A Spatiotemporal Situation Prediction Method for Ultra-fast Charging Load in Urban Distribution Networks under Heavy Rainfall Weather

Abstract

The present invention discloses a method for predicting the spatio-temporal situation of the ultra-fast charging load of urban distribution networks under heavy rainfall weather. S1. Construct an electrical performance failure probability model under ground water depth, rainfall intensity and time course change; S2. Construct a spatio-temporal transfer matrix model of the ultra-fast charging load considering the availability of ultra-fast charging piles and users' charging risk avoidance behaviors driven by the rainfall intensity time course; S3. Divide the charging radiation area with the ultra-fast charging station as the node, map the rainfall intensity and time course data to the nodes of the charging radiation area, combine the user electrical performance failure probability model and the spatio-temporal transfer matrix model of the ultra-fast charging load to obtain the operating state characteristics of the ultra-fast charging pile, the rainfall intensity characteristics and the traffic flow characteristics, construct a dynamic correlation model of the ultra-fast charging load based on spatio-temporal graph modeling, and obtain the corresponding load prediction value by inputting the current operating state characteristics of the ultra-fast charging pile, the rainfall intensity characteristics and the traffic flow characteristics into the dynamic correlation model of the ultra-fast charging load.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and particularly to a method for predicting the spatio-temporal situation of ultra-fast charging loads in urban distribution networks under heavy rainfall weather conditions. Background Art

[0002] In a new power system, the wide access of a high proportion of new energy promotes green and low-carbon development while also challenging the flexible regulation ability on the power supply side, and the power system's supply-demand balance faces greater uncertainties. With the increasing penetration rate of electric vehicles in urban distribution networks, their charging loads show high randomness and unevenness in the spatio-temporal dimension. As high-power centralized charging facilities, ultra-fast charging stations can effectively meet the demand for rapid energy replenishment, but they also significantly increase the load management and regulation pressure on the distribution system. Ultra-fast charging loads have the characteristics of significant large power, fast change, strong concentration, and high dependence on user choices. Their spatio-temporal distribution is affected by the superposition of multiple factors such as travel behavior, traffic network operation status, and meteorological conditions, showing strong uncertainty and volatility. Especially under heavy rainfall weather conditions, the spatial distribution of ultra-fast charging loads will face more complex disturbing factors, such as charging facilities being unable to operate normally due to waterlogging, the reconstruction of charging paths caused by user risk avoidance behaviors, and the instantaneous transfer and aggregation of loads in the spatial dimension. At the same time, factors such as the concentration of travel time periods during rainfall, traffic flow delays, and route adjustments further amplify the impact of load peaks on the distribution system, which may cause problems such as local overload, voltage fluctuations, and even over-limit, and may even induce feeder overload or equipment tripping in severe cases, affecting the safe and stable operation of the power grid.

[0003] Existing research on electric vehicle load forecasting mostly focuses on slow charging and fast charging scenarios, and there are obvious technical gaps in the modeling of ultra-fast charging load situations under heavy rainfall conditions. Moreover, existing methods generally ignore the dual disturbances of rainfall type characteristics such as rainfall intensity and time course distribution on the flooding failure of ultra-fast charging piles and user charging behaviors, making it difficult to accurately depict complex phenomena such as path interruption, spatial aggregation, and load sudden change under heavy rainfall, and lacking the collaborative forecasting ability for the dynamic changes of ultra-fast charging loads and multi-factor coupling. Intermittent heavy rainfall is characterized by strong suddenness and uncertain duration, which is likely to cause rapid accumulation of surface water in a short time. And when the water accumulation is not completely drained during the intermittent period, it may repeatedly trigger sensor actions, putting the charging piles in a complex environment of alternating flooding and recovery. At the same time, traffic congestion, road section closures, and decreased traffic efficiency caused by rainfall will significantly change the spatial accessibility of vehicles, easily triggering a dynamic charging behavior pattern of "rain avoidance - concentration - rain avoidance again" for users, further exacerbating the load impact during the rainfall time course intermittent period and amplifying the emergency bearing pressure when the charging piles resume operation. Summary of the Invention

[0004] In view of this, the present invention provides a method for predicting the spatio-temporal situation of the overcharging load of urban distribution networks under heavy rainfall weather, so as to at least solve the problem that the existing methods for predicting the spatio-temporal situation of the overcharging load of urban distribution networks are difficult to be applicable to heavy rainfall weather conditions.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for predicting the spatio-temporal situation of the overcharging load of urban distribution networks under heavy rainfall weather, comprising the following steps:

[0007] S1. Based on the water accumulation evolution process under heavy rainfall scenarios, considering surface runoff and local drainage capacity, dynamically simulate the relationship between the ground water depth in the area where the overcharging pile is located and the rainfall intensity and duration per unit time, establish a water level response threshold model for the sensor at the bottom of the overcharging pile, and, with the sensor trigger height as the boundary, combine the trigger delay to issue a control signal or power-off protection when the water level threshold is exceeded, and then construct an electrical performance failure probability model under the ground water depth, rainfall intensity and time course changes;

[0008] S2. Analyze and consider the dynamic transfer characteristics of the user charging selection space in the flooded sections of the urban traffic network, obtain the pulsed aggregation characteristics of the overcharging load caused by the superposition of the concentrated travel and charging behaviors of users under the intermittent rainfall time course, and construct a spatio-temporal transfer matrix model of the overcharging load considering the availability of overcharging piles and the charging risk avoidance behaviors of users driven by the rainfall intensity time course;

[0009] S3. Divide the charging radiation area with the overcharging station as the node, map the rainfall intensity and time course data to the nodes of the charging radiation area, combine the user electrical performance failure probability model and the spatio-temporal transfer matrix model of the overcharging load to obtain the operating state characteristics of the overcharging pile, the rainfall intensity characteristics and the traffic flow characteristics, construct a dynamic correlation model of the overcharging load based on spatio-temporal graph modeling, and obtain the corresponding load prediction value by inputting the current operating state characteristics of the overcharging pile, the rainfall intensity characteristics and the traffic flow characteristics into the dynamic correlation model of the overcharging load.

