A method for predicting short-term charging routes for electric vehicles based on environmental factors

Abstract

This invention belongs to the field of charging load technology, specifically relating to a method for predicting short-term charging routes for electric vehicles based on environmental factors. The method includes: constructing a traffic network model; collecting traffic flow data and denoising the data using the exponential moving average method; calculating the correlation of the denoised AC traffic flow data using the Spearman correlation coefficient formula; normalizing the traffic flow data based on the correlation and inputting it into the prediction model to obtain the traffic flow for the next time period; constructing a flow-density-speed relationship based on the predicted traffic flow and building a road impedance model; obtaining the shortest travel time path for the electric vehicle based on the road impedance model and calculating the mileage power consumption of the electric vehicle; processing the mileage power consumption using the Monte Carlo method to obtain the predicted results of the spatiotemporal distribution of the electric vehicle charging load; and determining the charging route for the electric vehicle based on the prediction results. This invention improves the convenience of user travel by predicting the charging load of electric vehicles with characteristics of mountainous urban areas.

Description

Technical Field

[0001] This invention belongs to the field of electric vehicle charging load technology, specifically relating to a method for predicting short-term charging routes for electric vehicles based on environmental factors. Background Technology

[0002] Under the dual pressures of energy conservation and environmental protection, electric vehicles (EVs) are developing rapidly worldwide, offering significant clean and environmental benefits. EVs are gradually becoming widely used, and their application greatly aligns with today's environmental awareness and effectively reduces fossil fuel consumption. EVs will become a crucial mode of transportation in the future. However, due to the intermittent, random, and uncertain nature of EV charging, the uncertainty of EV charging behavior poses a significant challenge to the power system. Large-scale, unregulated charging of EVs (randomly connecting to the grid) can affect grid stability, leading to peak charge accumulation and potentially threatening grid operation.

[0003] When studying the charging load of electric vehicles (EVs), it's easy to overlook weather factors and traffic conditions. Traffic flow exhibits randomness due to weather variations, but current research on weather is limited. Furthermore, many traffic flow prediction datasets are based on plains cities. Mountainous cities, built in mountainous terrain, have different spatial structures due to topographical constraints. Therefore, the impact of terrain on traffic flow should be considered, and datasets should be selected from traffic segments with similar spatial structures to mountainous cities to improve model prediction accuracy. Road traffic conditions significantly affect the energy consumption of EVs during movement, impacting their spatiotemporal load. They also significantly influence the final, orderly scheduling of EVs. Moreover, due to my country's vast territory, complex terrain, and numerous mountainous cities, traditional EV load prediction methods for plains cities neglect the impact of terrain on EV users' travel routes and energy consumption, resulting in poor applicability in cities with unique terrain. Additionally, travel temperature also affects the energy consumption of EVs during operation. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes a method for predicting short-term charging routes for electric vehicles based on environmental factors. This method includes: collecting traffic flow data; denoising the traffic flow data using the exponential moving average method; calculating the correlation of the denoised traffic flow data using the Spearman correlation coefficient formula; normalizing the traffic flow data based on the correlation; inputting the normalized data into a BP-LSTM-Attention prediction model to obtain the traffic flow for the next time period; constructing a flow-density-speed relationship based on the predicted traffic flow; and applying the flow-density-speed relationship... A road impedance model is constructed; the travel time path of the trolley is obtained based on the road impedance model, and the mileage power consumption of the trolley is calculated; the mileage power consumption is processed using the Monte Carlo method to obtain the prediction results of the spatiotemporal distribution of trolley charging load; a trolley link model is constructed, which consists of multiple charging pile nodes; the similarity between the charging pile node and the trolley node is calculated using an optimized similarity calculation formula, and the link between the two nodes is determined based on the similarity; the comprehensive similarity of the link is calculated using a comprehensive similarity formula based on the link and the prediction results of the spatiotemporal distribution of trolley charging load; the optimal travel route from the trolley node to the charging pile node is recommended based on the comprehensive similarity.

[0005] The beneficial effects of this invention are:

[0006] This invention provides preliminary preparation for exploring how to fully utilize the scheduling potential of electric vehicles in mountainous urban areas by predicting the charging load of electric vehicles with the characteristics of mountainous urban areas. The invention constructs a road impedance model based on the flow-density-speed relationship, obtains the shortest travel time path of the electric vehicle based on the road impedance model, calculates the mileage power consumption of the electric vehicle, and finally uses the Monte Carlo method to analyze the mileage power consumption of the electric vehicle, thereby improving the accuracy of the prediction results of the spatiotemporal distribution of electric vehicle charging load. Attached Figure Description

[0007] Figure 1 This is a flowchart of the Monte Carlo simulation of the present invention;

[0008] Figure 2 This is a flowchart of the training process for the BP-LSTM-Attention prediction model of the present invention.

