Electric vehicle charging station intelligent recommendation method and system based on power quality space-time optimization

Abstract

The application discloses a kind of based on electric energy quality space-time optimization electric vehicle charging station intelligent recommendation method and system, the recommendation method steps include: real-time acquisition of each area power grid operation data, charging station and electric vehicle state, road network operation data and traffic state data and individualized data;Extract the electric energy quality state parameter of each area, respectively calculate real-time electric energy quality factor;Real-time electric energy quality factor is input into electric energy quality prediction model, and the electric energy quality factor prediction value of each area is obtained;According to the current position of user, search local candidate subgraph along driving path, in the search process, according to electric energy quality factor prediction value, the weight of edge is adjusted;Fusion graph structure is constructed, according to electric energy quality factor prediction value, traffic cost and user preference data, select the optimal charging station as recommendation result.The application can improve electric energy quality, consider real-time traffic state, reduce the traffic cost of user, improve user experience simultaneously.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system regulation, in particular to an electric vehicle charging station intelligent recommendation method and system based on power quality space-time optimization. BACKGROUND

[0002] With the continuous increase in the number of current electric vehicles, the corresponding charging demand also increases sharply, thus higher requirements are put forward for the operation stability and power quality of the power grid. The intelligent recommendation of the charging station is to determine the optimal charging station according to the user's demand, the state of the charging station, etc., to provide personalized charging station recommendation for the user, so as to improve the user experience, optimize the power grid load, and at the same time improve the operation efficiency of the charging station.

[0003] For the intelligent recommendation of the charging station, the existing technology usually simply determines the recommendation scheme based on the user's travel demand and the idle state of the charging station, that is, the charging station with close distance and idle state in the user's travel path and the recommended path is selected, but this kind of method does not fully consider the influence of charging load on the power quality of the power grid, and a large number of electric vehicles concentrated charging may cause harmonic distortion, voltage fluctuation and power grid load imbalance and other problems, so the traditional charging station recommendation based on the user's travel demand and the idle state of the charging station will actually affect the service life of the power grid equipment and the electric vehicle battery. At the same time, the traffic state of the road network is constantly changing, and there may be serious congestion or temporary traffic control on some sections, making it inconvenient to travel, while some sections may have high traffic efficiency. Simply selecting the charging station with the closest distance according to the user's travel path may actually have problems such as inconvenient traffic, higher travel cost, and the need for longer waiting time, and if the traffic state is considered alone for charging station recommendation, the power quality problem cannot be considered. In addition, different electric vehicle users have different preferences, for example, some users may prefer charging speed to charging station distance, while some users may prefer charging station distance to charging speed, and the traditional intelligent recommendation scheme of the charging station cannot take into account the influence on the power quality of the power grid, the real-time state of the road network and the user's individual charging demand.

[0004] Therefore, an electric vehicle charging station intelligent recommendation method that can take into account the power quality optimization, real-time state of the road network and user's personalized charging demand is needed to improve the power quality and user experience, and at the same time reduce the user's travel cost. SUMMARY

[0005] The purpose of the present application is to solve the technical problems existing in the prior art, and provide an electric vehicle charging station intelligent recommendation method and system based on power quality space-time optimization, which can predict the power quality factor of the future time of the transformer area in real time, search for a local candidate subgraph in the road network module combined with the predicted power quality factor, and then screen out the optimal charging station combined with the power quality factor, the travel cost and the personalized charging demand data of the user, so as to improve the power quality, take into account the real-time traffic state of the road network and the personalized preference charging demand of the user, reduce the travel cost of the user, improve the user experience, and ensure the stable operation of the power grid and the charging network.

[0006] In order to achieve this purpose, the present application discloses an electric vehicle charging station intelligent recommendation method based on power quality space-time optimization, which adopts the following steps:

[0007] Real-time collection of power grid operation data of each transformer area in a specified area, charging station state information in each transformer area, electric vehicle state information, road network operation data and traffic state data, and current electric vehicle user personalized charging demand data;

[0008] According to the real-time collected power grid operation data, the power quality state parameters of each transformer area are extracted, including the harmonic distortion rate parameter, the voltage flicker parameter, the voltage fluctuation parameter and the voltage transient event frequency parameter, and the real-time power quality factor of each transformer area is calculated according to the power quality state parameters of each transformer area;

[0009] The real-time power quality factor of each transformer area is input into the pre-trained power quality prediction model to obtain the power quality factor prediction value of each transformer area, and the power quality prediction model is established by pre-training the ARMA model (Auto-Regression and Moving Average Model, ARMA model, self-regression moving average model) using the historical data of the power quality factor of each transformer area;

[0010] According to the current position and travel path of the current electric vehicle user, a local candidate subgraph is searched in the road network graph model, the road network graph model is constructed based on the graph structure according to the road network operation data of each transformer area, all road intersection positions, charging station positions and specified important road section start and end points are taken as nodes, and road sections are taken as edges, the weight of each edge is configured according to the length of the road section and the current traffic state data, and in the process of searching the local candidate subgraph, the weight of each edge is adjusted according to the power quality factor prediction value of the transformer area where the charging station is located;

[0011] According to the road network graph model and the feature vectors of the nodes, a fusion graph structure is constructed, the feature vectors of the nodes include power quality factor prediction values, charging load rates, passing costs and historical user preference charging demand data, according to the power quality factor prediction values of the area where the charging station is located, the passing cost from the current position of the electric vehicle to the charging station and the personalized preference charging demand data of the current electric vehicle user, the optimal charging station is selected from the local candidate sub-graph based on the fusion graph structure as the recommended result output.

[0012] As a further improvement of the application, the power grid operation data of the area includes area voltage , area current , the charging station state information includes idle state of the charging pile , load level of the charging pile , the electric vehicle state information includes vehicle type, real-time power, real-time position coordinates, driving path and charging power size and charging strategy of different electric vehicles, the personalized charging demand data includes preference for charging speed, preference for charging cost acceptance range, preference for charging station distance, the road network operation data includes location information of all road intersections within a specified area, location information of charging stations and passable road information, the traffic state data includes traffic flow, road passing speed, road congestion degree, signal lamp scheduling information and traffic event information affecting road passing capacity, the signal lamp scheduling information includes cycle and remaining time of each intersection red light, and the traffic event information includes traffic accident, construction closure and temporary traffic control event information.

[0013] As a further improvement of the application, the harmonic distortion rate parameter is a harmonic total distortion rate factor calculated according to voltage harmonic total distortion rate and current harmonic total distortion rate , for representing the comprehensive distortion state of voltage harmonic and current harmonic, the calculation expression of the harmonic total distortion rate factor is:

[0014] ,

[0015] ,

[0016] ,

[0017] wherein, represents harmonic order, represents voltage amplitude of the th harmonic, represents fundamental voltage amplitude, The adjustment index is set according to the harmonic order. Indicates the amplitude of the fundamental current;

[0018] The voltage flicker parameter is based on long-term flicker values. and short-time flicker value Calculated flicker factor The flicker factor is used to characterize the degree of flicker in a voltage signal. The calculation expression is:

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] in, Different time constants The flicker visual perception weighting coefficient is below. Indicates the number of sampling time points; Indicates the first The voltage fluctuation sensing value at each sampling time. It is based on the time constant The coefficients are set to be used for further adjustment of different time constants. The weighted effect under the following conditions Represents time constant Quantity, It is a statistical short-time flicker value. The number of Indicates the first A short-time flash value , This represents the maximum tolerable threshold for flicker. Indicates the value of long-term flickering;

[0024] The voltage fluctuation parameter is a voltage fluctuation factor calculated based on the voltage fluctuation amplitude. The voltage fluctuation factor The calculation expression is:

[0025]

[0026]

[0027]