[0010] Preferably, the specific content of the electrical performance failure probability model includes:

[0011] ,

[0012] wherein, is the electrical performance failure probability of the overcharging pile, is the failure rate of the overcharging pile at time is the time step; is the influence function of insulation deterioration on the performance failure of the overcharging pile after being flooded;

[0013] Among them, the specific content of the overcharging pile failure rate includes:

[0014] ,

[0015] wherein, is the electrical performance attenuation coefficient of the ultra-fast charging pile after being flooded; is the damping coefficient; is the water depth detected by the sensor; is the waterproof efficiency coefficient of the ultra-fast charging pile; is the n benchmark waterproof threshold when the protection level of the ultra-fast charging pile in the ultra-fast charging station is the benchmark waterproof level; is the soaking time coefficient; is the corrected cumulative immersion time detected by the sensor; is the anti-inlet time threshold of the ultra-fast charging pile;

[0016] The influence function of insulation deterioration on the performance failure of the ultra-fast charging pile after being flooded is related to the insulation resistance, and the specific content includes:

[0017] ,

[0018] ,

[0019] wherein, is the failure sensitivity coefficient; is the insulation resistance; is the safety threshold resistance; is the initial insulation resistance; is the water absorption aging coefficient; is the conductivity of water; is the water pressure coupling acceleration coefficient, indicating the insulation degradation rate after seal failure; is the step function, is the real-time water pressure borne by the seal structure, which is determined by the water depth in the area where the ultra-fast charging pile is located; is the water pressure reference value.

[0020] A preferred embodiment is that the specific content of obtaining the water depth in the area where the ultra-fast charging pile is located includes:

[0021] The heavy rainfall process is evolved into a piecewise function with alternating rainfall periods and intermittent periods:

[0022] ,

[0023] wherein, is the rainfall intensity sequence, k are respectively the rainfall intensity, start time and duration of the th rainfall; k is the k nd rainfall and the kThe intermittent time between +1 rainfall events;

[0024] The water depth in the area where the ultra-fast charging pile is located is determined by rainfall input, drainage output, and neighborhood overflow. During the time interval, rainfall pauses but the drainage system continues to work. A continuous rainfall intensity - drainage balance model is constructed:

[0025] ,

[0026] In the formula, is the water depth in the area of the ultra-fast charging pile at time t ; is a piecewise indicator function. If t is within the rainfall period , it takes 1, otherwise it takes 0, and is used to simulate rainfall intermittency; is the drainage rate, assumed to be proportional to the water level, f is the drainage capacity coefficient; is the set of geographical areas adjacent to the ultra-fast charging station, y refers to any surrounding area of is the surrounding area y 's overflow transmission coefficient to the current area x , which is determined by the slope between the ultra-fast charging pile area and the surrounding area; is the water depth in the surrounding area y at t time;

[0027] The rainfall input is obtained in the form of a piecewise function, and the water accumulation evolution process is affected by time-intermittency disturbances. Based on the continuous rainfall intensity - drainage balance model, discretization is performed to establish a time-step piecewise discretization model of water depth under intermittent heavy rainfall:

[0028] ,

[0029] In the formula, is the water depth in the area of the ultra-fast charging pile at time t after discretization, is the water depth in the surrounding area y at t time after discretization;

[0030] Affected by circuit delay and noise, the detected value of the actually deployed sensor and the corrected cumulative immersion time of the sensor detection are:

[0031] ,

[0032] ,

[0033] wherein, is the sensor response delay period; i is an intermediate variable, i = t - t 1 + 1, t 1 is the detection time threshold for the water immersion state of the ultra-fast charging pile; is the measurement noise, which follows a Gaussian distribution ; is the cumulative water immersion time; is i the indicator function of the water depth in the ultra-fast charging pile area at time , when is satisfied, the value is 1, and if not satisfied, the value is 0; is the sensor trigger threshold.

[0034] Preferably, the specific content of the ultra-fast charging load spatio-temporal transfer matrix model includes:

[0035] ,

[0036] ,

[0037] wherein, is the selection probability of users in area for the ultra-fast charging station n ; is the set of users in area , u represents any user in is the probability of users using ultra-fast charging; User u selects the probability of charging at the ultra-fast charging station n ; is the probability that user u selects the current path at the current time, OD refers to the path; is the ultra-fast charging load spatio-temporal transfer matrix model, is the probability that any user in area R selects the ultra-fast charging station n at each time, R is all 's set.

[0038] Preferably, the specific content of obtaining the probability u that user selects the current path at the current time includes:

[0039] Use the traffic network topology to reflect the dynamic changes of the waterlogged area. It is assumed that the passability in the road network is related to the water depth of the road section. When the water depth of the road section exceeds the set critical threshold, the road section is regarded as impassable during the user's path selection process. At this time, the traffic capacity of the road section is judged to be zero, the path impedance tends to infinity, and the user re-plans the path. Then, the passability model of urban roads under heavy rainfall is as follows:

[0040] ,

[0041] ,

[0042] In the formula, is the path impedance of path rs ; represents the length of path rs ; is the passing speed of path rs ; is the water depth of the road section; is the normal speed when the vehicle is driving; is the traffic capacity coefficient, which changes dynamically with the rainfall intensity, road water depth and traffic state, and reflects the travel efficiency; z is the passing sensitivity factor; is the water depth reference attenuation parameter;

[0043] Then the probability u that the user selects the current path at the current moment is:

[0044] ,

[0045] ,

[0046] In the formula, is the risk perception factor of the user for path rs , and are the travel choices of the user affected by the weather conditions and historical experience respectively, p is the total number of paths rs .