[0009] Figure 3 This is the overall flowchart of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] A method for predicting short-term charging routes for electric vehicles based on environmental factors, such as... Figure 3 As shown, the method includes: constructing a traffic network model; collecting traffic flow data and denoising the traffic flow data using the exponential moving average method; calculating the correlation of the denoised traffic flow data using the Spearman correlation coefficient formula; normalizing the traffic flow data based on the correlation and inputting the normalized data into the BP-LSTM-Attention prediction model to obtain the traffic flow for the next time period; constructing a flow-density-speed relationship based on the predicted traffic flow; constructing a road impedance model based on the flow-density-speed relationship; obtaining the trolleybus travel time path based on the road impedance model and calculating the trolleybus mileage power consumption; processing the mileage power consumption using the Monte Carlo method to obtain the prediction results of the spatiotemporal distribution of trolleybus charging load; constructing a trolleybus link model, which consists of multiple charging pile nodes; calculating the similarity between charging pile nodes and trolleybus nodes using an optimized similarity calculation formula, and determining the link between the two nodes based on the similarity; calculating the comprehensive similarity of the link using a comprehensive similarity formula based on the prediction results of the link and the spatiotemporal distribution of trolleybus charging load; and recommending the optimal travel route from the trolleybus node to the charging pile node based on the comprehensive similarity.

[0012] In real-world road networks, most traffic networks exhibit a mesh-like structure; therefore, graph theory methods are used to model these networks. The expression is:

[0013]

[0014] In the formula, G represents the transportation network of the mountain city; J represents the set of all nodes in graph G; E represents the set of all road segments in graph G; W is the set of road segment weights, i.e., road resistance; K represents the set of time periods, dividing the whole day into m time periods; Q is the traffic flow on the road; V is the vehicle speed on the road; θ is the road segment gradient of the mountain city; and n represents the number of nodes.

[0015] E is the set of road segments in the road network, and its adjacency matrix D = (d ij ) n×n The adjacency relationship between nodes and edges in a road network is described by the following element:

[0016]

[0017] In the formula, The value represents the road weight between node i and node j, which can be quantified using weights such as road segment length and travel time; inf indicates that there is no direct road connection between the two nodes.

[0018] The adjacent edge matrix can be represented as:

[0019]

[0020] The Spearman correlation coefficient is used to study the correlation between two groups of variables. The formula for calculating the Spearman correlation coefficient is:

[0021]

[0022]

[0023] Where n is the sample size; (U i V i Let be the rank of the two variables.

[0024] Real-time dynamic traffic flow is a complex system with high uncertainty and randomness, and its changes are often affected by various random factors such as weather, road conditions, travel costs, and travel time. Although the dynamic changes in traffic flow are complex and constantly fluctuating, there are still certain patterns and periodicities within it, and these characteristics can be revealed through in-depth analysis.

[0025] During the collection of traffic flow data, the complex and ever-changing external environment may lead to outliers in the collected data. To ensure successful modeling of traffic flow data, the impact of noise on the original data should be removed. The Exponential Moving Average (EMA) method is selected to denoise the original traffic flow data, preserving the trend of traffic flow data to the greatest extent possible without increasing algorithm complexity.

[0026] EMA t =αX t +(1-α)EMA t-1

[0027] Among them, EMA t X is the exponential moving average over time t, where α is the smoothing factor, a constant between 0 and 1 that determines the weight of the latest observation. t This represents the value at time t.

[0028] Traffic flow changes to some extent with variations in region and road segment. The Spearman correlation coefficient formula is used to calculate the correlation between current location, current time, and traffic flow. The Spearman correlation coefficient is chosen to construct the correlation matrix, and a significance test is used to evaluate the correlation strength between the two datasets. This explores the strong correlation between weather data such as temperature, humidity, visibility, wind direction, and rain / snow weather on short-term traffic flow.

[0029] Traffic flow data, geographic information, weather data, etc., are used to construct a traffic flow time series matrix of geographic and weather factors according to the observation time series, which is then used as the model data input.

[0030] Since the various attributes in the traffic and weather data used have different dimensional standards, this may affect the accuracy of subsequent prediction models. Therefore, it is necessary to constrain the values ​​of each attribute to a common range. Thus, the traffic flow data is normalized using min-max standardization.

[0031]

[0032] By employing an attention-based BP-LSTM prediction model, the backpropagation-capable BP model is combined with the LSTM model, which excels at uncovering latent temporal features, to capture long-term dependencies between data. An attention mechanism is introduced to allocate weights appropriately based on the strength of the correlation between geographical data, weather data, and traffic flow data, ultimately constructing a BP-LSTM-Attention prediction model.

