiple iterations until the first optimization termination condition is met. In each iteration, based on the fitness value of each wolf in the pack, the alpha wolf, scout wolf, and predator wolf are determined. For each scout wolf, the adaptive roaming strategy of the scout wolf is determined; S14, based on the alpha wolf of each wolf pack, determining whether the second optimization termination condition is met. If yes, proceed to S17; otherwise, proceed to S15; S15, for any two wolf packs, calculating the similarity between the alpha wolves of the two packs; S16, based on the similarity between the alpha wolves of any two packs, swapping the alpha wolves of multiple wolf packs, and proceeding to S13; S17, based on the alpha wolf of each wolf pack, determining multiple residual-related factors.
[0010] Furthermore, the adaptive roaming strategy of the scout wolf is determined, including: calculating the similarity between the scout wolf and the alpha wolf; calculating the mean similarity between the scout wolf and each other scout wolf; determining the scout wolf's roaming range based on the similarity between the scout wolf and the alpha wolf and the mean similarity between the scout wolf and each other scout wolf; determining the scout wolf's reference scout wolf based on the similarity between the scout wolf and each other scout wolf; and determining the scout wolf's adaptive roaming strategy based on the alpha wolf, the scout wolf's reference scout wolf, and the roaming range.
[0011] Furthermore, based on multiple sets of sample data and multiple residual value-related factors, a weighted knowledge graph is constructed, including: determining multiple environmental impact factors based on multiple sets of sample data and multiple residual value-related factors; determining multiple types of new energy used vehicles based on multiple environmental impact factors, and determining the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle based on multiple sets of sample data; and constructing a weighted knowledge graph based on the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle.
[0012] Furthermore, based on multiple sets of sample data and multiple residual value-related factors, multiple environmental impact factors are determined, including: determining multiple environmental factors, wherein the environmental factors include at least road condition factors; constructing a charging behavior profile for each sample of new energy used vehicles based on the multiple residual value-related factors; calculating the similarity between the charging behavior profiles of any two sample new energy used vehicles; grouping the multiple sample new energy used vehicles into multiple first sample groups based on the similarity between the charging behavior profiles of any two sample new energy used vehicles; for each environmental factor, determining the value of the environmental factor corresponding to each sample new energy used vehicle based on the historical charging data of the multiple sample new energy used vehicles; for each first sample group and each environmental factor, calculating the residual value impact coefficient of the environmental factor on the first sample group based on the value of each sample new energy used vehicle included in the first sample group corresponding to the environmental factor and the historical residual value of each sample new energy used vehicle; and selecting multiple environmental impact factors from the multiple environmental factors based on the residual value impact coefficient of each environmental factor on each first sample group.
[0013] Furthermore, based on multiple environmental impact factors, multiple types of new energy used vehicles are identified, and the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle are determined according to multiple sets of sample data. This includes: constructing a charging environment profile for each sample of new energy used vehicles based on multiple environmental impact factors; calculating the similarity between the charging environment profiles of any two sample new energy used vehicles; grouping multiple sample new energy used vehicles into multiple second sample groups based on the similarity between the charging behavior profiles of any two sample new energy used vehicles, where each second sample group corresponds to one type of new energy used vehicle; for each second sample group and each residual value-related factor, calculating the residual value impact coefficient of the residual value-related factor on the second sample group based on the value of each sample new energy used vehicle included in the second sample group and the historical residual value of each sample new energy used vehicle; and for each second sample group, determining the weights of multiple residual value-related factors for the new energy used vehicle type corresponding to the second sample group based on the residual value impact coefficient of each residual value-related factor on the second sample group.
[0014] Furthermore, through a residual value assessment model, the battery residual value of the new energy used vehicle is determined based on its charging behavior profile and weighted knowledge graph. This includes: constructing a charging environment profile of the new energy used vehicle based on its historical charging data; determining the type of new energy used vehicle based on its charging environment profile; determining the weights of multiple residual value-related factors corresponding to the new energy used vehicle based on its type and weighted knowledge graph; and determining the battery residual value of the new energy used vehicle based on its charging behavior profile and the corresponding weights of multiple residual value-related factors through the residual value assessment model.
[0015] Furthermore, the residual value assessment model includes a feature embedding layer, a factor weight modulation channel, a convolutional feature extraction branch, a global feature fusion layer, and a residual value output layer. The feature embedding layer encodes the charging behavior profile of the new energy used vehicle to be assessed and the weights of multiple corresponding residual value-related factors. The convolutional feature extraction branch extracts features from the encoded charging behavior profile of the new energy used vehicle to be assessed. The global feature fusion layer fuses the weights of the encoded multiple residual value-related factors and the features of the charging behavior profile of the new energy used vehicle to be assessed. The residual value output layer outputs the battery residual value of the new energy used vehicle to be assessed.
[0016] Compared to existing technologies, the method for profiling charging behavior and analyzing residual value of new energy used vehicles provided in this specification has at least the following beneficial effects:
[0017] 1. The core factors that significantly influence residual value are automatically selected from massive charging behavior data, and the differentiated weights of corresponding factors for different types of new energy used vehicles are recorded using a weighted knowledge graph, achieving refined modeling based on vehicle-specific weights. Secondly, the constructed residual value assessment model deeply embeds structured prior knowledge from the knowledge graph into the feature extraction process through a factor weight modulation channel. This allows the model to utilize both the data features of individual charging behavior profiles and the guidance of typified weight knowledge during prediction, achieving deep coupling between data-driven and knowledge-driven approaches. Finally, the residual value is output by combining the charging behavior profile and the weighted knowledge graph, significantly improving the accuracy, generalization ability, and interpretability of new energy used vehicle battery residual value assessment.