[0047] Preferably, the specific content of obtaining the probability that the user uses the supercharger includes:

[0048] When the user travels and the battery level is lower than the threshold, a charging demand is generated. The overall rainfall time series is solved into a series of rainfall segments and intermittent segments, and all rainfall intermittent periods are extracted, where and are thek Based on the start time and duration of each rainfall, and the start and end times, spatial positions, and durations of each rainfall intermittent segment, the kernel density estimation method is used to calculate the spatial aggregation factor of the charging load:

[0049] ,

[0050] ,

[0051] In the formula, represents the impact of intermittent rainfall on the concentrated travel and charging of users. The longer the intermittency, the more inclined users are to travel concentratedly; q is the influence parameter; represents t the aggregation intensity of user charging in the area where the supercharger station is located at time n ; is the number of traveling users in the area where the supercharger station is located during the intermittent period n ; is used to control the intensity of the pulsed influence; represents the pulsed intensity of the charging event during the intermittent period after the k th rainfall. and are the mean and standard deviation of the charging event intervals respectively;

[0052] Considering the influence of intermittent rainfall, users dynamically judge whether to preferentially choose to use a supercharger pile for charging and energy replenishment based on the duration of the current rainfall intermittent period:

[0053] ,

[0054] In the formula, is the influence factor of rainfall duration on user selection; is the maximum reference value of the rainfall intermittent period; is the state of charge of the user at time t; is the minimum state of charge required for the user to charge at time t; is a piecewise indicator function, which takes 1 if the condition is satisfied, otherwise it takes 0.

[0055] Preferably, obtaining the probability u that the user n chooses to charge at the supercharger station specifically includes:

[0056] Calculating the user charging probability based on the characteristics of user concentrated travel and charging and the availability of supercharger piles under intermittent rainfall time history:

[0057] ,

[0058] In the formula, User u The probability of choosing to charge at a supercharging station n ; Indicates t The agglomeration intensity of user charging in the area where the supercharging station is located at time n ; For the user u To the supercharging station n The length of the shortest passable path, which is infinite when the road is flooded; N Is the total number of supercharging stations.

[0059] Preferably, the specific content of S3 includes:

[0060] Taking N Supercharging stations as graph nodes to divide the charging radiation area of the urban distribution network, defining the node set as , discretizing time into a set of non-overlapping intervals , defining the time t The graph structure at time is , Indicates the set of features of all nodes in the time period ; E Indicates the set of connection edges between supercharging station nodes, which is used to represent the spatial proximity between regions. When the radiation areas of supercharging station i And supercharging station j The radiation areas of the nodes are adjacent, and the value of the connection edge Is defined as 1, otherwise 0;

[0061] Extract the basic state features of each supercharging station node to form a node state vector. The feature vector of each node s n,0 Includes the operating state features of the charging piles , rainfall intensity features And traffic flow features , using Sigmoid to construct a node adaptive dynamic adjacency matrix:

[0062] ,

[0063] ,

[0064] In the formula, Is the adjacency weight, which is jointly determined by the node feature similarity and spatial proximity; And Are learnable parameters that control the weights of feature differences and spatial proximity; Is the supercharging station i And supercharging station j The static adjacency relationship between;

[0065] Extract high-order spatio-temporal correlation features through a multi-layer graph network for each supercharger station n Extract the spatial correlation between it and adjacent nodes, and use the permutation invariance principle to aggregate and update the features of all adjacent supercharger station nodes:

[0066] ,

[0067] ,

[0068] wherein, is the aggregated feature of the adjacent nodes of the l -th layer supercharger station node n ; is the set of adjacent supercharger stations of supercharger station n , and is any node in; is the n -th layer adjacent feature transformation matrix of node l ; and are the input and output matrix orders; is the l- -st layer node feature vector; is the l -th layer supercharger station n feature vector; is the n [[ID= fifty-three]]-th layer node feature transformation matrix of node l ; L ]>is the number of layers of the graph neural network; all trainable parameters in each layer are shared on each node;

[0069] Obtain the t +1 moment TGE embedding vector:

[0070] ,

[0071] wherein, and are the outputs of the first and second hidden layers; and are the weight matrices of each hidden layer; and are the bias vectors of each output layer; is the time context embedding vector; is the current state variable, representing the time t +1 timing information;

[0072] Combine the node feature with the output layer weight matrix Multiply, perform non-linear transformation through an activation function, and multiply the result with the final prediction weight vector Multiply, and obtain the final predicted supercharging load through the activation function again , after passing through l layers of the network, obtain t the load prediction value of the node at time n +1:

[0073] ,

[0074] ,

[0075] wherein, is the output feature of the last layer of the graph network, Contact is the process of vector concatenation, is the L layer of the supercharging station n feature vector; is the supercharging station n at t the predicted supercharging load value at time +1; is the final prediction weight vector;

[0076] Through the above technical solutions, it can be seen that compared with the prior art, the present invention discloses a method for predicting the spatio-temporal situation of the supercharging load of urban distribution networks under heavy rainfall weather, and has the following beneficial effects:

[0077] 1. The present invention analyzes the pulsed aggregation characteristics of the spatio-temporal distribution of the supercharging load of electric vehicles in the distribution network under heavy rainfall weather. Existing models do not consider the dual disturbances of rainfall intensity, time course distribution and other rainfall type characteristics on the flooding failure of supercharging piles and user charging behaviors, and it is difficult to depict the spatio-temporal aggregation characteristics of supercharging load charging behaviors in a heavy rainfall environment. The present invention derives an electrical performance failure probability model for supercharging piles after flooding based on the water accumulation evolution process and waterproof structure threshold under heavy rainfall scenarios, analyzes the dynamic transfer characteristics of the user charging selection space considering the distribution of flooded sections of the urban traffic network, reveals the pulsed aggregation characteristics of supercharging load caused by the superposition of concentrated travel and charging behaviors of users under intermittent rainfall time courses, and constructs a spatio-temporal transfer matrix model of supercharging load considering the availability of supercharging piles and user charging risk avoidance behaviors driven by rainfall intensity time courses.