[0033] In this embodiment, the BP-LSTM-Attention prediction model is used to process the input data, including: filtering the collected traffic flow data and constructing a weather-factor traffic flow time series matrix based on the time series; normalizing the weather-factor traffic flow time series matrix to obtain a standard data matrix; sampling the standard data matrix using the sliding window method to obtain a time series; preprocessing the time series; inputting the preprocessed sequence into the LSTM layer to obtain the dynamic features of the time series; inputting the dynamic features of the time series into the hidden layer for feature enhancement; inputting all enhanced dynamic features into the Attention layer for weight fusion to obtain fused dynamic features; and inputting the fused dynamic features into the classification prediction layer to obtain the classification prediction result.

[0034] The time series matrix of traffic flow due to weather factors is as follows:

[0035]

[0036] Where m represents the number of dimensions, X1, X2, ..., X...m These are weather data from different dimensions, including traffic flow, atmospheric temperature, relative humidity, visibility, and wind direction, t1, t2, ..., t... n This represents the corresponding observation time. Since data of different dimensions have inconsistent dimensional units, this significantly reduces the training efficiency and prediction accuracy of the model. Therefore, the data is normalized so that the model's input is a normalized standard data matrix.

[0037] Using the standardized data samples obtained, a training sequence was established using the sliding window method. First, the time series matrix in the above equation was transformed into:

[0038]

[0039] Next, the shift() function is used to slide the window, and n=1 is set as the length of the sliding window. The data is input into the prediction model in batches according to the time sequence. Finally, after repeated learning and training on a large amount of data, the traffic flow data of 24 time periods per day is combined with the feature data. After feature selection, a traffic flow prediction model is established.

[0040] The model's structure includes an LSTM layer, hidden layers, and attention layers.

[0041] LSTM Layer: First, the LSTM layer is constructed. The input data (preprocessed sequence data) is directly fed into the LSTM layer. The LSTM layer captures the dynamic characteristics of the time series through its internal memory units and mechanisms such as forget gates, input gates, and output gates.

[0042] Hidden layers: As needed, additional fully connected layers (dense layers) or more LSTM layers can be added after the LSTM layers to further extract features or increase the non-linearity of the model.

[0043] Attention layer: The output of the LSTM layer (usually the hidden state of the last time step or the hidden states of all time steps) is fed into the Attention layer. The role of the Attention layer is to assign different weights according to the importance of the hidden state at each time step, thereby focusing on those time steps that have a greater impact on the prediction result.

[0044] like Figure 2 As shown, the model training process uses the backpropagation (BP) algorithm to train the entire model. During training, the model weights are updated using the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²).

[0045] Traffic flow on urban roads is predicted, yielding parameter Q. To evaluate the performance of the prediction model in traffic flow forecasting, mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are used as evaluation metrics. Their mathematical expressions are as follows:

[0046]

[0047]

[0048]

[0049] Among them, y i Indicates the predicted value. Let represent the actual traffic flow at time i, and m represent the predicted total number.

[0050] The unique terrain of mountainous cities affects vehicle speed primarily through road gradient and speed limits, namely:

[0051] V = V max ×e -αθ

[0052] In the formula, α is an adjustment coefficient, representing the degree of influence of road gradient on speed; V max The maximum speed limit is designed for mountain city roads.

[0053] Speed ​​v, flow rate Q, and density K are three parameters used to describe the characteristics of traffic flow. These three parameters are interrelated and together depict the dynamic characteristics of traffic flow. The relationship between the three traffic flow parameters is shown in the formula:

[0054] Q = VK

[0055] Traffic density refers to the number of vehicles per unit length of road within a specific time period, which can reflect the traffic density and congestion of a road segment.

[0056]

[0057] In the formula, K represents vehicle density; N is the number of vehicles in the road segment; and L represents the length of the road segment.

[0058] The relationship between velocity and density is:

[0059]

[0060] In the formula, v represents the speed of the electric vehicle; V f K represents the free-flow velocity. j This indicates the blockage density.

[0061] According to the criteria for classifying urban traffic conditions, road saturation s is used to represent the traffic conditions of a road, as shown in Table 1.

[0062] Table 1 Traffic Conditions Table

[0063]

[0064] When an electric vehicle travels in a road network, the road impedance is usually composed of two parts: road segment impedance and node impedance. Since there are differences in the traffic capacity of road intersections and road segments, road segment impedance and node impedance models corresponding to different saturation s can be established according to their respective traffic capacities.