[0018] 2. By calculating the residual influence coefficient of each charging behavior factor and the mutual influence coefficient between any two factors, the independent contribution and interaction effect of factors can be captured simultaneously, achieving accurate screening of effective factors. In the adaptive wolf pack algorithm, the wandering range of the scout wolf is driven by the similarity of the alpha wolf and the average similarity of the wolf pack. The larger the wandering range, the more factors are replaced, realizing adaptive adjustment of the search step size. At the same time, through the alpha wolf exchange mechanism, the similarity of alpha wolves among multiple wolf packs is used to promote information sharing and avoid premature convergence. The individual with the smallest similarity is selected by referring to the scout wolf to ensure that the maximum differentiated information is introduced to broaden the search space.
[0019] 3. By screening environmental impact factors and clustering charging behavior and environmental profiles, new energy used vehicles are finely divided into multiple types. For each type, the influence coefficients of residual value-related factors are independently calculated to determine differentiated weights. The constructed weight knowledge graph realizes structured prior storage for "weighting based on vehicle". During residual value assessment, the vehicle type is automatically matched based on the charging environment profile and the corresponding weight is retrieved from the knowledge graph. Then, the profile and weight are jointly encoded through the feature embedding layer. The factor weight modulation channel adaptively modulates the profile features with the retrieved weights. The convolutional branch extracts local behavior patterns. The global fusion layer deeply couples the modulated weight prior with the data features extracted by the convolution. Finally, the residual value is output, realizing the end-to-end fusion of knowledge-driven and data-driven approaches. This significantly improves the accuracy and interpretability of residual value assessment for new energy used vehicles under different usage scenarios. Attached Figure Description
[0020] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:
[0021] Figure 1 This is a flowchart illustrating a method for analyzing the correlation between charging behavior profiles and residual value of used new energy vehicles, as shown in one embodiment of this application.
[0022] Figure 2 This is a flowchart illustrating the determination of multiple residual value-related factors in one embodiment of this application;
[0023] Figure 3 This is a flowchart illustrating the determination of multiple environmental impact factors in one embodiment of this application;
[0024] Figure 4 This is a schematic diagram of a weighted knowledge graph shown in one embodiment of this application. Detailed Implementation
[0025] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below.
[0026] Figure 1 This is a flowchart illustrating a method for analyzing the correlation between charging behavior profiles and residual value of used new energy vehicles, as shown in one embodiment of this application. Figure 1 As shown, a method for profiling charging behavior and analyzing residual value of new energy used vehicles may include the following steps.
[0027] S110, acquire multiple sets of sample data.
[0028] The sample data includes historical charging data and historical residual value of the sample new energy used vehicles. Historical charging data can include at least two dimensions: vehicle operating environment data and basic charging data. Vehicle operating environment data refers to the objective record of external conditions during charging and vehicle driving. Examples include: ambient temperature data, i.e., the ambient temperature at each charging time and during driving. This data directly determines the battery's thermal operating state. High temperature environments accelerate electrolyte decomposition and SEI film thickening, leading to rapid capacity decay, while low temperature environments increase internal resistance and trigger lithium plating risk, making it the primary environmental factor affecting battery life; ambient humidity data, i.e., the relative humidity of the air during charging and driving. High humidity environments can easily lead to battery pack sealing failure and internal metal component corrosion, thereby increasing internal resistance and aggravating self-discharge; and road condition data, i.e., the type of road during vehicle driving (urban congestion / highway / mountain road / bumpy road, etc.). Frequent bumps and sudden acceleration and deceleration can cause vibration of the internal structure of the battery pack, aggravating electrode material fatigue and loosening of solder joints, thereby accelerating battery physical degradation.
[0029] Basic charging data refers to core quantitative indicators that reflect user charging habits and the actual charging and discharging stress the battery withstands. Examples include: charging frequency data, which is the number of charges per unit time. High-frequency charging means a high number of cycles, directly consuming battery cycle life; charging duration data, which is the duration of a single charge from start to finish. Too short a charging time usually corresponds to high-power fast charging, which poses a great risk of battery thermal shock and lithium plating. Too long a charging time reflects an increase in battery internal resistance, leading to a decrease in charging efficiency; and charging amount data, which is the actual amount of electricity charged in each charge (kWh). Combined with the total capacity of the vehicle battery, the SOC change of a single charge can be calculated. Frequent high-capacity fast charging will significantly accelerate battery aging.
[0030] Historical residual value is a normalized residual value coefficient (0-1) calculated based on battery capacity decay. It serves as a supervision label for model training, driving the model to learn the mapping relationship from multi-dimensional charging data to battery residual value.
[0031] S120, based on multiple sets of sample data, determines multiple residual-related factors.
[0032] Specifically, it includes:
[0033] Identify multiple charging behavior factors;
[0034] For each charging behavior factor, the value of the charging behavior factor corresponding to each sample of new energy used vehicles is determined based on the historical charging data of multiple sample new energy used vehicles.
[0035] Using an adaptive wolf pack algorithm, multiple residual value-related factors are determined based on the value of each sample of new energy used cars corresponding to each charging behavior factor.
[0036] Specifically, charging behavior factors are core indicators that can be quantified and extracted directly from charging data, representing user charging habits and the actual stress the battery experiences. Examples include: charging frequency, which is the number of times a battery is charged per unit time; high-frequency charging means a high number of cycles, directly consuming the battery's cycle life; the number of charging events, which is the cumulative total number of charging events in historical charging data, and is a basic indicator for measuring the degree of battery cycle aging; the average battery temperature during charging, which is the average temperature of the battery pack during each charge; this data directly reflects the thermal load state of the battery during charging; long-term high-temperature charging will accelerate SEI film growth and cathode material dissolution; temperature fluctuation, which is the standard deviation or range of battery temperature during a single charge; drastic temperature fluctuations indicate that the thermal management system is working frequently or the charging power is unstable, generating alternating thermal stress on the internal structure of the battery, leading to fatigue damage; and the average charging speed, which is the ratio of the amount of charge in a single charge to the charging time, reflecting the user's urgency to charge; the faster the charging speed, the larger the charging current usually corresponds to, and the greater the risk of lithium plating and thermal shock to the battery. In addition, factors such as the proportion of fast charging, SOC fluctuation amplitude, and resting time may also be included.