[0028]

[0029] wherein, represents the average voltage fluctuation value, represents the time window width, , respectively represent the effective value of the voltage signal , , is the normalized voltage fluctuation amplitude, is the mean value calculated from the samples selected from the historical average voltage fluctuation data according to the normalized voltage fluctuation amplitude , is the maximum value of the voltage fluctuation amplitude, , respectively represent the mean value and the standard deviation of the historical average voltage fluctuation data, is the period of the voltage signal;

[0030] The voltage transient event frequency parameter is a sag / swell factor calculated according to the statistical voltage sag event frequency and the voltage swell event frequency , for representing the frequency and depth of the voltage transient event, the voltage sag event frequency and the voltage swell event frequency are obtained by using adaptive threshold value, wherein when the voltage amplitude is lower than the dynamic threshold value and the duration exceeds the specified time window, it is determined as a voltage sag event, when the voltage amplitude is higher than the dynamic threshold value and the duration exceeds the specified time window, it is determined as a voltage swell event, , is a preset rated voltage amplitude threshold value, is the harmonic distortion rate and the weighted average value of the voltage amplitude influence factor and the load level of the transformer area, i.e. , is the load rate of the current transformer area, , , respectively are weight coefficients, is the voltage fluctuation reference value;

[0031] The calculation expression of the sag / swell factor is:

[0032] ,

[0033] wherein, is the sag event frequency, is the swell event frequency, , respectively are the depth weight of the voltage sag event and the voltage swell event, represents the maximum value of .

[0034] As a further improvement of the present application, the calculation expression of the real-time power quality factor of each substation is:

[0035] ,

[0036] wherein, is the real-time power quality factor of the i-th substation, is the harmonic distortion rate parameter of the i-th substation, is the voltage flicker parameter of the i-th substation, is the voltage fluctuation parameter of the i-th substation, is the voltage transient event frequency parameter of the i-th substation. is the harmonic distortion rate parameter of the i-th substation, is the voltage flicker parameter of the i-th substation, is the voltage fluctuation parameter of the i-th substation, is the voltage transient event frequency parameter of the i-th substation. is the voltage transient event frequency parameter of the i-th substation.

[0037] As a further improvement of the present application, the power quality prediction model is established by pre-training an ARMA model using the historical data of the power quality factor of each substation, comprising:

[0038] normalizing the historical data of the power quality factor of each substation and arranging them in time sequence to form a power quality factor time series;

[0039] dividing the power quality factor time series of each substation into multiple subsequences using a sliding window;

[0040] using an ARMA model as the prediction model to establish an independent power quality prediction model for each substation, wherein is the autoregressive order, is the moving average order. The determination of the model order uses ACF (autocorrelation function) and PACF (partial autocorrelation function) analysis. The ACF and PACF graphs of the sequence are calculated for preliminary identification: the ACF is used to measure the linear correlation between the time series and the corresponding own lag version to preliminarily determine the order of the moving average component ; the PACF is used to evaluate the partial correlation between the sequence and a certain lag after controlling the influence of the intermediate lag, which helps to preliminarily identify the order of the autoregressive component .

[0041] inputting each subsequence of each substation into the power quality prediction model ​​​​​The model is trained, and the error between the predicted value and the actual value is calculated during the training process, and the model is adjusted according to the error result The order of the model Or reselect the model until the model reaches the preset accuracy requirement, complete the training of the model, and obtain the trained power quality prediction model.

[0042] As a further improvement of the application, the local candidate subgraph is searched in the road network graph model according to the current position and driving path of the electric vehicle user, comprising:

[0043] The weight of each edge in the road network graph model is set according to the following formula:

[0044]

[0045] Wherein, Indicates the node The weight of the edge corresponding to the road section between the node At time , indicates the road section length between the node And the node , indicates the average travel speed of the road section between the node And the node At time , indicates the road congestion index of the road section between the node And the node At time , indicates the signal waiting time of the end intersection of the road section between the node And the node At time , indicates the burst event factor of the road section between the node And the node , , Adjust the weight; The current position of the current electric vehicle user is taken as the starting point, and all nodes and corresponding paths within the maximum acceptable travel time threshold in the target direction in the road network graph model are searched to form a local candidate subgraph; During the search of the local candidate subgraph, the weight of the edge corresponding to the current road section is adjusted according to the power quality factor prediction value of the transformer area according to the following formula:

[0046]

[0047]

[0048]

[0049] ​​​​​​wherein, representing a node and a node between which the corresponding edge of the road section is adjusted at the weight at the moment, representing a node and a node between which the power quality factor prediction value of the area to which the road section belongs is adjusted at the moment, representing a preset linkage coefficient between the power grid and the road network.

[0050] As a further improvement of the present application, the fusion graph structure constructed according to the road network graph model and the feature vector of each node comprises:

[0051] defining a fusion graph structure based on the road network graph model , representing a node in the road network graph model, representing an edge in the road network graph model, representing an edge weight matrix formed by the weights of each edge in the road network graph model, representing a node feature matrix composed of the node feature vectors of all nodes at the moment Each node in the fusion graph structure represents a charging station, and the node feature vector corresponding to each node contains multiple fusion features:

[0052]

[0053] wherein, representing the power quality factor prediction value of the area to which the candidate charging station belongs at the moment , representing the charging load rate of the candidate charging station at the moment , representing the travel cost value of the current electric vehicle user from the current location to the candidate charging station , , representing the path of the electric vehicle user from the current location to the candidate charging station , representing the weight of the corresponding edge of the road section between the node and the node at the moment , representing the electricity price of the area to which the candidate charging station belongs at the moment , , , , , ​​The candidate charging station The recommendation score of the moment is calculated by the expression:

[0054]

[0055] Wherein, represents the node The weight of the edge corresponding to the road section between the node represents the maximum value of the passing cost, represents the preference score of the first candidate charging station obtained according to the personalized charging demand data of the current electric vehicle user, is a weight coefficient.

[0056] As a further improvement of the present application, the fusion graph structure constructed according to the road network graph model and the feature vector of each node further comprises:

[0057] The node feature matrix is taken as the initial input matrix, that is, , and the time-normalized adjacency matrix is constructed according to the edge weight matrix ;

[0058] The first layer information transmission process of the fusion graph structure is defined as:

[0059]

[0060] In the formula, is the normalized adjacency matrix , is the node embedding representation of the first layer, is the trainable weight matrix of the current first layer, is an activation function;

[0061] When the graph attention mechanism is used to propagate information on the edge, different weights are assigned to different neighbors, wherein the attention weight represents the relative importance of the node receiving information from the neighbor node in the first layer, and is defined as:

[0062]

[0063] In the formula, is a learnable attention vector; represents a vector concatenation operation; is the node ​​The set of neighboring nodes, , Represents a node , The input feature vector, Representing neighboring nodes Its characteristics.

[0064] As a further improvement of the present invention, the step of selecting the optimal charging station from the local candidate subgraph based on the fusion graph structure as the recommended result output includes:

[0065] A Markov decision process is established based on the fusion graph structure, where the current state... Based on the user's current location and path direction The set of embedding vectors of candidate charging station nodes, and the action space. Let represent the set of selectable charging stations, and let the policy function be . , indicating the state Select action The probability, The parameters of the policy network are set with the objective function of maximizing the expected cumulative reward from the current moment into the future. :

[0066]

[0067] in, A discount factor representing future rewards; For the system in the first The immediate reward function obtained after a recommendation decision. In the policy function Cumulative reward for all possible trajectories Expected value;

[0068] The instant reward function is defined as follows:

[0069]

[0070] in, These are the weighting coefficients. for Moment Action The corresponding predicted power quality factor value, for Moment Action The corresponding charging load rate, for Moment Action The corresponding toll cost value, for Moment Action The corresponding preference rating.