[0065] The road segment impedance model includes the following: there is a certain relationship between the travel time of a vehicle on a road segment and the road's capacity. Furthermore, weather factors such as changes in temperature can affect vehicle speed, thus impacting travel time. The road segment impedance formulas considering different saturation levels under climatic factors are as follows:

[0066]

[0067]

[0068]

[0069] In the formula, K c This is a climate correction factor that reflects the impact of climate temperature on vehicle travel time on road segments; T c For real-time climate temperature, T max and T min To define the upper and lower limits of the suitable temperature range, we take 20℃ and 25℃ respectively; R vij (t) represents the road segment impedance, i.e., the actual travel time of the road segment considering the temperature effect; t0 represents the travel time under free flow; α and β are impedance influence factors; in formula (16), s represents the road segment saturation; Q represents the traffic flow of the road segment; C D The traffic capacity of the road section.

[0070] Nodal impedance model. In urban transportation networks, the delay of vehicles at road intersections is called nodal impedance, which is mainly determined by the capacity limitations of the intersection and the control of traffic lights.

[0071]

[0072] In the formula, C vi (t) represents the node impedance; c represents the signal period; λ represents the percentage of green light time at the node; and q represents the vehicle arrival rate of the road segment.

[0073] Impedance Model for Mountainous City Roads. Impedance, also known as travel time, consists of road travel time and intersection delay time in the urban traffic network. Considering the characteristics of mountainous city roads, the improved road impedance model calculation formula is shown below. The improved road impedance model reflects the impact of climate temperature and the unique terrain of mountainous areas on vehicle travel within the traffic network, thus more closely reflecting actual conditions.

[0074]

[0075] In the formula, θ represents the slope of the mountainous city road. The slope angle of the road affects the travel time mainly by changing the driving speed of the electric vehicle. On road sections with steep slopes, vehicles usually travel at lower speeds and may frequently change speeds, further increasing travel time and thus changing the user's travel time-space parameters.

[0076] In this embodiment, obtaining the shortest travel time path for the electric vehicle includes: In the unique terrain of mountainous cities, the traditional Floyd algorithm based on horizontal distance is insufficient to capture the true travel time of the EV in the road network. Therefore, an improved Floyd algorithm is adopted, using the actual travel time of the electric vehicle in the mountainous city to replace the traditional road distance weight, in order to determine the shortest travel time between each node. The improved algorithm steps include:

[0077] Step 1: Initialize the adjacency matrix w (0) Set the initial time cost between i and j to be If points i and j are not directly connected, then T total =∞;

[0078] Step 2, construct w (1) To optimize time cost, insert the first intermediate node between i and j, and calculate...

[0079] Step 3, construct w (2) Insert a second node between the two points and calculate...

[0080] Step 4: Construct w (n) ,calculate Then w (n) w represents the shortest time cost between nodes i and j after traversing all nodes. (n) This forms the shortest time cost matrix between each node.

[0081] In this embodiment, the energy consumption per unit mile of an electric vehicle is calculated using an energy consumption factor model, which determines the energy consumption per unit mile when the EV travels at a certain speed on different roads. The expression is:

[0082]

[0083] In the formula, V k The vehicle's speed is affected by road traffic conditions and speed limits based on road grade; ε k This represents the power consumption per unit distance when an EV travels on a k-level road, where k includes level 1, level 2, level 3, and level 4 roads.

[0084] Current research on the impact of travel temperature on EV energy consumption is limited in predicting EV charging load. The main effects of temperature changes on EV energy consumption are related to the battery and air conditioning. The correlation between EV energy consumption per unit driving distance and ambient temperature is shown in the following formula.

[0085]

[0086] In the formula, T represents the ambient temperature of the trip, in degrees Celsius (°C); C(T) is the power consumption per unit mileage at different temperatures. Table 2 gives the values ​​of the coefficients in the model.

[0087] Table 2 shows the values ​​of the coefficients in the model.

[0088]

[0089] When the ambient temperature is around 20℃, the energy consumption of the EV reaches its minimum. Therefore, a proportionality coefficient m for energy consumption variation at different temperatures is established with 20℃ as the baseline. T The power consumption ratio at temperature T is shown below:

[0090]

[0091] Taking into account the impact of traffic conditions and ambient temperature on power consumption, the power consumption per unit mile of an EV can be obtained, expressed as:

[0092]

[0093] In this embodiment, the impact of terrain on power consumption includes: given the frequent undulations and concentrated ups and downs in mountainous cities, when analyzing the actual power consumption of an EV driving in a mountainous city road network, the road should be divided into multiple segments, each with its unique slope and curvature. Calculating the actual distance between any two points requires considering the ups and downs, and is no longer simply a straight-line distance. The actual distance between two points is obtained by decomposing the road between the two points into multiple segments and calculating the distance of each segment, resulting in the actual cumulative distance, as shown in the following formula:

[0094]

[0095] In the formula, Taking into account the unique topography of mountainous cities, the actual spatial distance between points i and j; d c for The actual spatial distance of segment c is:

[0096]

[0097] In the formula, x c and h c These represent the horizontal distance and vertical height difference of the c-th road segment, respectively.