[0037] For each charging behavior factor, based on the historical charging data of multiple sample new energy used vehicles, the specific value of the charging behavior factor on each sample new energy used vehicle is calculated one by one. For example, for sample vehicle A, its charging frequency is 5.2 times per week, the average battery temperature during charging is 38.5℃, the standard deviation of temperature fluctuation is 4.2℃, and the average charging speed is 45kW / h, thus forming a multi-dimensional charging behavior factor vector corresponding to each sample new energy used vehicle.
[0038] As an example, the process of building a charging behavior profile is as follows:
[0039] First, for each charging behavior factor (including charging frequency, number of charges, average battery temperature during charging, temperature fluctuation, average charging speed, etc.), the factor value for each sample is calculated based on historical charging data of multiple sample new energy used vehicles. Specifically, charging frequency is defined as the number of charges per unit time (times / month), the number of charges is the total number of historical cumulative charges, the average battery temperature during charging is the arithmetic mean of the battery temperature during each charge, the temperature fluctuation is the standard deviation of the battery temperature during each charge, and the average charging speed is the average power (kW) of each charge. Then, the original values of the above five factors are set according to a preset fixed... The dimensions are arranged in a fixed order to form a five-dimensional vector, meaning the charging behavior profile is a multi-dimensional vector rather than a scalar, with the dimension order fixed as [charging frequency, number of charges, average temperature, temperature fluctuation, average charging speed]. Next, the five-dimensional vector is normalized using a min-max normalization method. After normalization, the value range of each dimension is mapped to the [0,1] interval to eliminate the influence of differences in the scale and numerical range of different factors on subsequent similarity calculations and model training. For missing dimensions in some samples, the median of the corresponding dimension in the training set is used to fill in the missing dimensions, ensuring that the charging behavior profile of each sample is a complete five-dimensional normalized vector.
[0040] In some embodiments, an adaptive wolf pack algorithm is used to determine multiple residual value-related factors based on the value of each charging behavior factor for each sample of new energy used vehicles, including:
[0041] For each charging behavior factor, the residual value influence coefficient of the charging behavior factor is calculated based on the value of the charging behavior factor for each sample of new energy used vehicles and the historical residual value of each sample of new energy used vehicles.
[0042] For any two charging behavior factors, calculate the mutual influence coefficient of the two charging behavior factors based on the value of each sample of new energy used vehicles corresponding to the two charging behavior factors.
[0043] Effective charging behavior factors are selected from multiple charging behavior factors by using the residual influence coefficient of each charging behavior factor and the mutual influence coefficient of any two charging behavior factors.
[0044] The adaptive wolf pack algorithm is used to determine multiple residual value-related factors based on several effective charging behavior factors.
[0045] Specifically, the value of the charging behavior factor for each sample of new energy used cars and the historical residual value of each sample of new energy used cars are substituted into the calculation formula of the correlation coefficient (e.g., Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) to calculate the correlation coefficient between the charging behavior factor and the residual value, and the absolute value is taken as the residual value influence coefficient of the charging behavior factor.
[0046] The values of the two charging behavior factors for each sample of new energy used cars are substituted into the formula for calculating the correlation coefficient (e.g., Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) to obtain the correlation coefficient of the two charging behavior factors. The absolute value of the correlation coefficient is taken as the mutual influence coefficient of the two charging behavior factors.
[0047] Charging behavior factors with residual influence coefficients greater than a residual influence coefficient threshold (e.g., 0.5) are designated as central charging behavior factors, and non-central charging behavior factors with mutual influence coefficients greater than a mutual influence coefficient threshold (e.g., 0.5) with at least one central charging behavior factor are designated as associated charging behavior factors. Valid charging behavior factors may include charging behavior factors of each central factor and each associated charging behavior factor.
[0048] Figure 2 This is a flowchart illustrating the determination of multiple residual value-related factors in one embodiment of this application, such as... Figure 2 As shown, in some embodiments, an adaptive wolf pack algorithm is used to determine multiple residual value-related factors based on multiple effective charging behavior factors, including:
[0049] S11. Initialize multiple wolf packs based on the residual influence coefficients of multiple effective charging behavior factors and the mutual influence coefficients of any two effective charging behavior factors.
[0050] S12. Construct the fitness function;
[0051] S13. For each wolf pack, perform multiple iterations until the first optimization termination condition is met. In each iteration, determine the alpha wolf, scout wolf, and predator wolf based on the fitness value of each wolf in the pack. For each scout wolf, determine the scout wolf's adaptive roaming strategy.
[0052] S14. Based on the alpha wolf of each wolf pack, determine whether the second optimization termination condition is met. If yes, proceed to S17; otherwise, proceed to S15.
[0053] S15. For any two wolf packs, calculate the similarity between the alpha wolves of the two packs.
[0054] S16. Based on the similarity of the alpha wolves of any two wolf packs, swap the alpha wolves of multiple wolf packs and execute S13.
[0055] S17. Determine multiple residual-related factors based on the alpha wolf of each wolf pack.
[0056] Specifically, initializing multiple wolf packs may include the following steps:
[0057] For each effective charging behavior factor, the mean of the mutual influence coefficients between the effective charging behavior factor and each other effective charging behavior factor is calculated as the mean of the mutual influence coefficients. The mean of the mutual influence coefficients and the residual value influence coefficients of the effective charging behavior factors are weighted and summed to obtain the sampling probability of the effective charging behavior factor. The larger the mean of the mutual influence coefficients and the residual value influence coefficients of the charging behavior factors, the greater the sampling probability of the effective charging behavior factor.