[0078] 2. The present invention proposes a method for predicting the spatio-temporal situation of ultra-fast charging load in urban distribution networks based on spatio-temporal graph deep learning. The charging radiation area is divided with ultra-fast charging stations as nodes, and the rainfall intensity and time course data are mapped to the nodes of the charging radiation area. Combining the characteristics of user ultra-fast charging aggregation and the spatial dynamic reconstruction process of traffic flow, a dynamic association model of ultra-fast charging load based on spatio-temporal graph modeling is constructed. The graph network and time-guided embedding algorithm are used to encode the periodic fluctuations and mutation response characteristics of ultra-fast charging load, analyze the temporal characteristics of ultra-fast charging behavior under different weather conditions, and adaptively construct the adjacency matrix of ultra-fast charging stations according to the similarity of behavior characteristics to capture the spatial diffusion effect under the influence of weather, so as to realize the spatio-temporal situation prediction of electric vehicle ultra-fast charging load under heavy rainfall weather. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0080] Figure 1 It is a flowchart of a method for predicting the spatio-temporal situation of ultra-fast charging load in urban distribution networks under heavy rainfall weather provided by the present invention;

[0081] Figure 2 It is a schematic diagram of the spatio-temporal distribution result of electric vehicle ultra-fast charging load provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0082] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0083] The present invention provides a method for predicting the spatio-temporal situation of ultra-fast charging load in urban distribution networks under heavy rainfall weather, as Figure 1 shown, including the following steps:

[0084] S1. Based on the water accumulation evolution process under heavy rainfall scenarios, considering surface runoff and local drainage capacity, dynamically simulate the relationship between the ground water depth in the area where the ultra-fast charging pile is located and the rainfall intensity and duration per unit time, establish a water level response threshold model for the sensor at the bottom of the ultra-fast charging pile, and use the trigger height of the sensor as the boundary. Combining the trigger delay, a control signal or power-off protection is issued when the water level threshold is exceeded, and then an electrical performance failure probability model under the change of ground water depth, rainfall intensity and time course is constructed;

[0085] S2. Analyze the dynamic transfer characteristics of the user charging selection space considering the distribution of flooded sections in the urban traffic network, obtain the pulsed aggregation characteristics of the supercharging load caused by the superposition of concentrated user travel and charging behaviors under intermittent rainfall time courses, and construct a spatio-temporal transfer matrix model of the supercharging load considering the availability of supercharging piles and user charging risk avoidance behaviors driven by the rainfall intensity time course;

[0086] S3. Divide the charging radiation area with the supercharging station as the node, map the rainfall intensity and time course data to the nodes of the charging radiation area, combine the user electrical performance failure probability model and the spatio-temporal transfer matrix model of the supercharging load to obtain the operating state characteristics of the supercharging piles, rainfall intensity characteristics, and traffic flow characteristics, construct a dynamic correlation model of the supercharging load based on spatio-temporal graph modeling, and input the current operating state characteristics of the supercharging piles, rainfall intensity characteristics, and traffic flow characteristics into the dynamic correlation model of the supercharging load to obtain the corresponding load prediction values.

[0087] To further implement the above technical solutions, a supercharging station electrical performance failure probability model is constructed based on the influence of water level and flooding time on the working state of supercharging piles under heavy rainfall. This model describes the charging pile failure probability considering the influence of the charging pile immersion failure rate and the change of insulation performance with the environment. The specific content of the electrical performance failure probability model includes:

[0088] ,

[0089] In the formula, is the electrical performance failure probability of the supercharging pile, is the failure rate of the supercharging pile at time is the time step; is the influence function of insulation deterioration on the performance failure of the supercharging pile after flooding;

[0090] The working state of the supercharging pile is affected by the water level and flooding time. The rising water level will directly affect the insulation performance of the supercharging pile body part. If the water accumulation depth exceeds a certain threshold, the supercharging pile may encounter electrical faults such as short circuits, leakage, or equipment damage. The flooding time is closely related to the aging process of electrical equipment. As time goes by, the insulation performance of the supercharging pile will gradually decrease, resulting in an increased risk of failure. When water accumulates in the charging pile heat dissipation pipeline, the conductivity of the coolant increases, triggering a short circuit risk. Assuming that the water level height of the same supercharging pile is consistent, the waterproof performance of the supercharging pile is quantified, and a waterproof efficiency coefficient is established considering the IP protection level of the supercharging pile to calculate the failure rate of the supercharging pile in the flooded state.

[0091] The specific content of the supercharging pile failure rate includes:

[0092] ,

[0093] In the formula, is the attenuation coefficient of the electrical performance of the supercharger pile after being flooded; is the damping coefficient; is the water depth detected by the sensor; is the waterproof efficiency coefficient of the supercharger pile, where ; is the n benchmark waterproof threshold when the protection level of the supercharger pile in the supercharger station is the benchmark waterproof level. In this embodiment, the benchmark waterproof level is set to IPV67; is the immersion time coefficient; is the corrected cumulative immersion time detected by the sensor; is the anti-water inlet time threshold of the supercharger pile;

[0094] The aging process of the insulating material affects the performance of the supercharger pile. In a humid environment, the insulation resistance of the material will decay exponentially due to water absorption, and the decay rate is accelerated by water pressure triggering, which further affects the failure rate of the supercharger pile in the flooded state. Considering the aging process of the material in a humid and hot environment, an insulation failure model is established. The influence function of insulation deterioration on the performance failure of the supercharger pile after being flooded The specific content of is as follows:

[0095] ,

[0096] ,

[0097] In the formula, is the failure sensitivity coefficient; is the insulation resistance; is the safety threshold resistance; is the initial insulation resistance; is the water absorption aging coefficient; is the conductivity of water; is the water pressure coupling acceleration coefficient, indicating the insulation degradation rate after seal failure; is the step function, is the real-time water pressure borne by the seal structure, which is determined by the water depth in the area where the supercharger pile is located; is the water pressure reference value.