[0071] The application also discloses an electric vehicle charging station intelligent recommendation system based on power quality space-time optimization, comprising a processor and a memory, the memory is used for storing a computer program, and the processor is used for executing the computer program to execute the method as described above.

[0072] Compared with the prior art, the application has the advantages that: the application calculates the power quality state parameters of each area in real time according to the power grid operation data, and then calculates the power quality factors of each area to evaluate the power quality state of the area, and further uses the power quality prediction model constructed based on the ARMA model to predict the power quality factors of each area at future time, so that the change trend of the power quality state of the area can be predicted, when it is necessary to recommend a charging station for the current electric vehicle user, the local candidate subgraph is searched in the real-time road network graph model according to the current position and the driving path of the user, and the weight of the edge corresponding to each road section is adjusted in combination with the power quality factors of each area during the searching process, so that the recommendation efficiency can be improved, the searching time can be reduced, and the charging station in the area with better power quality can be preferentially selected, finally, the optimal charging station is selected in combination with the predicted value of the power quality factor, the passing cost from the current position of the current electric vehicle to the candidate charging station and the individualized charging demand data of the current electric vehicle user, so that the power quality optimization, the passing cost of the user, the real-time road network traffic state and the individualized charging demand of the user can be considered, the power grid load can be balanced, the real-time traffic state can be considered, the passing cost of the user can be reduced, the user experience can be improved, and the stable operation of the power grid and the charging network can be ensured. BRIEF DESCRIPTION OF DRAWINGS

[0073] In the following, the application will be described in more detail based on embodiments and with reference to the drawings. In which:

[0074] Figure 1 is a flowchart of the implementation of the electric vehicle charging station intelligent recommendation method based on power quality space-time optimization in the embodiments of the application. DETAILED DESCRIPTION

[0075] The application will be further described in detail below with reference to the drawings and specific embodiments, but the protection scope of the application is not limited by this.

[0076] The application firstly calculates the electric energy quality state parameters of each area, and obtains the electric energy quality factor for evaluating and predicting the electric energy quality level of the area, and the prediction model constructed based on the ARMA model can judge the change trend of the electric energy quality at the future time in advance, and when recommending the charging station for the electric vehicle user, the system searches the local candidate subgraph in the real-time road network map according to the current position and the driving path of the user, and dynamically adjusts the road segment weight in the searching process in combination with the electric energy quality factor of each area, so as to improve the recommendation efficiency, shorten the searching time, and preferentially consider the charging station in the area with better electric energy quality. Finally, the optimal charging station is selected from the candidate subgraph for recommendation by comprehensively considering the electric energy quality prediction value, the user's access cost to the candidate station and the individualized charging demand, so that the electric energy quality optimization, the user's access cost, the real-time road network traffic state and the individualized charging demand of the user are considered, the electric network load is balanced, the real-time traffic state is considered, the user's access cost is reduced, the user experience is improved, and the stable operation of the electric network and the charging network is ensured.

[0077] In one aspect, the application can use the real-time evaluation of the electric energy quality state of the area by introducing the harmonic distortion, voltage fluctuation, flicker and voltage transient event depth in the process of intelligent recommendation of the charging station, adjusting the weight of the edge in the road network graph model according to the electric energy quality state in the process of searching the candidate charging station, fully combining the real-time electric energy quality state to adjust the search path of the charging station, not only improving the search efficiency, but also enabling the user to avoid the high load or low electric energy quality area in real time, and preferentially selecting the charging station in the area with better electric energy quality, avoiding the electric energy quality problems such as harmonic distortion, voltage fluctuation and unbalanced load of the electric network caused by the charging behavior of the electric vehicle, and further improving the electric energy quality of the electric network, reducing the influence on the service life of the electric network equipment and the electric vehicle battery, and ensuring the stability of the electric network operation.

[0078] On the other hand, on the basis of searching the road network graph model in combination with the prediction value of the electric energy quality factor, the optimal charging station is selected by further combining the prediction value of the electric energy quality factor, the access cost and the charging demand data, a traffic-energy deep integration regulation mode is formed, the real-time traffic state and the intelligent deep integration of the energy system are fully utilized to dynamically determine the recommendation result, the real-time electric energy quality state, the road network traffic state and the individualized charging demand of the user are comprehensively considered, the three-in-one collaborative recommendation optimization of “path selection-power quality-intelligent guidance” is realized, so as to improve the flexibility and stability of the power distribution network, and improve the charging efficiency and user experience of the electric vehicle user.

[0079] Figure 1The detailed process of the intelligent recommendation method for the electric vehicle charging station based on the power quality space-time optimization of the embodiment of the application is shown, and the steps include:

[0080] Step S01. Real-time collection of power grid operation data of each substation in the specified area, charging station state information in each substation, electric vehicle state information, road network operation data and traffic state data, and current individual charging demand data of electric vehicle users.

[0081] Specifically, the power grid operation data of the substation includes substation voltage , substation current , etc., the charging station state information includes the idle state of the charging pile , the load level of the charging pile , etc., the electric vehicle state information includes the vehicle type, real-time power, real-time position coordinates, driving path, and charging power size of different electric vehicles, charging control strategy, etc., and the individual charging demand data of the user includes the preference for charging speed, the preference for the acceptable range of charging cost, the preference for the distance to the charging station, etc.

[0082] For example, for the charging station state information, the working state (idle / busy rate, represented by for idle, for busy) and the current charging power of each charging pile in the charging station can be obtained in real time, and then the overall load level of the charging station is calculated according to the following formula :

[0083] (1)

[0084] wherein represents the output power of the th charging pile in the charging station, is the total design capacity of the charging station, is the number of charging piles in the charging station.

[0085] For the electric vehicle state information, the current position coordinates, driving path, real-time power, etc. of the electric vehicle are obtained in real time during the driving of the electric vehicle user, and the charging power of different types of electric vehicles and the charging control strategy of each brand are recorded, etc., which are stored in the user database. When intelligent recommendation is needed, the real-time state information of the electric vehicle user can be obtained from the user database.

[0086] For the personalized charging demand data of the user, a questionnaire or a preference option can be provided to the user through the charging APP program, and the user selects his / her preference for the charging speed (such as fast charging priority, ordinary charging is acceptable, etc.), the acceptance range preference for the charging cost (such as low price priority, charging speed priority regardless of the price, etc.), and the distance preference for the charging station (such as priority consideration of the charging station with short distance or no requirement for the distance, etc.). After obtaining the selection results submitted by the user, the selection results are stored in the user database, and when intelligent recommendation is needed, the preference charging demand data of the user can be obtained from the user database.

[0087] For the road network operation data and the traffic state data, the traffic open platform, the navigation system API, the city Internet of Things terminal, and the intelligent traffic control system can be used for acquisition. Specifically, the road network operation data includes the position information of all road intersections in a specified area, the position information of the charging station, and the passable road information, etc., and the traffic state data includes the traffic flow, the road passing speed, the road congestion degree, the signal lamp scheduling information, and the traffic event information affecting the road passing capacity, etc., wherein the road passing speed is the average speed of a vehicle on a certain road in a unit time, which is used to reflect the passing efficiency of the vehicle; the road congestion degree can be calculated using the ratio of the actual traffic density to the road design capacity, and when congested, the value is close to 1, which means that the passing capacity is reduced; the signal lamp scheduling information includes the cycle and the remaining time of each intersection, which is used to evaluate the waiting delay of the vehicle at the intersection; the traffic event information includes the abnormal event information such as traffic accidents, construction closure, and temporary traffic control, which will affect the passing efficiency of the road.