[0058] When initializing each wolf in each wolf pack, the number of valid charging behavior factors included in the wolf is randomly sampled from a range of factor numbers (e.g., 3-10). Based on the sampling probability of each valid charging behavior factor, multiple valid charging behavior factors are sampled from the multiple valid charging behavior factors, with the number of valid charging behavior factors included in the wolf being equal to the number of valid charging behavior factors included in the wolf. The larger the sampling probability, the greater the probability that a valid charging behavior factor is selected, thereby generating wolves.
[0059] Assuming there are 10 valid charging behavior factors, which are sequentially encoded as F1, F2, ..., F10, when initializing a wolf, the number of factors contained in the wolf is first randomly sampled from the factor quantity range [3,10]. For example, if 5 factors are sampled, the wolf must contain 5 valid charging behavior factors. Then, non-uniform sampling is performed according to the sampling probability of each factor. Assuming that F2, F5, F7, F1, and F9 are selected after sampling, the generated wolf is a set of encoded sequences [2,5,7,1,9]. This sequence represents the subset of factors currently carried by the wolf. If another wolf randomly selects 4 factors, and the sampling result is F3, F6, F8, and F10, then the encoded sequence of the wolf is [3,6,8,10]. Thus, each wolf is represented by encoded sequences of different lengths and combinations, forming differentiated initial search individuals.
[0060] By weighted summing the residual influence coefficient and the mean of the mutual influence coefficient of each effective charging behavior factor to obtain the sampling probability, the joint driving force of factor importance and factor coupling is realized. This ensures that factors that contribute significantly to the residual and have a high degree of coupling with other factors (i.e., contain more collaborative information) have a higher probability of being selected, thus prioritizing the retention of high-value factor combinations in the initialization phase. At the same time, each wolf randomly determines the number of factors it contains from the range of factor numbers, and then performs biased sampling based on non-uniform sampling probability. This ensures that the factor subsets carried by each wolf in the initial population have differentiated scale and combination, and guides the overall population towards high residual relevance through the sampling probability mechanism, avoiding the key factor omission or factor redundancy problems caused by traditional uniform random initialization. In addition, this data-driven probabilistic initialization method gives the initial wolf pack a naturally high search quality, enabling it to converge to the global optimum more quickly and improving efficiency.
[0061] The fitness function is related to the residual influence coefficients of the multiple effective charging behavior factors included in the wolf and the mutual influence coefficients of any two effective charging behavior factors. For example, the mean of the residual influence coefficients of the multiple effective charging behavior factors included in the wolf is obtained, and the mean of the mutual influence coefficients of any two effective charging behavior factors included in the wolf is obtained. The fitness function is a weighted summation function of the mean residual influence coefficients and the mean mutual influence coefficients of the wolf. The larger the mean residual influence coefficients and the mean mutual influence coefficients of the wolf, the larger the fitness value.
[0062] The first optimization termination condition refers to the criterion for determining whether the wolf pack optimization iteration process should terminate. Specifically, it can include a combination of one or more of the following: First, termination occurs when the number of iterations reaches a preset maximum number of iterations, for example, setting the maximum number of iterations to 100 to prevent the algorithm from running indefinitely; Second, termination occurs when the improvement in the optimal fitness value of the wolf pack is less than a preset threshold (e.g., 0.001) in a number of consecutive iterations (e.g., 20 consecutive iterations), indicating that the algorithm has entered a convergence plateau and further iterations will not yield meaningful optimization improvements. Meeting any of the above conditions triggers the first optimization termination condition, stopping the current wolf pack iteration.
[0063] In some embodiments, determining the adaptive roaming strategy of the wolf finder includes:
[0064] Calculate the similarity between the scout wolf and the alpha wolf;
[0065] Calculate the mean similarity between the scout wolf and each of the other scout wolves;
[0066] The range of movement of the scout wolf is determined by the average similarity between the scout wolf and the alpha wolf, as well as the average similarity between the scout wolf and other scout wolves.
[0067] Based on the similarity between the scout wolf and each other scout wolf, determine the reference scout wolf;
[0068] Based on the alpha wolf, the scout wolf's reference scout wolf, and the roaming range, determine the scout wolf's adaptive roaming strategy.
[0069] Specifically, the cosine similarity between the coding sequences of the scout wolf and the alpha wolf can be calculated as the similarity between the scout wolf and the alpha wolf.
[0070] Calculate the cosine similarity between the coded sequences of the wolf and each other wolf, and take the mean as the mean of the similarity.
[0071] The greater the similarity between the scout wolf and the alpha wolf, and the greater the mean similarity between the scout wolf and other scout wolves, the wider the scout wolf's movement range, allowing it to search over a larger area. The scout wolf's movement range characterizes the number of effective charging behavior factors that can be substituted within the scout wolf population.
[0072] The range of movement of the wolf can be determined using the following formula:
[0073]
[0074] in, Let i be the range of movement of the i-th scout wolf. The preset range of movement (e.g., 1, 2, etc.). Let be the similarity between the i-th scout wolf and the alpha wolf. Let be the mean similarity between the i-th scout wolf and the other scout wolves. To Round up.
[0075] This indicates the consistency of the exploratory wolf's and alpha wolf's directions. The larger the value, the more consistent the exploratory wolf's and alpha wolf's search directions are. Larger following walks should be encouraged to accelerate convergence. The similarity value represents the average similarity between the scout wolf and other scout wolves. A higher value indicates that the scout wolf is in a densely populated area of the pack, with high information redundancy within the group, requiring it to move more extensively to avoid getting trapped in local optima. First, when a scout wolf is highly similar to the alpha wolf and other scout wolves, it moves extensively, accelerating its convergence towards the global optimum. Second, when a scout wolf is similar to the alpha wolf but significantly different from other scout wolves, it moves moderately, balancing following and independent exploration. Third, the rounding up operation ensures that the scout wolf replaces at least one factor in each iteration, avoiding a dead loop due to zero movement, ensuring minimum exploration activity, and thus achieving a dynamic balance between global exploration and local development.