[0098] To further implement the above technical solution, the specific content of obtaining the water depth in the area where the supercharger pile is located includes:

[0099] The heavy rainfall process is evolved into a piecewise function with alternating rainfall periods and intermittent periods:

[0100] ,

[0101] In the formula, is the rainfall intensity sequence, are respectively thek The rainfall intensity, start time, and duration of the ith rainfall; For the k ith rainfall and the k (i + 1)th rainfall, the intermittent time between them;

[0102] The water depth in the area where the supercharger pile is located is jointly determined by rainfall input, drainage output, and neighborhood overflow. During the time interval, rainfall pauses but the drainage system continues to work. A continuous rainfall intensity - drainage balance model is constructed:

[0103] ,

[0104] In the formula, is the water depth in the area of the supercharger pile at time t t; is a piece - wise indicator function. If t t is in the rainfall period [t_j, t_{j + 1}], it takes 1, otherwise it takes 0, which is used to simulate rainfall intermittency; is the drainage rate, assumed to be proportional to the water level, f is the drainage capacity coefficient; is the set of geographical areas adjacent to the supercharger station, y refers to any surrounding area in N_j, y and x is the overflow transfer coefficient of the surrounding area N_j to the current area y N_i, determined by the slope between the supercharger pile area and the surrounding area; t is the water depth in the surrounding area

[0105] Rainfall input is obtained in the form of a piece - wise function, and the water depth evolution process is affected by time - course intermittency perturbations. Based on the continuous rainfall intensity - drainage balance model, discretization is performed to establish a time - step piece - wise discretization model of water depth under intermittent heavy rainfall:

[0106] ,

[0107] In the formula, is the water depth in the area of the supercharger pile at the discretized time t t_k, is the water depth in the surrounding area y N_j at the discretized time t t_k;

[0108] The electrical failure probability of the supercharger pile is positively correlated with the water accumulation depth and the soaking time. The drainage system works continuously during the time interval, resulting in a decrease in the water level. When the water accumulation depth in the area where the charging pile is located exceeds the waterproof threshold, the failure rate of the charging pile rises rapidly. Affected by circuit delay and noise, there is a dynamic deviation between the detection value of the actually deployed water level sensor and the true water accumulation depth. It is necessary to correct the cumulative time based on the photoelectric water level detection value. Sensor detection value and the sensor detection corrected cumulative soaking time are as follows:

[0109] ,

[0110] ,

[0111] In the formula, is the sensor response delay period; i is an intermediate variable, i = t - t 1 + 1, t 1 is the detection time threshold for the supercharger pile in the soaked state; is the measurement noise, following a Gaussian distribution ; is the cumulative soaking time; is i the indicator function of the water accumulation depth in the area of the supercharger pile at time . When it meets , the value is 1. If it does not meet, the value is 0;

[0112] To further implement the above technical solution, the specific content of the supercharging load spatio-temporal transfer matrix model includes:

[0113] ,

[0114] ,

[0115] In the formula, is the selection probability of users in area for the supercharging station n ; is the set of users in area . u represents any user in; is the probability of users using supercharging; User u selects the probability of charging at the supercharging station n ; is the probability that user u selects the current path at the current moment, ODReference path; is the matrix model for spatio-temporal transfer of supercharging load, is the area R The probability that any user in selects a supercharging station at each time n is R is all of the set.

[0116] To further implement the above technical solution, obtaining the probability that a user u selects the current path at the current moment specifically includes:

[0117] Using the traffic network topology to reflect the dynamic changes of the flooded area, it is set that the passability in the road network is related to the water depth of the road section. When the water depth of the road section exceeds the set critical threshold, the road section is regarded as impassable in the user's path selection process. At this time, the traffic capacity of the road section is judged to be zero, the path impedance tends to infinity, and the user re-plans the path. Then the passability model of urban roads under heavy rainfall is considered as:

[0118] ,

[0119] ,

[0120] In the formula, is the path impedance of path rs ; represents the length of path rs ; is the passing speed of path rs ; is the water depth of the road section; is the conventional speed when the vehicle is driving; is the traffic capacity coefficient, which changes dynamically with the rainfall intensity, road water depth and traffic state, and reflects the travel efficiency; z is the traffic sensitivity factor; is the water depth reference attenuation parameter;

[0121] The user's path selection is not a one-time decision, but a dynamic adjustment process. Since closed road sections are not considered in path search, when facing roads that may be flooded or closed, the user will immediately activate the path re-planning mechanism. Users often adjust their travel paths according to the principle of minimum impedance, that is, select a path with the least resistance and the highest efficiency.

[0122] Then the probability that user u selects the current path at the current moment is:

[0123] ,

[0124] ,

[0125] wherein, is the risk perception factor of the user for the path rs , and are the travel choices of the user affected by weather conditions and historical experience respectively, p is the path rs total number of.

[0126] To further implement the above technical solution, the specific content of obtaining the probability of the user using the supercharger is as follows:

[0127] When the user travels and the battery level is lower than the threshold, a charging demand is generated, and the overall rainfall time series is solved into a series of rainfall segments and intermittent segments, and all rainfall intermittent periods are extracted ,wherein, and are the start time and duration of the k th rainfall respectively. Based on the start and end times, spatial positions and durations of each rainfall intermittent segment, the kernel density estimation method is used to calculate the charging load spatial aggregation factor:

[0128] ,

[0129] ,

[0130] wherein, represents the impact of intermittent rainfall on the concentrated travel and charging of users. The longer the intermittency, the more inclined the user is to concentrated travel; q is the influence parameter; represents t the aggregation intensity of user charging in the area where the supercharger n is located at the moment; is the number of traveling users in the area where the supercharger n is located during the intermittent period; is used to control the intensity of the pulsating influence; represents the pulsating intensity of the charging event during the intermittent period after the k th rainfall, and are the mean and standard deviation of the charging event interval respectively;

[0131] Considering the influence of intermittent rainfall, the user dynamically judges whether to preferentially choose to use the supercharger pile for charging and energy replenishment according to the duration of the current rainfall intermittent period:

[0132] ,

[0133] Wherein, is the influence factor of rainfall duration on user selection; is the maximum reference value during the rainfall intermission period; is the state of charge of the user at time t; is the minimum state of charge value that the user needs to charge at time t; is a piecewise indicator function, which takes 1 if the condition of is satisfied, otherwise it takes 0.