[0088] Step S02. According to the real-time collected power grid operation data, the power quality state parameters of each area are extracted, including the harmonic distortion rate parameter, the voltage flicker parameter, the voltage fluctuation parameter, and the voltage transient event frequency parameter, and the power quality factor of each area is calculated according to the power quality state parameters of each area.

[0089] In this embodiment, the harmonic distortion rate parameter, the voltage flicker parameter, the voltage fluctuation parameter, and the voltage transient event frequency parameter, etc. are extracted according to the real-time collected power grid operation data, so that the power quality state of each area can be evaluated in real time.

[0090] In this embodiment, the harmonic distortion rate parameter is the harmonic total distortion rate factor calculated according to the voltage harmonic total distortion rate and the current harmonic total distortion rate , which is used to represent the comprehensive distortion state of the voltage harmonic and the current harmonic.

[0091] Specifically, the harmonic total distortion rate factor can be calculated according to the following process:​

[0092] Real-time collected transformer substation voltage Current in the transformer area Perform Fast Fourier Transform (FFT) on each signal to convert the time-domain signal to the frequency-domain signal. Assume the transformer substation voltage signal... The frequency domain representation is obtained after FFT. Current signal The frequency domain representation is obtained after FFT. Based on the frequency domain signal analysis, the content of each harmonic can be determined. The total voltage harmonic distortion rate can be calculated using the following formula:

[0093] (2)

[0094] in, Indicates the harmonic order. Indicates the first The voltage amplitude of the second harmonic Indicates the amplitude of the fundamental voltage. This is the adjustment index related to the harmonic order.

[0095] Calculate the total harmonic distortion of current. The expression can be represented as:

[0096] (3)

[0097] in, Indicates the harmonic order. Indicates the first The voltage amplitude of the second harmonic Indicates the amplitude of the fundamental current. This is the adjustment index related to the harmonic order.

[0098] Total harmonic distortion of voltage and total harmonic distortion of current Obtain the total harmonic distortion factor :

[0099] (4)

[0100] As shown in equation (4) above, the total harmonic distortion factor Capable of integrating total voltage harmonic distortion and total harmonic distortion of current The influence of these factors is used to comprehensively reflect the quality status of the total harmonic distortion of voltage and current in the power grid data.

[0101] In this embodiment, the voltage flicker parameter is based on the long-term flicker value. and short-time flicker value Calculated flicker factor , for characterizing the flicker degree of the voltage signal. Specifically, the flicker factor The calculation process can be as follows:

[0102] The envelope detection is performed on the real-time collected voltage signal of the transformer area to obtain the amplitude change rate of the voltage fluctuation , which is input into the following perception function to calculate the voltage fluctuation perception value :

[0103] (5)

[0104] wherein, is the voltage fluctuation reference value, which is used to standardize the perception intensity.

[0105] The flicker visual sensitivity weighting coefficients under different time constants are calculated using the sliding window integration method:

[0106] (6)

[0107] wherein, represents the number of sampling time points; represents the voltage fluctuation perception value of the i-th sampling time point. Finally, the short-time flicker value is calculated according to the following formula:

[0108]

[0109] (7)

[0110] wherein, is the flicker visual sensitivity weighting coefficient under different time constants , which is used to reflect the sensitivity of the human eye to the light flicker caused by the voltage fluctuation under different time scales, for example, the human eye is more sensitive to fast flicker, and therefore the voltage fluctuation that changes rapidly in a short time corresponds to a larger coefficient related to the time constant , which is used to further adjust the weighting effect under different time constants, represents the number of time constants .

[0111] In this embodiment, the short-time flicker value is calculated according to the formula (5)~(7), and the coefficient can adjust the flicker visual sensitivity weighting coefficient under different time scales in the short-time flicker value ​​​​The contribution degree in the value is calculated, thereby improving the calculation accuracy and reliability of the short-time flicker value .

[0112] The long-time flicker value is further calculated according to the plurality of short-time flicker values . The long-time flicker value is calculated according to the plurality of short-time flicker values .The calculation formula is as follows:

[0113] (8)

[0114] wherein, is the number of the statistical short-time flicker values , that is, the number of the values calculated according to a certain time interval in a specified time period, represents the i-th short-time flicker value .

[0115] Finally, the flicker factor is obtained by comprehensively considering the short-time flicker value and the long-time flicker value :

[0116] (9)

[0117] wherein, is the maximum tolerable threshold of the flicker, for example, 4.0.

[0118] According to the above formula (9), the value of the flicker factor can be mapped to the range of (0, 1], if the sum of the short-time flicker value and the long-time flicker value of a certain area reaches or exceeds the maximum tolerable threshold of the flicker , corresponding to the worst power quality, it is indicated that there is a serious voltage flicker problem in the area, which obviously affects the lighting quality and user comfort; when the sum of the short-time flicker value and the long-time flicker value is less than the maximum tolerable threshold of the flicker , the flicker factor , corresponding to the best power quality, thereby the flicker factor can be used to effectively evaluate the flicker quality state of the power.

[0119] In the embodiment, the voltage fluctuation parameter is a voltage fluctuation factor calculated according to the voltage fluctuation amplitude, and the voltage fluctuation factor can be calculated by the following process:

[0120] ​​The effective voltage value refers to the DC voltage value that generates the same amount of heat as the AC voltage across the same resistor within one cycle. This applies to the real-time acquired transformer substation voltage. Signal Calculate its effective value for:

[0121] (10)

[0122] in, Let be the period of the voltage signal. In practical calculations, the effective value of the voltage within the window can be approximated by applying a sliding window processing technique to the voltage time series data. .

[0123] Then calculate the average voltage fluctuation value. :

[0124] (11)

[0125] Among them, the adjustment coefficient The width of the time window. , Representing voltage signals respectively , The effective value. By setting a reasonable time window, the voltage variation within a certain period can be accurately reflected. The average voltage fluctuation value is calculated according to the formula above. It can reflect the relativity of voltage fluctuations.

[0126] Then normalize the voltage fluctuation amplitude according to the following formula:

[0127] (12)

[0128] in, , The mean and standard deviation of historical average voltage fluctuation data are used to standardize the fluctuation amplitude.

[0129] This embodiment, based on normalization, considers the normalized voltage fluctuation amplitude. The mean value is obtained by selecting samples from historical average voltage fluctuation data, removing outliers, and recalculating. To further improve calculation accuracy. For example, it is possible to eliminate those that satisfy the following conditions. The extreme fluctuation values ​​were sampled, and only samples within the normal range were retained for recalculating the mean. .

[0130] Finally, the voltage fluctuation factor was calculated. :

[0131] (13)

[0132] wherein, is the maximum value of the voltage fluctuation amplitude. The value of can be mapped to the range of (0, 1] according to equation (13).

[0133] The voltage fluctuation factor is calculated according to the above equations (10)~(13) in this embodiment, the average voltage fluctuation value is calculated first. The physical quantification of the degree of voltage disturbance can be provided, and then the normalized voltage fluctuation amplitude is used to eliminate outliers, and the scale difference can be eliminated based on the normalization operation, so that the fluctuation amplitudes of different areas or different time periods have unified comparability, so that the final voltage fluctuation factor can stably reflect the power quality state of the area.

[0134] In this embodiment, the voltage transient event frequency parameter is the sag / surge factor calculated according to the statistical voltage sag event frequency and voltage surge event frequency, which is used to represent the frequency and depth of voltage transient events, and the voltage sag event frequency and voltage surge event frequency are obtained by adaptive threshold statistics. When the voltage amplitude is lower than the dynamic threshold and the duration exceeds the specified time window, it is determined as a voltage sag event, and the sag depth is recorded at this time.

[0135] When the voltage amplitude is higher than the dynamic threshold and the duration exceeds the specified time window, it is determined as a voltage surge event, and the surge amplitude is recorded at this time.