[0076] Other scout wolves with the lowest similarity can be used as reference scout wolves. Based on the scout wolves' movement range, some effective charging behavior factors among the scout wolves can be replaced with effective charging behavior factors from the alpha wolf and the reference scout wolves. The adaptive movement strategy of the scout wolves is precisely this scheme of replacing some effective charging behavior factors among the scout wolves with effective charging behavior factors from the alpha wolf and the reference scout wolves.
[0077] For example, if a scout wolf has a walk range of 2, its adaptive walk strategy is as follows: randomly select 2 factors (i.e., walk range of 2) from its 5 factors for replacement. The replacement sources are the factor pools of the alpha wolf and the reference scout wolf. For example, randomly select factors 1 and 7 from the scout wolf for replacement, select factor 2 from the alpha wolf's factor pool {2,5,7,1,9}, and select factor 10 from the reference scout wolf's factor pool {4,6,8,10,2} (ensuring it does not overlap with factors 3, 5, and 9 retained in the scout wolf). The final generated new scout wolf encoding sequence is [2,3,5,10,9]. The core idea of this strategy is: the larger the walk range, the more factors are replaced. The scout wolf moves towards both the alpha wolf and the reference scout wolf with the greatest difference, utilizing both the global guidance information of the alpha wolf and the complementary information of the reference scout wolf with the greatest difference to broaden the search space, thereby achieving an adaptive balance between exploration and development.
[0078] Calculate the similarity between the alpha wolves of any two wolf packs, and take the average of these similarities as the population similarity mean. The second optimization termination condition can be that the population similarity mean is greater than the population similarity mean threshold (e.g., 0.7).
[0079] For each population, calculate the similarity between the alpha wolf of that population and the alpha wolf of every other population, and replace the alpha wolf of that population with the alpha wolf of the other population with the lowest similarity.
[0080] Based on the fitness value of the alpha wolf in each wolf pack, the alpha wolf with the highest fitness value is determined, and the effective charging behavior factors included in the alpha wolf with the highest fitness value are used as multiple residual correlation factors.
[0081] By calculating the residual value influence coefficient of each charging behavior factor, the independent contribution of each factor to the residual value of used cars can be quantified, and irrelevant factors can be quickly eliminated. By calculating the mutual influence coefficient of any two charging behavior factors, the interaction effect between factors can be captured, avoiding feature redundancy or omission caused by neglecting the coupling relationship between factors in traditional methods. Combined with the adaptive wolf pack algorithm, the swarm intelligence search capability of wolves is used to adaptively walk and filter in the effective factor pool, and finally determine the optimal combination of residual value-related factors. This scheme realizes automated and refined dimensionality reduction from the original charging behavior factors to residual value-related factors, which not only retains the core factors that are most explanatory for residual value prediction, but also eliminates redundancy and noise, significantly improving the accuracy and generalization ability of the subsequent residual value prediction model.
[0082] S130 constructs a weighted knowledge graph based on multiple sets of sample data and multiple residual-related factors.
[0083] Among them, the weighted knowledge graph is used to record the weights of multiple residual value-related factors corresponding to different types of new energy used cars.
[0084] Specifically, it includes:
[0085] Based on multiple sets of sample data and multiple residual value-related factors, multiple environmental impact factors were determined;
[0086] Based on multiple environmental impact factors, multiple types of new energy used vehicles were identified, and the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle were determined based on multiple sets of sample data.
[0087] A weighted knowledge graph is constructed based on the weights of multiple residual value-related factors corresponding to each type of new energy used car.
[0088] Figure 3 This is a flowchart illustrating the determination of multiple environmental impact factors in one embodiment of this application, such as... Figure 3 As shown, in some embodiments, multiple environmental impact factors are determined based on multiple sets of sample data and multiple residual value-related factors, including:
[0089] Multiple environmental factors are identified, including at least road condition factors, and may also include average annual temperature, annual precipitation, average annual air pressure, etc.
[0090] Based on multiple residual value-related factors, a charging behavior profile for each sample of new energy used vehicles is constructed, wherein the charging behavior profile of the sample of new energy used vehicles includes the value of each residual value-related factor;
[0091] Calculate the similarity of charging behavior profiles for any two sample new energy used vehicles;
[0092] Based on the similarity of the charging behavior profiles of any two sample new energy used vehicles, multiple sample new energy used vehicles are grouped to determine multiple first sample groups. For example, by using a clustering algorithm, multiple sample new energy used vehicles are clustered based on the similarity of the charging behavior profiles of any two sample new energy used vehicles to determine multiple first sample groups.
[0093] For each environmental factor, based on the historical charging data of multiple sample new energy used vehicles, the value of the environmental factor corresponding to each sample new energy used vehicle is determined. As an example, the quantification method of multiple environmental factors is as follows: For road condition factors, road surface roughness is used as a continuous value for quantification. Specifically, the road surface roughness index of the corresponding road segment is extracted from the driving trajectory associated with the historical charging data of the sample new energy used vehicles. The quantification method of road surface roughness is as follows: Based on the triaxial acceleration data collected by the vehicle acceleration sensor during the last segment of driving before charging, the standard deviation of acceleration is calculated as a measure of the road surface bumpiness, and then it is linearly mapped to the [0,1] interval. For the three continuous environmental factors of annual average temperature, annual precipitation, and annual average air pressure, the annual average temperature value (unit ℃), annual precipitation value (unit mm), and annual average air pressure value (unit hPa) corresponding to the region where the sample new energy used vehicle is located are directly taken as the value of the environmental factor corresponding to the sample new energy used vehicle.
[0094] For each first sample group and each environmental factor, the environmental factor's influence coefficient on the residual value of the first sample group is calculated based on the value of each sample of new energy used cars included in the first sample group corresponding to the environmental factor and the historical residual value of each sample of new energy used cars. For example, the correlation coefficient between the environmental factor and the historical residual value is calculated using the correlation coefficient calculation formula, and its absolute value is taken as the environmental factor's influence coefficient on the residual value of the first sample group.