[0098] Therefore, taking into account the impact of traffic network impedance on travel time, temperature, road conditions, etc. in mountainous cities, the power consumption of electric vehicles driving on mountainous city roads is as follows:

[0099]

[0100] In the formula, a i,j This is the efficiency coefficient for an electric vehicle in terms of work done against gravity or energy recovery per unit relative height, expressed in m / kWh. Where a i,j It can be shown by the following formula.

[0101]

[0102] When an electric car travels uphill, it needs to do work against gravity, then a i,j Let a be the climbing coefficient. c When an electric vehicle is driving downhill, it can recover some energy due to braking, then a i,j Let a be the energy recovery coefficient. d .

[0103] In this embodiment, the prediction of the spatiotemporal distribution of charging load includes: based on the analysis of EV user behavior, the probability distribution of spatiotemporal variables of user travel behavior can be obtained as input for the Monte Carlo method. When simulating user charging demand, it is necessary to comprehensively consider the influence of external factors and spatiotemporal variables, and to alternately sample the spatiotemporal variables.

[0104] In this embodiment, as Figure 1As shown, the Monte Carlo method is used to process mileage-based energy consumption, including: constructing a probability distribution of user travel behavior for future time periods using the Monte Carlo method; generating various potential charging scenarios through extensive random simulations; summarizing all charging scenarios to obtain the load probability distribution; planning user travel routes based on the load probability distribution; using the Floyd algorithm to plan user travel routes with the shortest travel time as the objective; obtaining the travel time of each road based on the speed-flow-density model; updating the spatiotemporal variables and state of charge of the EV after it reaches its destination; calculating travel energy consumption based on the duration and mileage of the current trip, combined with the influence of terrain and temperature on energy consumption; updating the remaining battery power based on travel energy consumption; analyzing user charging needs to obtain the spatiotemporal distribution of each electric vehicle's travel and information on each destination; when a vehicle arrives at a destination, determining whether the user needs to charge based on the remaining battery power; if charging is needed, determining the charging duration based on the dwell time and battery power; calculating the electric vehicle charging load for each destination, i.e., each functional area; if charging is not needed, continuing to drive.

[0105] After travel demand is generated, the user's travel route needs to be planned. The Floyd algorithm is used to plan the user's travel route with the shortest travel time as the objective. The travel time of each road is obtained based on the speed-flow-density model. After the EV arrives at its destination, its spatiotemporal variables and state of charge are updated.

[0106] The formula for the charging probability is:

[0107]

[0108] In the formula, Indicates whether the EV will charge at destination 's', 1 indicates the user has a charging need, 0 indicates the user will not charge; remaining range. The power consumption per unit mile for electric vehicles in mountainous cities takes into account the effects of road conditions and temperature; L s,s+1 This represents the actual driving distance for the user's next trip within the mountainous city road network.

[0109] When an EV user arrives at each destination, the energy consumption is calculated based on the trip's duration and mileage, combined with the impact of terrain and temperature on energy consumption. The remaining battery capacity is then updated to estimate the charging load during the stop at that destination and integrate it into the corresponding area's charging demand. After each trip, the EV user's initial state of charge upon arrival at the destination is determined based on the EV's mileage and route, combined with an established mountainous city energy consumption model. The expression is:

[0110]

[0111] In the formula, SOC(T) represents the amount of electricity used to reach the current destination and leave the previous location; d_0 The battery level at the moment of the first trip is related to the user's habits; E max The maximum capacity of an EV depends on battery parameters; D d and ε d N represents the distance traveled by the electric vehicle and the power consumption per kilometer under corresponding road conditions. S-1,S This represents the total number of road segments traversed from the starting point to the destination based on the shortest travel time route planning.

[0112] Assuming an EV user chooses to charge at a certain location, the user will immediately connect to the charging facility upon arrival and continue charging until the battery reaches the predetermined ideal charge level or maximum dwell time. The charging time is as follows:

[0113]

[0114] In the formula, η is the charging efficiency of the EV; P s Indicates the EV charging power level at point s; t stay E represents the length of time an EV stays at a destination; exp This represents the user's desired battery level, typically 1.