[0136] (14)

[0137] wherein, is the weighted average of the harmonic distortion rate, the voltage amplitude influence factor and the load level of the area, which comprehensively reflects the power quality disturbance risk level, is the preset rated voltage amplitude threshold. The calculation formula is as follows:

[0138] (15)

[0139] In the formula, is the load rate of the current area, , , are weight coefficients respectively. By setting appropriate weight coefficients, the identification threshold of voltage transient event can be dynamically adjusted under different operating scenarios, so that the system is more sensitive to high-risk areas, and the accuracy and intelligence of event detection are improved.

[0140] This embodiment combines the operating status of the transformer area, such as load level and harmonic distortion rate, and introduces an adaptive threshold model to dynamically adjust the judgment conditions for voltage sags and swells. This can fully consider the dynamic harmonic distortion and load level of the transformer area, thereby improving the recognition accuracy of voltage sag and swell events.

[0141] In a specific application embodiment, the transient decrease / increase factor The calculation expression can be represented as:

[0142] (16)

[0143] in, To temporarily reduce the frequency of incidents, To temporarily increase the frequency of incidents, , For the depth weights of voltage sag events and voltage swell events, express The maximum value. For example, suppose The typical range is 0 to 1000, so we take 1000. According to formula (16), we can... The value is mapped to the range (0,1].

[0144] After extracting the aforementioned power quality status parameters, this embodiment further calculates the power quality factor for different transformer substations by integrating these parameters. It can comprehensively consider the impact of factors such as harmonics, flicker, voltage sags and swells on power quality, and accurately reflect the power quality status of the distribution area.

[0145] In specific application examples, the power quality factor of each transformer area can be calculated using the following formula:

[0146] (17)

[0147] in, The weights for the harmonic distortion rate parameter, The weights for voltage flicker parameters, The weights for voltage fluctuation parameters, The weights for the frequency parameter of voltage transient events. For the first Harmonic distortion rate parameters for each transformer area For the first Voltage flicker parameters for each transformer area For the first Voltage fluctuation parameters for each distribution area For the first Frequency parameters of voltage transient events in each transformer substation.

[0148] Step S03. The power quality factors of each area are respectively input into the pre-trained power quality prediction model to predict the power quality factors of each area at future time, and the power quality prediction model is established by pre-training ARMA model using the historical data of power quality factors of each area.

[0149] In this embodiment, the power quality prediction model is pre-trained based on ARMA model, and the steps include:

[0150] Step S301. The historical data of the power quality factor of each area is normalized and arranged in time sequence to form a power quality factor time series.

[0151] Specifically, the voltage, current, harmonic, flicker and other data collected in step S02 can be classified according to the area number to generate independent time series data sets:

[0152] (18)

[0153] Wherein is the total number of areas, is the total number of time points.

[0154] The power quality factor of each area is independently preprocessed: removing missing values, outliers (such as ). The data is mapped to the interval [0, 1]:

[0155] (19)

[0156] Step S302. The power quality factor time series of each area is divided into multiple subsequences by using a sliding window.

[0157] Specifically, the normalized data of each area is arranged in time sequence to form an independent time series, wherein the power quality factor sequence of each area is . The time series of each area is divided into multiple subsequences by using a sliding window technique.

[0158] For example, the window length (such as 24 hours of data points), the step (such as 1 hour) is divided, and each window data is: , which is used to predict the power quality factor at the next time.

[0159] Step S303. Use the ARMA (Autoregressive Moving Average) model as the prediction model, and determine the order of the ARMA model based on the autocorrelation function ACF and the partial autocorrelation function PACF. ,in Let the order be the autoregressive order. To determine the moving average order, an independent moving average is established for each substation. The model is used to predict the power quality factor of the transformer area at future times.

[0160] Specifically, model order The determination of the moving average component is achieved using ACF (autocorrelation function) and PACF (partial autocorrelation function) analysis. Preliminary identification is performed by calculating the ACF and PACF plots of the series. The ACF measures the linear correlation between the time series and its corresponding lagged version, thus providing an initial assessment of the order of the moving average component. PACF is used to assess the partial correlation between a sequence and a specific lag term after controlling for the effects of intermediate lags, and it helps to initially identify the order of the autoregressive component. .

[0161] Specifically, establish an independent [system / mechanism] for each [station / area]. The model can be represented as:

[0162] (20)

[0163] in For the first Each district The power quality factor at any given time. is a constant term representing the intercept of the model. The autoregressive coefficient represents the past... Each time point The effect of the value on the current value. The moving average coefficient represents the past... The impact of the error term at each time point on the current value. For time points The error term (white noise), assuming .

[0164] Step S304. Input each sub-sequence of each transformer area into the ARMA model for training. During the training process, calculate the error between the predicted value and the actual value, and adjust the model according to the error results. Model order Alternatively, a new model can be selected until the model reaches the preset accuracy requirements, thus completing the model training and obtaining a well-trained power quality prediction model.

[0165] Specifically, the mean squared error (MSE) can be used to calculate the error between the predicted and actual values. The calculation expression is as follows:

[0166] (twenty one)

[0167] in For the first Each district Predicted power quality factor at time [time]. This represents the number of samples.

[0168] Then, based on the error analysis results, the order of the ARMA model is adjusted. Alternatively, a different model can be selected to improve prediction accuracy.

[0169] After the model is trained, the trained ARMA model is used to predict the future of each transformer area. Power quality factor at time:

[0170] (twenty two)

[0171] in, For the first Time point of each station area The predicted value of the power quality factor. The autoregressive coefficient represents the past... Each time point The effect of the value on the current value. The moving average coefficient represents the past... The impact of the error term at each time point on the current value. For time points The error term follows the order of distributed.

[0172] This embodiment predicts the power quality factor of each transformer area at future times. Based on the predicted power quality factor It can analyze the trend of power quality changes in the future and determine whether power quality will decline or rise and the magnitude of the change.

[0173] Step S04. Search for local candidate subgraphs in the road network graph model based on the current location and driving route of the electric vehicle user. The road network graph model is constructed based on the road network operation data of each substation area, with all road intersection locations, charging station locations, and the start and end points of designated important road segments as nodes and road segments as connecting nodes as edges. The weight of each edge is configured according to the length of the road segment and the current traffic status data. During the search for local candidate subgraphs, the weight of each edge is adjusted according to the real-time power quality factor prediction value of the substation where the charging station is located.

[0174] The embodiment first constructs a road network graph model based on electronic map data of a target area The graph model takes all road intersections, charging station locations, and important road segment start and end points as nodes in the graph Each node corresponds to a spatial position with a unique geographic coordinate to reflect a set of passable spatial points. Road segments are abstracted as a set of directed edges Each edge represents that a vehicle can travel from node to node along an actual road, and the directionality of the edge follows road travel rules such as one-way street settings, traffic light guide sequences, and fork priority, etc. Each edge is also attached with a dynamic attribute weight , which represents the travel cost of the road segment at the current time, that is, the path cost. The above road network graph model can be represented as .

[0175] In the embodiment, the real-time running state of the roads in the target area is continuously collected and updated, and the road network graph model is synchronously updated. For example, real-time road network running data and traffic state data are collected and standardized and time-synchronized to form a traffic state matrix , which is bound to the road network graph model , so that the attributes of each road edge can be dynamically adjusted in real time and over time, and the road network graph structure has dynamic perceptibility. Through charging station search based on the above constructed road network graph model, dynamic perception of traffic factors and path guidance can be achieved.