[0134] To further implement the above technical solution, obtaining the probability that a user u chooses to charge at a supercharger station n specifically includes: The specific content is as follows:

[0135] Calculating the user charging probability based on the characteristics of user concentrated travel and charging and the availability of supercharger piles under intermittent rainfall duration:

[0136] ,

[0137] ]>Wherein, The probability that the user u chooses to charge at a supercharger station n ; represents t the agglomeration intensity of user charging in the area where the supercharger station n is located at time For the user u to the supercharger station n the shortest accessible path length, which is infinite when the road is flooded; N is the total number of supercharger stations.

[0138] To further implement the above technical solution, the specific content of S3 includes:

[0139] Taking N supercharger stations as graph nodes to divide the charging radiation area of the urban distribution network, defining the node set as , discretizing time into a set of non-overlapping intervals , defining the graph structure at time t as , represents the set of features of all nodes in the time period , E represents the set of connection edges between supercharger station nodes used to represent the spatial proximity between regions. When the radiation areas of supercharger station i and supercharger station j nodes are adjacent, the value of the connection edge is defined as 1, otherwise it is 0;

[0140] Extract the basic state features of each ultra-fast charging station node to form a node state vector, and the feature vector of each node s n,0 includes the operating state features of the ultra-fast charging piles , rainfall intensity features and traffic flow features , and use Sigmoid to construct a node adaptive dynamic adjacency matrix:

[0141] ,

[0142] ,

[0143] where is the adjacency weight, jointly determined by node feature similarity and spatial proximity; and are learnable parameters that control the weights of feature differences and spatial proximity; is the static adjacency relationship between ultra-fast charging station i and ultra-fast charging station j ;

[0144] Extract high-order spatio-temporal correlation features through a multi-layer graph network. For each ultra-fast charging station n extract the spatial correlation between it and adjacent nodes, and use the permutation invariant principle to aggregate and update the features of all its adjacent ultra-fast charging station nodes:

[0145] ,

[0146] ,

[0147] where is the aggregated feature of the adjacent nodes of the ultra-fast charging station node at the l th layer n ; is the set of adjacent ultra-fast charging stations of ultra-fast charging station n , is any node in; is the feature vector of node n at the l th layer; and are the input and output matrix orders; is the feature vector of the node l- at the first layer ; is the feature vector of the ultra-fast charging station l at the n th layer; is the feature transformation matrix of node n at the l th layer;L is the number of layers of the graph neural network; all trainable parameters in each layer are shared on each node;

[0148] The time-guided embedding extracts hidden stationary features by learning the conditional distribution of the time context of the prediction target, and then captures the potential patterns of the user's ultra-fast charging behavior on time scales such as daily cycles and weekly cycles. First, the graph neural network is used to extract the temporal feature information of the nodes, and then these feature information are concatenated with the time-guided embedding vector. At t the +1 moment, the TGE embedding vector realizes the perception of the time evolution trend by fusing information such as the rainfall intensity, water accumulation depth, user's ultra-fast charging demand, traffic status, and station function status at the current time, thereby providing temporal consistency support for the downstream graph neural network.

[0149] Obtain t the TGE embedding vector at the +1 moment:

[0150] ,

[0151] In the formula, and are the output of the first and second hidden layers; and are the weight matrices of each hidden layer; and are the bias vectors of each output layer; is the time context embedding vector; is the current state variable, representing the temporal information of time t +1;

[0152] Multiply the node feature by the output layer weight matrix , perform a non-linear transformation through the activation function, multiply the result by the final prediction weight vector , and obtain the final predicted ultra-fast charging load through the activation function again . After passing through l layers of the network, obtain t the load prediction value of node n at the +1 moment:

[0153] ,

[0154] ,

[0155] In the formula, is the output feature of the last layer of the graph network, Contact is the process of vector concatenation, is the L th layer n ultra-fast charging station is a supercharging station n at t the predicted supercharging load value at the +1 moment; is the final predicted weight vector; is the output layer weight matrix.

[0156] The present invention will be further described below through simulation experiments:

[0157] To verify the prediction accuracy of the spatio-temporal distribution of the supercharging load of the proposed model under heavy rainfall conditions, 25 supercharging stations in a typical urban area are selected, and the load prediction results of the traditional LSTM time series prediction model, the static graph model under normal weather, and the graph network time-guided embedding TGE model incorporating the impact of heavy rainfall constructed by the present invention are compared in an intermittent heavy rainfall scenario. The historical load data of the past 30 days and the corresponding rainfall and waterlogging conditions are selected as the training set to predict the spatio-temporal distribution of the supercharging load, and the evaluation indicators include mean square error, mean absolute percentage error, and spatial distribution consistency index.

[0158] The spatio-temporal distribution results of the supercharging load of electric vehicles under heavy rainfall are as Figure 2 shown. The supercharging load of electric vehicles shows obvious spatial agglomeration characteristics. Due to waterlogging, the traffic on some roads is restricted, causing vehicles to detour to areas with higher accessibility, forming a superposition of local congestion and charging demand. Secondly, the intermittent rainfall makes the charging behavior of users tend to be in the modes of avoiding rain, concentrating, and dispersing. A large number of vehicles flock to the available supercharging stations in a short time, exacerbating the load fluctuation. In addition, due to intermittent heavy rainfall, some supercharging stations are out of service in some areas, further restricting the available charging resources and driving the load to gather at a few stable nodes.

[0159]

[0160] The comparison of the prediction results under different methods is shown in Table 1. Under continuous heavy rainfall conditions, the traditional model has obvious lag and prediction deviation at the load mutation point, and the ability to identify the spatial aggregation effect is insufficient. However, the model of the present invention significantly improves the sensitivity to load fluctuations and the ability to capture the spatio-temporal aggregation trend by introducing the rainfall-traffic-load coupling mechanism. It is superior to the comparative model in all three indicators, demonstrating good practicability and adaptability.