[0115] When an EV user leaves its current destination, if the EV chooses not to charge after a charging decision at the current destination, its state of charge upon departure will be the same as when it arrived at the current destination; if the EV is charged at the destination, it will choose to fully charge the EV or leave the current destination.

[0116]

[0117] In the formula Indicates the vehicle's dwell time; η represents the charging efficiency; P s This indicates the charging power level at point s.

[0118] By analyzing users' charging needs, we can obtain the spatiotemporal distribution of each electric vehicle's journey and information on its destinations, including arrival time, dwell time, departure time, and charge level at each time. This allows us to calculate the electric vehicle charging load for each destination, i.e., each functional area. When a vehicle arrives at a destination, the user makes a charging decision based on the remaining battery power. If charging is required, the charging duration is determined based on the dwell time and battery level. The charging load at location f can be expressed as:

[0119]

[0120] In the formula, This represents the charging state flag of the electric vehicle at time t, with 1 indicating that it is charging and 0 indicating that it is not charging; N f (t) represents the number of electric vehicles in this functional area at time t; This represents the total charging load of electric vehicles in this functional area at time t.

[0121] A trolleybus link model is constructed, consisting of multiple charging pile nodes. This model recommends suitable charging stations for trolleybuses and incorporates information such as load forecasting, time-of-use pricing, charging service fees, and passenger source forecasting as internal node attributes into the overall network topology. In this research, electric taxis and charging stations are abstracted as nodes in the network, and encounter records between nodes are transformed into links within the network, thereby enabling the prediction and recommendation of potential future links.

[0122] The trolley link model consists of two types of nodes: trolley nodes and charging station nodes. The definition and attributes of the nodes are as follows: Trolley node (v i (∈V) represents every tram running in the city. Location P i : indicates tram v i The location coordinates in time slot t are usually expressed in latitude and longitude. Current load forecast L j This reflects the vehicle's charging needs and is mainly used to assess whether charging should be a priority for the vehicle.

[0123] Current load forecast L i The expression is:

[0124]

[0125] Among them, C max C represents the maximum battery capacity of the electric vehicle. i (t) represents the tram v i The remaining battery capacity in time slot t, This represents the current percentage of battery charge consumed, reflecting the electric vehicle's charging needs; T i (t) represents the tram v i The current waiting time in time slot t is used to assess the potential time cost of the charging station; β1 is an adjustment coefficient used to balance the impact of remaining power and waiting time.

[0126] User D is expected to i It indicates the potential passenger demand in the area where the vehicle is located, and is used to assess the passenger carrying capacity of the area during a certain period of time, affecting whether the vehicle chooses to charge.

[0127] D i =α1·H i (t)+α2·S i (t)+α3·P i (t)

[0128] Among them, H i (t) represents historical user boarding data, indicating the historical user demand for the tram at its current location. i (t) represents the pedestrian density in the area where the vehicle ahead is located, which can be obtained through real-time ride-hailing application data or pedestrian flow monitoring data; P i (t) represents the predicted impact of special factors such as weather and holidays, which are usually predicted using statistical models or machine learning methods; α1, α2, and α3 are adjustment coefficients used to balance the degree of influence of different factors.

[0129] Charging station node: Represents the various charging stations in a city. Each charging station node contains the following attributes:

[0130] Position P j : indicates charging station c j The fixed position coordinates.

[0131] Current load forecast L j : indicates charging station c j The current load forecast reflects the busyness of the charging station and is mainly used to assess the congestion situation of the charging station.

[0132]

[0133] Where: N j (t) represents charging station c j The current number of vehicles in the queue at time slot t; N max This refers to the maximum service capacity of a charging station, i.e., the maximum number of vehicles it can serve simultaneously. The current queuing rate reflects the busyness of the charging station; W j (t) represents charging station c j The average waiting time is used to further evaluate the waiting cost of vehicles at charging stations; γ1 is an adjustment factor used to balance the impact of queuing rate and waiting time.

[0134] Time-of-use electricity price P j This reflects the charging cost at the current moment, as electricity prices change over time.

[0135] Charging service fee F j This indicates additional charging service fees, which are usually a fixed fee or charged according to the type of service.

[0136] The connection between trolley nodes and charging station nodes is defined as a link (e ij Link weights are used to represent possible charging behaviors. The weights of links are determined by the similarity between nodes and their respective internal attributes. The calculation of link weights combines historical encounter records and internal node information, enabling a more accurate reflection of possible future charging behaviors.