[0176] In the embodiment, the road network graph model supports continuous weighting of edge attributes, and the weight of each edge in the graph is calculated based on road segment length, traffic state data (real-time vehicle speed, congestion degree, signal light delay, construction closure state, etc.), so that the constructed road network graph model can comprehensively describe the traffic accessibility and impedance characteristics of the road network in the current area at the current time. Using the road network graph model , the shortest path and travel time between an electric vehicle user and a charging station can be estimated, and it also serves as the basis layer for subsequent construction of a “road network-power distribution network-charging network” integrated graph to realize adjacency propagation calculation of a subsequent graph neural network and determination of a recommendation strategy.

[0177] In a specific application embodiment, after the road network graph model is constructed and real-time traffic state data is collected, the weight of each edge is set according to the following formula to accurately describe the time, resistance, or energy consumption required for a vehicle to travel from one node to another node:

[0178] (23)

[0179] wherein, denotes the node and the edge weight of the road segment between the node and the node at the time point denotes the node and the length of the road segment between the node and the node at the time point denotes the node and the average travel speed of the road segment between the node and the node at the time point denotes the node and the road congestion index of the road segment between the node and the node at the time point denotes the node and the signal waiting time of the end intersection of the road segment between the node and the node at the time point denotes the node and the event factor of the road segment between the node and the node , , , is the adjustment weight of the corresponding factor, to support on-demand flexible adjustment of the influence degree of different factors on the edge weight.

[0180] By setting each edge weight value in the above manner, the embodiment assigns each weight to the edge attribute set in the road network graph model , and combines the subsequent edge update mechanism of the graph neural network to realize real-time feedback regulation of road condition changes on the charging recommendation result, which can effectively improve the response ability of the charging recommendation system to the actual road traffic environment, make the path planning and recommendation result more accurate and reasonable, and is particularly suitable for intelligent navigation and load guidance in dynamic environments such as morning and evening peak hours, special events or disaster weather.

[0181] To further improve the system calculation efficiency and recommendation response speed, the embodiment further adopts a user travel path clipping mechanism to generate a local candidate subgraph within the user's reachable range in real time on the basis of constructing the road network graph model as the global road network graph structure. Specifically, the user's current location Starting from this point, and combining the user's input of the target direction or navigation intent, the path planning module uses a graph model... Search for all accessible nodes within its drivable range. Based on the preset maximum acceptable passage time threshold Only charging station nodes that meet the following conditions will be retained:

[0182] (twenty four)

[0183] In the formula, This indicates the route from the user's current location to the candidate charging station, taking into account the current traffic conditions. The estimated travel time can be obtained by summing the edge weights of all edges in the path:

[0184] (25)

[0185] Finally, all charging station nodes that meet the time threshold requirements, along with their paths, are combined to form a local candidate subgraph for the current user request. Using this graph as the input region for graph neural network processing can significantly reduce computational complexity and avoid global traversal on complex graphs.

[0186] Furthermore, this embodiment searches for local candidate subgraphs. During the process, the corresponding edge weights are dynamically adjusted based on the predicted power quality factor of the transformer area, so that the route recommendation process is not only affected by traffic conditions, but also takes into account the operation quality of the distribution system of the transformer area along the route. This realizes the positive guidance and coordinated control of electric vehicle charging behavior on the power grid operation status, forming a linkage control mechanism between the road network and the distribution network.

[0187] Specifically, during the search process, if the power quality of the area traversed by the path is poor, i.e., the predicted power quality factor... The value is lower than the set safety threshold If so, then the edge weights corresponding to the road segments surrounding that area will be adjusted as a penalty. For example, suppose a certain road The connected area belongs to the transformer substation. The weights of its edges can be adjusted according to the following formula:

[0188] (26)

[0189] In the formula, Represents a node With nodes The corresponding edge of the road segment between them The weights are adjusted over time. express Time Node the node belongs to the power quality factor prediction value of the area where the road section between the node is located, the preset linkage coefficient between the power grid and the road network, the larger the value, the more sensitive to the power quality.

[0190] As shown in formula (26), when the power quality factor of a certain area approaches 1 (i.e. very good), the travel cost is almost unchanged; and when approaches 0 (i.e. very poor power quality), the cost of the travel path will be significantly increased, so that the recommendation system automatically avoids the area, achieving the goal of "avoiding weak power areas and guiding balanced load distribution".

[0191] Through the above regulation mechanism, the embodiment can effectively guide the charging behavior to the area with stronger power grid carrying capacity and better operation quality without affecting the efficiency of the user's basic travel path, and realize the "path selection-power distribution quality-smart guidance" trinity collaborative recommendation optimization in combination with the regulation mode of traffic-energy deep integration, thereby effectively improving the resilience and stability of the power distribution network.

[0192] Step S05. Construct a fusion graph structure according to the road network graph model and the feature vectors of each node, wherein the feature vectors of the nodes include the power quality factor prediction value, the charging load rate, the travel cost, and the historical user preference charging demand data. According to the power quality factor prediction value of the area where the charging station is located, the travel cost from the location of the electric vehicle to the charging station, and the individualized charging demand data of the electric vehicle user, the optimal charging station is selected from the local candidate subgraph based on the fusion graph structure as the recommended result output.

[0193] In the embodiment, the fusion graph structure is constructed according to the road network graph model and the feature vectors of each node, so as to fuse and express the three types of heterogeneous information of the power distribution network operation state, the charging station business load, and the road network travel condition in a unified graph structure, and construct a dynamic heterogeneous graph supporting state updating and decision optimization.

[0194] Specifically, the fusion graph structure is defined based on the road network graph model , represents a node in the road network graph model, represents an edge in the road network graph model, represents an edge weight matrix formed by the weights of each edge in the road network graph model, represents the node feature matrix composed of the node feature vectors of all nodes at the moment Each node in the fusion graph structure represents a charging station, and the node feature vector corresponding to each node The plurality of fusion features include: , wherein represents a predicted value of the power quality factor of the th candidate charging station at the th time point, and the specific value range is [0, 1], represents the charging load rate of the th candidate charging station at the th time point, represents the electricity price of the th candidate charging station at the th time point, represents a preference score of the th candidate charging station in the th time point according to the individualized charging demand data of the current electric vehicle user, is the recommendation score of the th candidate charging station at the th time point, represents a traffic accessibility index, that is, a travel cost value of the current electric vehicle user from the current location to the th candidate charging station, and can be calculated according to the following formula:

[0195] (27)

[0196] wherein, represents a path from the current location of the current electric vehicle user to the th candidate charging station.

[0197] Further, the node feature matrix is taken as an initial input matrix , and a time-normalized adjacency matrix is constructed according to the edge weight matrix :

[0198] (28)

[0199] The adjacency matrix is used for information propagation and graph convolution calculation of the graph neural network, so that the system can identify the travel impedance, influence range and service capacity difference between different stations.

[0200] In the embodiment, a comprehensive service quality score function is further defined to calculate the recommendation score , which is used to evaluate the overall recommendation value of each charging station at the th time point, and the calculation expression is:

[0201] (29)

[0202] wherein, represents the One candidate charging station Recommended score at any time Represents a node With nodes The weights of the corresponding edges between road segments. This represents the maximum value of the passage cost. This indicates that the data obtained based on the personalized charging needs of current electric vehicle users is used to analyze the first... Preference ratings for each candidate charging station These are the weighting coefficients, and For the weighting coefficients, satisfying .

[0203] In this embodiment, the first step of the fusion graph structure is further defined. The layer information transmission process is as follows:

[0204] (30)

[0205] In the formula, normalized adjacency matrix , For the first Layer node embedding representation, The trainable weight matrix of the current layer, This is the activation function.

[0206] To enhance the model's expressive power under heterogeneous node attributes, this embodiment further employs a graph attention mechanism to assign differentiated weights to different neighbors when propagating information along edges, wherein the attention weights... Indicates the first Layer nodes Receive from neighboring nodes The relative importance of information is defined as:

[0207] (31)

[0208] In the formula, This is a learnable attention vector; This represents a vector concatenation operation; For nodes The set of neighboring nodes, , Represents a node , The input feature vector, nodes It is a node The neighboring nodes, Representing neighboring nodes Its characteristics.