[0095] Based on the residual impact coefficient of each environmental factor on each first sample group, multiple environmental impact factors are screened from multiple environmental factors.
[0096] Specifically, for each environmental factor, the mean of the residual value impact coefficient of the environmental factor on each first sample group is calculated, and this mean is used as the mean of the residual value impact coefficient of the corresponding environmental factor. Environmental factors whose mean residual value impact coefficient is greater than the threshold of the mean residual value impact coefficient (e.g., 0.5) are considered as environmental impact factors.
[0097] In some embodiments, multiple types of new energy used vehicles are determined based on multiple environmental impact factors, and the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle are determined based on multiple sets of sample data, including:
[0098] Based on multiple environmental impact factors, a charging environment profile is constructed for each sample of new energy used vehicles. The charging environment profile may include the value of each environmental impact factor.
[0099] Calculate the similarity of the charging environment profiles of any two sample new energy used vehicles;
[0100] Based on the similarity of the charging behavior profiles of any two sample new energy used vehicles, multiple sample new energy used vehicles are grouped to determine multiple second sample groups. Each second sample group corresponds to a type of new energy used vehicle. For example, by using a clustering algorithm to cluster multiple sample new energy used vehicles based on the similarity of the charging behavior profiles of any two sample new energy used vehicles, multiple second sample groups can be determined.
[0101] For each second sample group and each residual value-related factor, the residual value-related factor's residual value impact coefficient on the second sample group is calculated based on the residual value-related factor's value for each sample of new energy used cars included in the second sample group and the historical residual value of each sample of new energy used cars included in the second sample group. Specifically, using the correlation coefficient calculation formula, the correlation coefficient between the residual value-related factor and the historical residual value of the second sample group is calculated based on the residual value-related factor's value for each sample of new energy used cars included in the second sample group and the historical residual value of each sample of new energy used cars included in the second sample group. The absolute value of this coefficient is taken as the residual value impact coefficient of the residual value-related factor on the second sample group.
[0102] For each second sample group, the weights of multiple residual value-related factors for the new energy used car type corresponding to the second sample group are determined based on the residual value influence coefficient of each residual value-related factor on the residual value of the second sample group. Specifically, the ratio of the residual value influence coefficient of the residual value-related factor on the second sample group to the sum of the residual value influence coefficients of all residual value-related factors on the second sample group is used as the weight of the residual value-related factor for the new energy used car type corresponding to the second sample group.
[0103] Figure 4 This is a schematic diagram of a weighted knowledge graph shown in one embodiment of this application, such as... Figure 4 As shown, the weighted knowledge graph uses the type of new energy used car as the core node, residual value-related factors as associated nodes, and the weight is the attribute value of the edge. For example, the node "Urban High-Frequency Fast Charging Type" has a weight of 0.55 for the connected factor "Battery Health"; a weight of 0.30 for the connected factor "Fast Charging Percentage"; and a weight of 0.15 for the connected factor "Annual Charging Times". The node "Suburban Low-Frequency Slow Charging Type" has a weight of 0.45 for the connected factor "Battery Health"; a weight of 0.35 for the connected factor "Fast Charging Percentage"; and a weight of 0.20 for the connected factor "Annual Charging Times".
[0104] S140 constructs a residual value assessment model based on multiple sets of sample data, multiple residual value-related factors, and a weighted knowledge graph.
[0105] S150: Obtain historical charging data for the new energy used vehicle to be evaluated.
[0106] S160 constructs a charging behavior profile of the new energy used vehicles to be evaluated based on multiple residual value-related factors and historical charging data of the new energy used vehicles to be evaluated.
[0107] S160 uses a residual value assessment model to determine the battery residual value of the new energy used vehicle to be assessed based on the charging behavior profile and weighted knowledge graph of the new energy used vehicle to be assessed.
[0108] Specifically, it includes:
[0109] Based on the historical charging data of the new energy used vehicles to be evaluated, a charging environment profile of the new energy used vehicles to be evaluated is constructed.
[0110] Based on the charging environment profile of the new energy used vehicles to be evaluated, determine the type of new energy used vehicle to be evaluated.
[0111] Based on the knowledge graph of new energy used vehicles and their weights, the weights of multiple residual value-related factors corresponding to the new energy used vehicles to be evaluated are determined.
[0112] The residual value of a new energy used vehicle's battery is determined by using a residual value assessment model, based on the charging behavior profile of the new energy used vehicle to be assessed and the weights of several corresponding residual value-related factors.
[0113] Specifically, the similarity between the charging environment profile of the new energy used vehicle to be evaluated and the charging environment profile of the sample new energy used vehicle corresponding to the cluster center of each second sample group is calculated. The new energy used vehicle type corresponding to the second sample group with the highest similarity is taken as the new energy used vehicle type of the new energy used vehicle to be evaluated.
[0114] In some embodiments, the residual value assessment model includes a feature embedding layer, a factor weight modulation channel, a convolutional feature extraction branch, a global feature fusion layer, and a residual value output layer. The feature embedding layer is used to encode the charging behavior profile of the new energy used vehicle to be assessed and the weights of the corresponding multiple residual value-related factors. The convolutional feature extraction branch is used to extract the features of the encoded charging behavior profile of the new energy used vehicle to be assessed. The global feature fusion layer is used to fuse the weights of the encoded multiple residual value-related factors and the features of the charging behavior profile of the new energy used vehicle to be assessed. The residual value output layer is used to output the battery residual value of the new energy used vehicle to be assessed.