[0161] The above embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A spatio-temporal situation prediction method for the ultra-charge load of urban distribution networks under heavy rainfall weather, characterized in that, Including the following steps: S1. Based on the water accumulation evolution process under heavy rainfall scenarios, considering surface runoff and local drainage capacity, dynamically simulate the relationship between the water depth on the ground in the area where the ultra-fast charging pile is located and the rainfall intensity and duration per unit time, establish a response threshold model for the water level of the sensor at the bottom of the ultra-fast charging pile. Taking the trigger height of the sensor as the boundary, combined with the trigger delay, send a control signal or cut off the power for protection when the water level threshold is exceeded, and then construct a failure probability model of electrical performance under the variation of ground water depth, rainfall intensity and time course; S2. Analyze the dynamic transfer characteristics of the user charging selection space considering the distribution of waterlogged sections in the urban traffic network, obtain the pulsed aggregation characteristics of the ultra-fast charging load caused by the superposition of concentrated user travel and charging behaviors during intermittent rainfall time courses, and construct a spatio-temporal transfer matrix model of the ultra-fast charging load considering the availability of ultra-fast charging piles and user charging risk avoidance behaviors driven by the rainfall intensity time course; S3. Divide the charging radiation area with the ultra-fast charging station as the node, map the rainfall intensity and time course data to the nodes of the charging radiation area, combine the user electrical performance failure probability model and the spatio-temporal transfer matrix model of the ultra-fast charging load to obtain the operating state characteristics of the ultra-fast charging pile, rainfall intensity characteristics and traffic flow characteristics, construct a dynamic correlation model of the ultra-fast charging load based on spatio-temporal graph modeling, and obtain the corresponding load prediction value by inputting the current operating state characteristics of the ultra-fast charging pile, rainfall intensity characteristics and traffic flow characteristics into the dynamic correlation model of the ultra-fast charging load.

2. A method for predicting the spatio-temporal situation of the ultra-charging load of urban distribution networks under heavy rainfall weather according to claim 1, characterized in that The specific content of the electrical performance failure probability model includes: ， Wherein, is the electrical performance failure probability of the ultra-fast charging pile, is the failure rate of the ultra-fast charging pile at time is the time step; is the influence function of insulation deterioration on the performance failure of the ultra-fast charging pile after being flooded; Among them, the failure rate of the ultra-fast charging pile The specific content includes: ， Wherein, is the attenuation coefficient of the electrical performance of the supercharger pile after being flooded; is the damping coefficient; is the water depth detected by the sensor; is the waterproof efficiency coefficient of the supercharger pile; is the supercharger station n the reference waterproof threshold when the protection level of the supercharger pile in it is the reference waterproof level; is the immersion time coefficient; is the corrected cumulative immersion time detected by the sensor; is the anti-inflow time threshold of the supercharger pile; Influence Function of Insulation Deterioration on Performance Failure of Ultra-fast Charging Piles after Flooding Related to the insulation resistance, the specific content includes: ， ， Wherein, is the failure sensitivity coefficient; is the insulation resistance; is the safety threshold resistance; is the initial insulation resistance; is the water absorption aging coefficient; is the conductivity of water; is the water pressure coupling acceleration coefficient, indicating the insulation degradation rate after seal failure; is the step function, is the real-time water pressure borne by the sealing structure, which is determined by the water depth in the area where the overcharging pile is located; is the water pressure reference value.

3. A method for predicting the spatio-temporal situation of the ultra-charging load of an urban distribution network under heavy rainfall weather according to claim 2, characterized in that, The specific content of obtaining the water depth in the area where the ultra-fast charging pile is located includes: Evolve the heavy rainfall process into a piecewise function alternating between rainfall periods and intermittent periods: ， In the formula, is the rainfall intensity sequence, are respectively the rainfall intensity, start time and duration of the k -th rainfall; is the interval time between the k -th rainfall and the k +1-th rainfall; The water depth in the area where the ultra-fast charging pile is located is jointly determined by rainfall input, drainage output and neighborhood overflow. During the time course intermittent period, rainfall pauses but the drainage system continues to work, and a continuous rainfall intensity-drainage balance model is constructed: ， Wherein, is the water depth of the supercharging pile area at time t ; is a piecewise indicator function, which takes 1 if t is during the rainfall period and 0 otherwise, used to simulate the intermittency of rainfall; is the drainage rate, assumed to be proportional to the water level, f is the drainage capacity coefficient; is the set of geographical areas adjacent to the supercharging station, y which refers to any surrounding area within ; y is the overflow transmission coefficient of the surrounding area x to the current area, determined by the slope between the supercharging pile area and the surrounding area; is the water depth of the surrounding area y at time t ; Rainfall input is obtained in the form of a piecewise function and the water accumulation evolution process is affected by time course intermittency disturbances. Based on the continuous rainfall intensity-drainage balance model, discretization processing is carried out to establish a time step segmented discretization model of water depth under intermittent heavy rainfall: ， In the formula, is the accumulated water depth in the fast charging pile area at the discretized t time, is the accumulated water depth in the surrounding area at the discretized y at t time; Affected by circuit delay and noise, the detected value of the actually deployed sensor and the corrected cumulative immersion time detected by the sensor are as follows: ， ， In the formula, is the sensor response delay period; i is an intermediate variable, i = t - t 1 + 1, t where 1 is the detection time threshold for the water immersion state of the supercharger pile; is the measurement noise, which follows a Gaussian distribution ; is the cumulative water immersion time; is i the indicator function of the water depth in the supercharger pile area at time , which takes the value of 1 when the condition is met, and 0 if not; is the sensor trigger threshold.

4. A method for predicting the spatio-temporal situation of the ultra-charging load of urban distribution networks under heavy rainfall weather according to claim 1, characterized in that, The specific content of the spatio-temporal transfer matrix model of the ultra-fast charging load includes: ， ， Wherein, is the selection probability of the supercharger station by the users in the area n ; is the set of users in the area ; u represents any one user in ; is the probability of a user using a supercharger; u The probability that the user n chooses to charge at the supercharger station ; u is the probability that the user OD chooses the current path at the current moment, where represents the path; is the spatio-temporal transfer matrix model of the supercharger load, is the probability that any user in the area R chooses the supercharger station n at each time, R is the set of all .