[0209] In this embodiment, a fused graph structure is further used. Based on the node representations of its graph neural network output, a Markov decision process framework is established. By autonomously learning recommendation strategies through interaction with the environment, the adaptability to optimal recommendation paths under complex and multi-constraint conditions can be further improved.

[0210] Specifically, set the current state. Based on the user's current location and path direction The set of embedding vectors of candidate charging station nodes, and the action space. Let represent the set of selectable charging stations, and let the policy function be . , indicating the state Select action The probability, These are the parameters of the policy network. That is, the current state of the system is defined as:

[0211] (32)

[0212] In the formula Indicates the current electric vehicle user Location at any given moment Indicates the user's destination direction (such as going home, heading to the office area or business district, etc.); This represents the candidate charging station node embedding vector, which is the final output of the graph neural network. This represents the set of candidate charging station nodes in the current cropped subgraph. The state space integrates user behavioral intent, graph structure encoding, and node feature information, enabling a comprehensive characterization of the environmental awareness state in the current recommendation scenario.

[0213] The action space is defined as the set of all reachable candidate charging station nodes in the current subgraph:

[0214] (33)

[0215] Furthermore, the recommendation objective function is to maximize the expected cumulative reward from the current moment into the future.

[0216] (34)

[0217] in, This represents a discount factor for future rewards. In the policy function Cumulative reward for all possible trajectories Expected value; For the system in the first The immediate reward function obtained after the recommendation decision can be specifically defined as follows:

[0218] (35)

[0219] in, These are the weighting coefficients. for Reward value at any moment for Moment Action The corresponding predicted power quality factor value, for Moment Action The corresponding charging load rate, for Moment Action The corresponding toll cost value, for Moment Action The corresponding recommendation score. Specifically, it is calculated according to formula (29), that is, the recommended score. On the one hand, it will be embedded into the graph structure as a node feature, participating in the graph attention propagation and information fusion process between adjacent charging stations. On the other hand, it is used to calculate the immediate reward function in the Markov decision process to reflect the immediate comprehensive benefits brought by choosing each charging station, thereby guiding the optimization of reinforcement learning strategies. This can not only improve recommendation accuracy, but also achieve a continuous connection from multi-factor evaluation to graph structure learning and dynamic decision optimization, ensuring the unity of recommendation strategies in terms of data fusion and rationality of behavioral decisions.

[0220] As shown in equation (32) above, the larger the reward function value, the higher the comprehensive value of the recommendation in the current state. The system will continuously improve the probability of selecting high-quality sites through the reinforcement learning strategy optimization process, thereby guiding the strategy network to evolve towards "grid friendly, traffic efficient, and user satisfied", and finally selecting the optimal recommendation scheme that can take into account power quality, real-time traffic status and user charging needs.

[0221] This embodiment is based on the current user state and the optimal policy function after training. The system selects the optimal target station from the set of candidate charging stations in a local candidate subgraph and outputs the recommendation result. The recommendation process is based on the user's current location. Real-time traffic status map Embedded Graph Nodes Together they constitute the current system state Policy network according to Output recommended actions That is, select the target site .

[0222] The embodiment also provides an electric vehicle charging station intelligent recommendation system based on power quality space-time optimization, including a processor and a memory, the memory is used for storing a computer program, and the processor is used for executing the computer program to execute the method as described above.

[0223] By introducing power quality optimization, the real-time power quality state, road network traffic state and user personalized charging demand are comprehensively considered, the trinity of path selection-power distribution quality-intelligent guidance is realized, the charging station intelligent recommendation is realized by considering the power quality optimization, user travel cost, real-time road network traffic state and user personalized charging demand, the stability of power grid operation and user charging experience are significantly improved, and the charging station intelligent recommendation can be applied to the application scenarios of intelligent power grid, electric vehicle charging network and the like.