[0115] Specifically, the feature embedding layer receives the charging behavior profile of the new energy used vehicle to be evaluated and the residual value-related factor weights corresponding to the vehicle type retrieved from the weight knowledge graph. Through embedding mapping, the discrete encoding is converted into a dense vector, and the weight values are encoded as weight vectors with the same dimension as the profile vector. This allows the model to simultaneously perceive "what kind of vehicle it is" and "how important each factor is." The factor weight modulation channel receives the embedded charging behavior profile vector and weight vector. Through element-wise multiplication or attention weighting, the weight vector modulates each dimension of the profile vector, thereby achieving differentiated emphasis and suppression of the contribution of different factors at the feature level, enabling the model to... The model focuses on the factors that have the greatest impact on the residual value of this type of vehicle. The convolutional feature extraction branch performs a one-dimensional convolution operation on the charging behavior profile vector after weight modulation. It extracts local feature patterns by sliding through multiple convolution kernels of different sizes, and then compresses them through a pooling layer to obtain a high-level abstract feature vector of the charging behavior profile. The global feature fusion layer concatenates or weights the modulated weight vector with the profile feature vector extracted by convolution, so that the model can simultaneously utilize the typified priors provided by the weight knowledge graph and the individualized information provided by the actual profile. Finally, the residual value output layer maps the fused features to a scalar value through a fully connected layer, which is the battery residual value prediction result of the new energy used car.
[0116] The training process for the residual value assessment model is as follows:
[0117] First, a training set is constructed based on a large number of historical samples of new energy used vehicles with completed battery residual value testing. Each sample includes a charging behavior profile, the type of new energy used vehicle to which the sample belongs, the weight vector of multiple residual value-related factors corresponding to the type retrieved from the weight knowledge graph, and the actual battery residual value (scalar, as a supervision signal) obtained by the actual testing of the sample.
[0118] During training, the feature embedding layer receives discrete feature indices from the charging behavior profile and discrete type indices from each residual-related factor. It maps these to dense vectors using a learnable embedding matrix. Simultaneously, the factor weights provided by the weight knowledge graph are encoded as weight vectors with the same dimension as the profile vector. The embedding layer parameters are jointly optimized with the full model during training. The factor weight modulation channel performs element-wise multiplication on the embedded profile vector and the weight vector to obtain the modulated feature vector. This step amplifies the profile dimension corresponding to high-weight factors and suppresses the low-weight dimension. The convolutional feature extraction branch applies multiple... One-dimensional convolutional kernels of different sizes (e.g., kernel_size=3, 5, 7) are used. Each convolutional kernel is followed by ReLU activation and max pooling. The pooling results are concatenated and compressed into a fixed-length high-level abstract feature vector through a fully connected layer. The global feature fusion layer concatenates the modulated weight vector with the portrait feature vector output by the convolutional kernel along the feature dimension, and then performs non-linear fusion through a fully connected layer to output a fused feature vector of a unified dimension. The residual output layer takes this fused vector as input, passes through two fully connected layers (with Dropout regularization in between), and finally outputs a scalar as the battery residual value prediction.
[0119] The residual evaluation model uses the mean squared error (MSE) between the predicted residual and the actual residual as the loss function. It employs the Adam optimizer for end-to-end backpropagation training. The gradient is backpropagated through the global feature fusion layer to the convolutional branch, modulation channel, and feature embedding layer. The embedding matrix of the embedding layer is updated during training to learn a better feature representation.
[0120] After training, during inference, the input is the charging behavior profile of the vehicle to be evaluated and the factor weight vector corresponding to its matching type. After the same forward process described above, the battery residual value prediction result can be output. Throughout the process, the weight knowledge graph provides prior factor weights of various types as supervision and guidance during the training phase, and provides retrieval weights as conditional input during the inference phase, enabling the model to have the ability to integrate typified prior knowledge and individual profile information.
[0121] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.
Claims
1. A method for analyzing the correlation between charging behavior profiles and residual value of used new energy vehicles, characterized in that, include: Acquire multiple sets of sample data, including historical charging data and historical residual value of sample new energy used vehicles; Based on multiple sets of sample data, several residual-related factors were determined; Based on multiple sets of sample data and multiple residual value-related factors, a weighted knowledge graph is constructed, wherein the weighted knowledge graph is used to record the weights of multiple residual value-related factors corresponding to different types of new energy used cars; A residual value assessment model is constructed based on multiple sets of sample data, multiple residual value-related factors, and a weighted knowledge graph. Obtain historical charging data for the new energy used vehicles to be evaluated; Based on multiple residual value-related factors and historical charging data of the new energy used vehicles to be evaluated, a charging behavior profile of the new energy used vehicles to be evaluated is constructed. The residual value of the battery of the new energy used vehicle to be evaluated is determined by the residual value assessment model based on the charging behavior profile and weighted knowledge graph of the new energy used vehicle to be evaluated.
2. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 1, characterized in that, Based on multiple sets of sample data, several residual-related factors were identified, including: Identify multiple charging behavior factors; For each charging behavior factor, the value of the charging behavior factor corresponding to each sample of new energy used vehicles is determined based on the historical charging data of multiple sample new energy used vehicles. Using an adaptive wolf pack algorithm, multiple residual value-related factors are determined based on the value of each sample of new energy used cars corresponding to each charging behavior factor.
3. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 2, characterized in that, Using an adaptive wolf pack algorithm, multiple residual value-related factors are determined based on the value of each charging behavior factor for each sample of new energy used vehicles, including: For each charging behavior factor, the residual value influence coefficient of the charging behavior factor is calculated based on the value of the charging behavior factor for each sample of new energy used vehicles and the historical residual value of each sample of new energy used vehicles. For any two charging behavior factors, calculate the mutual influence coefficient of the two charging behavior factors based on the value of each sample of new energy used vehicles corresponding to the two charging behavior factors. Effective charging behavior factors are selected from multiple charging behavior factors by using the residual influence coefficient of each charging behavior factor and the mutual influence coefficient of any two charging behavior factors. The adaptive wolf pack algorithm is used to determine multiple residual value-related factors based on several effective charging behavior factors.
4. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 3, characterized in that, Using an adaptive wolf pack algorithm, several residual value-related factors are determined based on multiple effective charging behavior factors, including: S11. Initialize multiple wolf packs based on the residual influence coefficients of multiple effective charging behavior factors and the mutual influence coefficients of any two effective charging behavior factors. S12. Construct the fitness function; S13. For each wolf pack, perform multiple iterations until the first optimization termination condition is met. In each iteration, determine the alpha wolf, scout wolf, and predator wolf based on the fitness value of each wolf in the pack. For each scout wolf, determine the scout wolf's adaptive roaming strategy. S14. Based on the alpha wolf of each wolf pack, determine whether the second optimization termination condition is met. If yes, proceed to S17; otherwise, proceed to S15. S15. For any two wolf packs, calculate the similarity between the alpha wolves of the two packs. S16. Based on the similarity of the alpha wolves of any two wolf packs, swap the alpha wolves of multiple wolf packs and execute S13. S17. Determine multiple residual-related factors based on the alpha wolf of each wolf pack.
5. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 4, characterized in that, Determine the adaptive roaming strategy for the wolf scout, including: Calculate the similarity between the scout wolf and the alpha wolf; Calculate the mean similarity between the scout wolf and each of the other scout wolves; The range of movement of the scout wolf is determined by the average similarity between the scout wolf and the alpha wolf, as well as the average similarity between the scout wolf and other scout wolves. Based on the similarity between the scout wolf and each other scout wolf, determine the reference scout wolf; Based on the alpha wolf, the scout wolf's reference scout wolf, and the roaming range, determine the scout wolf's adaptive roaming strategy.
6. A method for analyzing the correlation between charging behavior profiles and residual value of used new energy vehicles according to any one of claims 1-5, characterized in that, Based on multiple sets of sample data and multiple residual-related factors, a weighted knowledge graph is constructed, including: Based on multiple sets of sample data and multiple residual value-related factors, multiple environmental impact factors were determined; Based on multiple environmental impact factors, multiple types of new energy used vehicles were identified, and the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle were determined based on multiple sets of sample data. A weighted knowledge graph is constructed based on the weights of multiple residual value-related factors corresponding to each type of new energy used car.
7. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 6, characterized in that, Based on multiple sets of sample data and multiple residual-related factors, several environmental impact factors were identified, including: Multiple environmental factors are identified, wherein the environmental factors include at least road condition factors; Based on multiple residual value-related factors, a charging behavior profile of each sample of new energy used vehicles is constructed; Calculate the similarity of charging behavior profiles for any two sample new energy used vehicles; Based on the similarity of the charging behavior profiles of any two sample new energy used vehicles, multiple sample new energy used vehicles are grouped to determine multiple first sample groups; For each environmental factor, the value of the environmental factor corresponding to each sample of new energy used vehicles is determined based on the historical charging data of multiple samples of new energy used vehicles. For each first sample group and each environmental factor, the residual value of each sample of new energy used cars included in the first sample group corresponding to the environmental factor is calculated as the residual value of each sample of new energy used cars and the historical residual value of each sample of new energy used cars. Based on the residual impact coefficient of each environmental factor on each first sample group, multiple environmental impact factors are screened from multiple environmental factors.
8. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 7, characterized in that, Based on multiple environmental impact factors, several types of new energy used vehicles were identified, and based on multiple sets of sample data, the weights of multiple residual value-related factors corresponding to each type of new energy used vehicle were determined, including: Based on multiple environmental influencing factors, a charging environment profile is constructed for each sample of new energy used vehicles; Calculate the similarity of the charging environment profiles of any two sample new energy used vehicles; Based on the similarity of the charging behavior profiles of any two sample new energy used vehicles, multiple sample new energy used vehicles are grouped to determine multiple second sample groups, where each second sample group corresponds to a type of new energy used vehicle. For each second sample group and each residual value-related factor, the residual value impact coefficient of the residual value-related factor on the residual value of the second sample group is calculated based on the value of each sample of new energy used cars included in the second sample group corresponding to the residual value-related factor and the historical residual value of each sample of new energy used cars. For each second sample group, the weights of multiple residual value-related factors for the corresponding new energy used car type are determined based on the residual value influence coefficient of each residual value-related factor on the residual value of the second sample group.
9. The method for analyzing the correlation between charging behavior profiles and residual value of new energy used vehicles according to claim 8, characterized in that, Using a residual value assessment model, based on the charging behavior profile and weighted knowledge graph of the new energy used vehicles to be assessed, the residual value of the batteries of the new energy used vehicles is determined, including: Based on the historical charging data of the new energy used vehicles to be evaluated, a charging environment profile of the new energy used vehicles to be evaluated is constructed. Based on the charging environment profile of the new energy used vehicles to be evaluated, determine the type of new energy used vehicle to be evaluated. Based on the knowledge graph of new energy used vehicles and their weights, the weights of multiple residual value-related factors corresponding to the new energy used vehicles to be evaluated are determined. The residual value of a new energy used vehicle's battery is determined by using a residual value assessment model, based on the charging behavior profile of the new energy used vehicle to be assessed and the weights of several corresponding residual value-related factors.
10. The method for profiling charging behavior and analyzing residual value of new energy used vehicles according to claim 9, characterized in that, The residual value assessment model includes a feature embedding layer, a factor weight modulation channel, a convolutional feature extraction branch, a global feature fusion layer, and a residual value output layer. The feature embedding layer is used to encode the charging behavior profile of the new energy used vehicle to be assessed and the weights of the corresponding multiple residual value-related factors. The convolutional feature extraction branch is used to extract the features of the encoded charging behavior profile of the new energy used vehicle to be assessed. The global feature fusion layer is used to fuse the weights of the encoded multiple residual value-related factors and the features of the charging behavior profile of the new energy used vehicle to be assessed. The residual value output layer is used to output the battery residual value of the new energy used vehicle to be assessed.