[0137] To predict potential links between EVs and charging stations, similarity calculations are based on a modified CAR-based Adamic / Adar Index (CAA) combined with a time decay function. The specific similarity calculation formula is as follows:

[0138]

[0139] Where S(x,y) is the similarity score between nodes x and y, Γ(x) is the set of neighboring nodes of node x, Γ(y) is the set of neighboring nodes of node y, k(i) is the degree value of the common neighbor node i, γ(i) is the subset of the neighbors of node i, and p(t,t) is the similarity score between nodes x and y. i,x ,t i,y ) is the time decay function, λ is the time decay coefficient, and t is the current time slot. i,x Let t be the time slot where neighbor node i and node x last met. i,y Let y be the time slot where neighbor node i and node y last met.

[0140] Based on the basic similarity calculation, information from nodes such as load forecasting, time-of-use pricing, charging service fees, and passenger demand forecasting is incorporated into the similarity calculation to form a comprehensive similarity formula:

[0141]

[0142] Wherein, S′(v i ,c j (v) is for tram i and charging station c j The overall similarity score between them, S(v) i ,c j (v) is for tram i and charging station c j The similarity values ​​between them, where α, β, γ, and δ are adjustment coefficients, and L j For charging station c j The current load forecast value, P j For charging station c j The current time-of-use electricity price, F j For charging service fees, D i To meet the needs of electric vehicle users.

[0143] In this embodiment, the recommended objective is to minimize the additional travel distance recommended by the charging station, so that the electric vehicle can deviate from its original planned route as little as possible while charging. The formula for the recommended objective is as follows:

[0144]

[0145] in, To determine the additional travel distance for recommended charging stations, ||c j -p i (t)|| represents the tram v i From the current position p i (t) to charging station c j The distance, ||p i (t+τ)-c j ||After charging, from charging station c j Move to future destination p i The distance (t+τ), ||p i (t+τ)-p i (t)|| represents the shortest path length from the current time t to the future time t+τ.

[0146] To improve recommendation performance, a Top-K recommendation method is used, ranking similarity scores and selecting the K charging stations with the highest scores as the final recommendation results. The specific steps are as follows: Calculate the similarity matrix A. x Each element in the similarity matrix represents the similarity score between nodes; the results in the similarity matrix are sorted in descending order of score; the top K charging stations with the highest scores are selected to form the recommendation sequence Ω. x (t)=(ω1,ω2,…,ω K ).

[0147] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting short-term charging routes for electric vehicles based on environmental factors, characterized in that, include: Constructing a transportation network model for mountainous cities; Traffic flow data is collected and denoised using the exponential moving average method. The correlation of the denoised traffic flow data was calculated using the Spearman correlation coefficient formula. Traffic flow data is normalized based on correlation, and the normalized data is input into the BP-LSTM-Attention prediction model to obtain traffic flow for the next time period. A flow-density-speed relationship is constructed based on the predicted traffic flow. A road impedance model is then built based on this relationship. The shortest travel time path for trolleybuses is obtained from the road impedance model, and the mileage power consumption is calculated. The Monte Carlo method is used to process the mileage power consumption to obtain the prediction results of the spatiotemporal distribution of trolleybus charging load. A trolleybus link model is constructed, consisting of multiple charging pile nodes. The similarity between charging pile nodes and trolleybus nodes is calculated using an optimized similarity calculation formula, and the link between the two nodes is determined based on the similarity. Finally, a comprehensive similarity formula is used to calculate the comprehensive similarity of the link based on the prediction results of the link and the spatiotemporal distribution of trolleybus charging load. The optimal route from the trolley node to the charging pile node is recommended based on the overall similarity. The road impedance model constructed based on the flow-density-velocity relationship includes: the road segment impedance formula corresponding to different saturation levels under climatic factors is as follows: In the formula, K c T is the climate correction factor. c For real-time climate temperature, T max and T min The upper and lower limits of the suitable temperature range; t0 is the travel time under free flow; α and β are impedance influence factors; s represents the road segment saturation; Q is the traffic flow of the road segment; C D The traffic capacity of the road section; An improved road impedance model is constructed based on road segment impedance: Where θ is the slope of the mountain city road, and C vi (t) represents the node impedance; The optimized similarity calculation formula is used to calculate the similarity between charging pile nodes and trolleybus nodes, including: Where S(x,y) is the similarity score between nodes x and y, Γ(x) is the set of neighboring nodes of node x, Γ(y) is the set of neighboring nodes of node y, k(i) is the degree value of the common neighbor node i, γ(i) is the subset of the neighbors of node i, and p(t,t) is the similarity score between nodes x and y. i,x ,t i,y Let λ be the time decay function, t be the time decay coefficient, and t be the current time slot. i,x Let t be the time slot where neighbor node i and node x last met. i,y Let y be the time slot where neighbor node i and node y last met.

2. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, Constructing a traffic network model involves using graph theory methods to model the traffic network, and its expression is as follows: Where G represents the transportation network of the mountain city; J represents the set of all nodes in graph G; E represents the set of all road segments in graph G; W represents the set of road segment weights, i.e. road resistance; K represents the set of time periods, dividing the whole day into m time periods; Q represents the traffic flow on the road; V represents the vehicle speed on the road; θ represents the road segment gradient of the mountain city; and n represents the number of nodes.

3. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, The BP-LSTM-Attention prediction model is used to process the input data, including: filtering the collected traffic flow data and constructing a weather-factor traffic flow time series matrix based on the time series; normalizing the weather-factor traffic flow time series matrix to obtain a standard data matrix; sampling the standard data matrix using the sliding window method to obtain a time series; preprocessing the time series; inputting the preprocessed sequence into the LSTM layer to obtain the dynamic features of the time series; inputting the dynamic features of the time series into the hidden layer for feature enhancement; inputting all enhanced dynamic features into the Attention layer for weight fusion to obtain the fused dynamic features; and inputting the fused dynamic features into the classification prediction layer to obtain the classification prediction result.

4. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, Constructing the flow-density-velocity relationship includes: The velocity expression is: V=V max ×e -αθ The relationship between velocity, flow rate, and density is expressed as follows: Q = VK The density expression is: The relationship between velocity and density is: Where α is the adjustment coefficient, V max The maximum design speed limit for mountain city roads; L is the length of the road segment; K represents vehicle density; N is the number of vehicles in the road segment; V f K represents the free-flow velocity. j This indicates the blockage density.

5. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, The shortest travel time path for the tram is obtained by processing the data collected from each node using an improved Floyd algorithm. The specific steps include: Step 1: Initialize the adjacency matrix w (0) Set the initial time cost between i and j to be If points i and j are not directly connected, then T total =∞; Step 2, construct w (1) To optimize time cost, insert the first intermediate node between i and j, and calculate... Step 3, construct w (2) Insert a second node between the two points and calculate... Step 4: Construct w (n) ,calculate Then w (n) w represents the shortest time cost between nodes i and j after traversing all nodes. (n) This forms the shortest time cost matrix between each node.

6. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, The calculation of trolley power consumption per mile includes: Among them, E d Electricity consumption per kilometer of the tram. Let m represent the energy consumption per unit distance when the EV travels on a Class k road, and m be the travel distance from the starting point to the destination, x be the energy consumption per unit distance. i,j Let a be the horizontal distance between points i and j. i,j Let a be the climbing coefficient. c h i,j Let ε be the relative height difference between points i and j in the vertical direction. k m is the power consumption per unit distance when an EV is driven on a Class k road. T This represents the proportionality coefficient of energy consumption variation at different temperatures.

7. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, The Monte Carlo method is used to process mileage-based energy consumption, including: constructing a probability distribution of user travel behavior for future time periods using the Monte Carlo method; generating various potential charging scenarios through extensive random simulations; summarizing all charging scenarios to obtain the load probability distribution; planning user travel routes based on the load probability distribution; using the Floyd algorithm to plan user travel routes with the shortest travel time as the objective; obtaining the travel time of each road based on the speed-flow-density model; updating the spatiotemporal variables and state of charge of the EV after it reaches its destination; calculating the travel energy consumption based on the duration and mileage of the trip, combined with the impact of terrain and temperature on energy consumption; updating the remaining battery power based on the travel energy consumption; analyzing user charging needs to obtain the spatiotemporal distribution of each electric vehicle's travel and information on each destination; when a vehicle arrives at a destination, determining whether the user needs to charge based on the remaining battery power; if charging is needed, determining the charging duration based on the dwell time and battery power; calculating the electric vehicle charging load for each destination, i.e., each functional area; if charging is not needed, continuing to drive.

8. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, The formula for comprehensive similarity is: Wherein, S′(v i ,c j (v) is for tram i and charging station c j The overall similarity score between them, S(v) i ,c j (v) is for tram i and charging station c j The similarity values ​​between them, where α, β, γ, and δ are adjustment coefficients, and L j For charging station c j The current load forecast value, P j For charging station c j The current time-of-use electricity price, F j For charging service fees, D i To meet the needs of electric vehicle users.

9. The method for predicting short-term charging routes for electric vehicles based on environmental factors according to claim 1, characterized in that, The formula for recommending the optimal route from the trolley node to the charging station node is: in, To determine the additional travel distance for recommended charging stations, ||c j -p i (t)|| represents the tram v i From the current position p i (t) to charging station c j The distance, ||p i (t+τ)-c j ||After charging, from charging station c j Move to future destination p i The distance (t+τ), ||p i (t+τ)-p i (t)|| represents the shortest path length from the current time t to the future time t+τ.

Images