[0224] Although the present application has been described with reference to the preferred embodiments, various modifications can be made to the application without departing from the scope of the application, and equivalent parts can be substituted for the parts thereof. In particular, the technical features mentioned in each embodiment can be combined in any manner as long as there is no structural conflict. The present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A power quality space-time optimization-based intelligent recommendation method for electric vehicle charging stations, characterized by the steps of The application relates to a charging station recommendation method and device. Real-time acquisition of power grid operation data of each area in a specified area, charging station state information in each area, electric vehicle state information, road network operation data and traffic state data and current electric vehicle user personalized charging demand data; According to the real-time acquisition of the power grid operation data, the power quality state parameters of each area are extracted, the power quality state parameters including a harmonic distortion rate parameter, a voltage flicker parameter, a voltage fluctuation parameter and a voltage transient event frequency parameter, and the real-time power quality factors of each area are calculated according to the power quality state parameters of each area. The real-time power quality factors of each area are input into a pre-trained power quality prediction model to obtain power quality factor prediction values of each area, and the power quality prediction model is established by pre-training an ARMA model using historical power quality factor data of each area. According to the current location and driving path of the electric vehicle user, a local candidate subgraph is searched in a road network graph model, the road network graph model being constructed based on a graph structure according to road network operation data of each area, with all road intersection locations, charging station locations and start and end points of specified important road sections as nodes, and road sections as connecting nodes, the weights of each edge being configured according to the length of the road section and the current traffic state data, and in the process of searching the local candidate subgraph, the weights of each edge are adjusted according to the power quality factor prediction values of the charging station area. A fusion graph structure is constructed according to the road network graph model and the feature vectors of each node, the feature vectors of each node including power quality factor prediction values, charging load rates, passing costs and historical user preferred charging demand data, and according to the power quality factor prediction values of the charging station area, the passing cost from the current location of the electric vehicle to the charging station and the current electric vehicle user's personalized preferred charging demand data, the optimal charging station is selected from the local candidate subgraph as a recommended result output based on the fusion graph structure. The construction of the fusion graph structure according to the road network graph model and the feature vectors of each node includes: The fusion graph structure is defined based on the road network map model. , This represents a node in a road network diagram model. Represents the edges in the road network diagram model. This represents the edge weight matrix, which is formed by the weights of each edge in the road network graph model. express Node feature vectors of all nodes at time 1 The resulting node feature matrix, where each node in the fused graph structure represents a charging station, and the corresponding node feature vector for each node... It contains multiple fusion features; The selection of the optimal charging station from the local candidate subgraph as the recommended result output based on the fusion graph structure includes: establishing a Markov decision process based on the fusion graph structure, wherein a current state is a set of embedding vectors of the candidate charging station nodes , a path direction of the user at the current time , an action space representing a set of selectable charging station targets, and a policy function is a probability of selecting an action in a state , and parameters of the policy network are used to maximize an expected cumulative reward from the current time to the future as a recommendation objective function : where, is a discount factor representing future rewards; is the immediate reward function obtained by the system after recommending a decision at time is the immediate reward function obtained by the system after recommending a decision at time is the expected value of the cumulative reward of all possible trajectories under the policy function . 2.The method of claim 1, wherein, The power grid operation data of the transformer area includes transformer area voltage , transformer area current , the charging station state information includes idle state of charging pile , load level of charging pile , the electric vehicle state information includes vehicle type, real-time power, real-time position coordinates, driving path, and charging power size and charging strategy of different electric vehicles, the personalized charging demand data includes preference for charging speed, preference for charging cost acceptance range, and preference for charging station distance, the road network operation data includes location information of all road intersections within a specified area range, location information of charging stations, and passable road information, the traffic state data includes traffic volume, road traffic speed, road congestion degree, signal lamp scheduling information, and traffic event information affecting road traffic capacity, the signal lamp scheduling information includes cycle and remaining time of each intersection red-green light, and the traffic event information includes traffic accident, construction closure, and temporary traffic control event information. 3.The method of claim 1, wherein, The harmonic distortion rate parameter is a harmonic total distortion rate factor calculated according to the voltage harmonic total distortion rate and the current harmonic total distortion rate The harmonic total distortion rate factor calculated for representing the comprehensive distortion state of the voltage harmonic and the current harmonic, the harmonic total distortion rate factor The calculation expression of the harmonic total distortion rate factor is: ， ， ， wherein, denotes the harmonic order, denotes the voltage amplitude of the harmonic, denotes the fundamental voltage amplitude, is an adjustment index set according to the harmonic order, denotes the fundamental current amplitude, denotes the current amplitude of the harmonic; The voltage flicker parameter is a flicker factor calculated from a long-time flicker value and a short-time flicker value The calculation expression of the flicker factor for characterizing the degree of flicker of the voltage signal is: ​ ， ， ， ， wherein is a flicker perceptibility weighting factor for different time constants , denotes the number of sampling time points; denotes the voltage fluctuation perceptibility value for the th sampling time point, is a coefficient set according to the time constant for further adjusting the weighting effect for different time constants , denotes the number of time constants , is the number of statistical short-time flicker values , denotes the short-time flicker value for the th short-time flicker value , is the maximum tolerable threshold value for flicker, denotes the long-time flicker value; The voltage fluctuation parameter is a voltage fluctuation factor calculated according to a voltage fluctuation amplitude The calculation expression of the voltage fluctuation factor is: ， ， ， ， wherein, denotes the average voltage fluctuation value, denotes the time window width, , denote the effective values of the voltage signals , , is the normalized voltage fluctuation amplitude, is the mean value calculated from the samples selected from the historical average voltage fluctuation data according to the normalized voltage fluctuation amplitude , is the maximum value of the voltage fluctuation amplitude, , are the mean value and the standard deviation of the historical average voltage fluctuation data, respectively, is the period of the voltage signal; The voltage transient event frequency parameter is a sag / swell factor calculated according to the frequency of voltage sag events and the frequency of voltage swell events , for characterizing the frequency and depth of voltage transient events, the frequency of voltage sag events and the frequency of voltage swell events are obtained by adaptive threshold statistics, wherein when the voltage amplitude is lower than the dynamic threshold and the duration exceeds the specified time window, it is determined as a voltage sag event, when the voltage amplitude is higher than the dynamic threshold and the duration exceeds the specified time window, it is determined as a voltage swell event, , is a preset rated voltage amplitude threshold, is the harmonic distortion rate and the weighted average of the voltage amplitude influence factor and the load level of the transformer area, that is , is the load rate of the current transformer area, , , are weight coefficients, respectively, is the voltage fluctuation reference value; The dip / rise factor The calculation expression is: ， wherein, is the frequency of voltage sag events, is the frequency of voltage swell events, , are the depth weightings for voltage sag events and voltage swell events, respectively, denotes the maximum value. 4.The method of claim 1, wherein, The calculation expression of the real-time power quality factor of each area is: ， wherein, is the real-time power quality factor for the th zone, is the weight of the harmonic distortion parameter, is the weight of the voltage flicker parameter, is the weight of the voltage fluctuation parameter, is the weight of the voltage transient event frequency parameter, is the harmonic distortion parameter for the th zone, is the voltage flicker parameter for the th zone, is the voltage fluctuation parameter for the th zone, is the voltage transient event frequency parameter for the th zone. 5.The method of claim 1, wherein, The power quality prediction model is established by pre-training an ARMA model using historical power quality factor data of each area, including: The historical data of the power quality factor of each area is normalized and arranged in time sequence to form a power quality factor time sequence; The power quality factor time sequence of each area is divided into multiple subsequences by using a sliding window. ARMA model is used as the prediction model, and an independent model is established for each area Model, wherein is the autoregressive order, is the moving average order, and the model order The determination of the model order is performed using ACF and PACF analysis, preliminary identification is performed by calculating the ACF and PACF graphs of the sequence, the order of the moving average component is preliminarily judged using the ACF , and the order of the autoregressive component is preliminarily identified using the PACF ; Input each subsequence of each transformer area into... The model is trained, and the error between the predicted and actual values ​​is calculated during the training process. Adjustments are made based on the error results. Model order Alternatively, a new model can be selected until the model reaches the preset accuracy requirements, thus completing the model training and obtaining the trained power quality prediction model. 6.The method of claim 1 or 2 or 3 or 4 or 5, wherein, The search of the local candidate subgraph in the road network graph model according to the current location and driving path of the electric vehicle user includes: The weight of each edge in the road network graph model is set according to the following formula: ， in, Represents a node With nodes The corresponding edge of the road segment between them Weight of time, Represents a node With nodes The length of the road segment between them For nodes With nodes The section between Average traffic speed at any given time For nodes With nodes The section between Road congestion index at any given time. Represents a node With nodes The end point of the road section is at Traffic light waiting time at any time For nodes With nodes Unexpected event factors in the road sections between them , , To adjust the weights; The current location of the current electric vehicle user is taken as a starting point, all nodes and corresponding paths in the target direction within the maximum acceptable passing time threshold in the road network graph model are searched to form a local candidate subgraph. In the process of searching for a local candidate subgraph, the weight of an edge corresponding to a current road segment is adjusted according to a power quality factor prediction value of a transformer area according to the following formula: ， in, Represents a node With nodes The corresponding edge of the road segment between them The weights are adjusted over time. express Time Node With nodes The section between the two roads belongs to the same district. The predicted value of the power quality factor. This represents the preset linkage coefficient between the power grid and the road network. 7.The method of claim 1-5, wherein, A node feature vector corresponding to each node The plurality of fusion features include: ， wherein denotes the predicted value of the power quality factor of the th candidate charging station at the th time point, denotes the charging load rate of the th candidate charging station at the th time point, denotes the travel cost value of the current electric vehicle user from the current location to the th candidate charging station, denotes the path of the electric vehicle user from the current location to the th candidate charging station, denotes the weight of the edge corresponding to the road segment between the node and the node at the th time point, denotes the electricity price of the th candidate charging station at the th time point, denotes the preference score of the th candidate charging station according to the individualized charging demand data of the current electric vehicle user, is the recommendation score of the th candidate charging station at the th time point, and the calculation expression is: wherein, represents the maximum value of the passage cost, is a weight coefficient. 8.The method of claim 7, wherein, The constructing a fusion graph structure according to the road network graph model and the feature vectors of the nodes further includes: The node feature matrix is constructed as the initial input matrix, i.e. and the edge weight matrix is constructed as ; The first The layer information passing process is defined as: wherein is a normalized adjacency matrix , is a first layer node embedding representation, is a trainable weight matrix for a current layer, is an activation function; The graph attention mechanism is used to assign different weights to different neighbors when propagating information on the edges, where the attention weight represents the relative importance of the node in the layer when receiving information from neighbor nodes is defined as: ， wherein is a learnable attention vector; denotes a vector concatenation operation; is a node , , denotes an input feature vector of a node , , denotes a feature of a neighbor node . 9.The method of claim 8, wherein, The instant reward function is defined as: ， wherein, is a weight coefficient, is the time action the corresponding power quality factor prediction value, is the time action the corresponding charging load rate, is the time action the corresponding passage cost value, is the time action the corresponding recommendation score.

10. An intelligent recommendation system for electric vehicle charging stations based on power quality spatiotemporal optimization, comprising a processor and a memory, the memory being configured to store a computer program, characterized in that, The processor is configured to execute the computer program to perform the method of any one of claims 1-9.

Images