5. A method for predicting the spatio-temporal situation of the ultra-fast charging load of an urban distribution network under heavy rainfall weather according to claim 4, characterized in that, Obtain user u Probability of selecting the current path at the current moment The specific content includes: Use the traffic network topology structure to reflect the dynamic changes of the waterlogged area. Set the passability in the road network to be related to the water depth of the road section. When the water depth of the road section exceeds the set critical threshold, the road section is regarded as impassable in the user's path selection process. At this time, the traffic capacity of the road section is judged to be zero, the path impedance tends to infinity, and the user re-plans the path. Then the passability model of urban roads under heavy rainfall is: ， ， Wherein, is the path rs impedance; represents the path rs length; is the passing speed of the path rs ; is the accumulated water depth of the road section; is the conventional speed when the vehicle is running; is the passing capacity coefficient, which changes dynamically with the rainfall intensity, road accumulated water depth and traffic state, and reflects the travel efficiency; z is the passing sensitivity factor; is the accumulated water depth reference attenuation parameter; Then the user u The probability of selecting the current path at the current moment is as follows: ， ， In the formula, is the risk perception factor of the user for the path rs , and are the travel choices of the user affected by weather conditions and historical experience respectively, p is the path rs total number.

6. A method for predicting the spatio-temporal situation of the ultra-fast charging load of urban distribution networks under heavy rainfall weather, as claimed in claim 4, wherein Obtain the probability of users using ultra-fast charging The specific content includes: When the user travels and the battery level is lower than the threshold, a charging demand is generated, and the overall rainfall time series is solved as a series of rainfall segments and intermittent segments, and all rainfall intermittent segments are extracted , where and are respectively the start time and duration of the k th rainfall. Based on the start and end times, spatial positions, and durations of each rainfall intermittent segment, the kernel density estimation method is used to calculate the spatial agglomeration factor of the charging load: ， ， In the formula, represents the impact of intermittent rainfall on the concentrated travel and charging of users. The longer the intermittency, the more inclined users are to travel concentratedly; q is the impact parameter; represents t the agglomeration intensity of user charging in the area where the supercharging station is located at time n ; is the number of traveling users in the area where the supercharging station is located during the intermittent period n ; is used to control the intensity of the pulsed impact; represents the pulsed intensity of the charging event during the intermittent period after the k th rainfall, and are the mean and standard deviation of the charging event intervals respectively; Considering the influence of intermittent rainfall, users dynamically judge whether to prefer to use ultra-fast charging piles for charging and energy replenishment according to the duration of the current rainfall intermittent period: ， In the formula, is the influence factor of rainfall duration on the user's choice; is the maximum reference value during the rainfall intermission period; is the state of charge of the user at time t; is the minimum value of the state of charge that the user needs to charge at time t; is a piecewise indicator function, which takes 1 if the condition is satisfied, otherwise it takes 0.

7. A method for predicting the spatio-temporal situation of the ultra-charge load of urban distribution networks under heavy rainfall weather according to claim 4, characterized in that Obtain user u Select the probability of charging n at the ultra-fast charging station The specific content includes: Calculate the user charging probability based on the concentrated travel and charging characteristics of users and the availability of ultra-fast charging piles under intermittent rainfall time courses: ， In the formula, user u the probability of choosing to charge at a n supercharger station; represents t the agglomeration intensity of user charging in the area where the n supercharger station is located at time is the user u to the supercharger station n the length of the shortest accessible path, which is infinite when the road is flooded; N is the total number of supercharger stations.

8. A method for predicting the spatio-temporal situation of the ultra-charge load of urban distribution networks under heavy rainfall weather according to claim 1, characterized in that The specific content of S3 includes: Taking N supercharging stations as graph nodes to divide the charging radiation area of the urban distribution network, the node set is defined as , discretize time into a set of non-overlapping intervals , define the graph structure at time t as , represents the set of features of all nodes in the time period , E represents the set of connection edges between the supercharging station nodes, which is used to represent the spatial proximity between regions. When the radiation areas of the supercharging stations i and the supercharging station j are adjacent, define the value of the connection edge as 1, otherwise 0; Extract the basic state features of each ultra-fast charging station node to form a node state vector, and the feature vector of each node s n,0 includes the operating state features of the ultra-fast charging pile , rainfall intensity features and traffic flow features , and use Sigmoid to construct a node adaptive dynamic adjacency matrix: ， ， wherein, is the adjacency weight, jointly determined by the node feature similarity and the spatial proximity; and are learnable parameters, controlling the weights of the feature difference and the spatial proximity; is the supercharging station i and the supercharging station j is the static adjacency relationship therebetween; Extract high-order spatio-temporal correlation features through a multi-layer graph network, for each supercharger station n Extract its spatial correlation with adjacent nodes, and use the principle of permutation invariance to aggregate and update the features of all its adjacent supercharger station nodes: ， ， In the formula, is the l layer supercharging station node n adjacent node aggregation feature; is the set of adjacent supercharging stations of the supercharging station n , is any one node in; is the node n the l layer adjacent feature transformation matrix; and are the input and output matrix orders; is the l- 1st layer node feature vector; is the l layer supercharging station n feature vector; is the node n the l layer node feature transformation matrix; L is the number of layers of the graph neural network; all trainable parameters in each layer are shared on each node. Obtain t TGE embedding vector at time +1: ， Wherein, and are the output of the first and second hidden layers; and are the weight matrices of each hidden layer; and are the bias vectors of each output layer; is the time context embedding vector; is the current state variable, representing the timing information at time t +1; Multiply the node features by the output layer weight matrix and perform a non - linear transformation through the activation function. Multiply the result by the final prediction weight vector and obtain the final predicted overcharge load by passing through the activation function again . After passing through l layers of the network, obtain t the load prediction value of the node at the n +1 moment: ， ， In the formula, is the output feature of the last layer of the graph network, and Contact is the process of vector concatenation. is the L th layer of the ultra-fast charging station n 's feature vector; is the predicted ultra-fast charging load value of the ultra-fast charging station n at t +1 moment; is the final predicted weight vector; is the weight matrix of the output layer.

Images