Charging load forecasting method based on ADN dynamic reconstruction and active and reactive joint optimization model

Abstract

The charging load prediction method based on ADN dynamic reconstruction and active and reactive joint optimization model includes the following steps: Step 1: Establish an EV unit mileage energy consumption model and the charge state of EV; Step 2: Consider the subjective will of the owner and establish a spatiotemporal distribution prediction model of EV charging load, which is applicable to charging load prediction in different scenarios; Step 3: Consider the comprehensive operating cost and voltage deviation and other indicators to determine the objective function, and establish a mathematical model containing ADN dynamic reconstruction and active and reactive joint optimization; Step 4: Based on the constraint conditions, the second-order cone relaxation and variable product linearization method are used to transform the joint optimization mathematical model in step 3. The present invention improves the prediction accuracy of EV charging load and ensures the safety and economy of the distribution network operation containing large-scale EV charging load.

Description

Technical Field

[0001] The present invention relates to the technical field of EV charging load prediction, and in particular to a charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model. Background Art

[0002] As the global electric vehicle (EV) industry, especially in China, is entering a new stage of accelerated development, although it has been clearly required that more charging facilities be deployed in various areas of cities, electric vehicles (EVs) have the dual attributes of road traffic network load and power network load, and their driving and charging behaviors will be affected alternately by the road network and the power grid. Active distribution networks are an effective solution to the problems of distributed new energy grid-connected operation control, interaction between the power grid and EVs and their charging and discharging facilities, and intelligent power distribution. At present, domestic and foreign scholars have conducted extensive research on the network reconstruction of active distribution networks containing DG and EVs, but few studies have considered the spatiotemporal distribution characteristics of EV loads and the coupling between EVs and transportation systems and distribution networks. Summary of the invention

[0003] The present invention provides a charging load prediction method based on ADN dynamic reconstruction and active and reactive joint optimization model. The method takes into account the coupling of vehicles, roads and networks, establishes an EV time-space transfer model and applies it to charging load prediction in different scenarios. For ADN containing EVs, a mathematical model of ADN dynamic reconstruction and reactive and active joint optimization is established; at the same time, this joint optimization model is converted into a mixed integer second-order cone programming problem, which greatly improves the solution efficiency.

[0004] The technical solution adopted by the present invention is:

[0005] A charging load prediction method based on active distribution network (ADN) dynamic reconstruction and active and reactive joint optimization model includes the following steps:

[0006] Step 1: Establish an EV unit mileage energy consumption model;

[0007] Step 2: Considering the subjective will of the car owner, a spatiotemporal distribution prediction model of EV charging load is established to predict charging load in different scenarios;

[0008] Step 3: Taking into account the comprehensive operating cost and voltage deviation index, determine the objective function and establish a mathematical model including ADN dynamic reconstruction and active and reactive joint optimization;

[0009] Step 4: Based on the constraints, the joint optimization mathematical model in step 3 is transformed using the second-order cone relaxation and variable product linearization method.

[0010] In step 1, based on the speed and flow of transportation capacity, the EV unit mileage energy consumption model considering the actual temperature and transportation conditions and the EV charge state are as follows:

[0011]

[0012] Where: is the speed of the EV at point x on the road (i, j), where i and j are the two sides of the road, represents the traffic volume of road (i, j), Tr i,j represents the transportation capacity, a1, a2, a3 are the road adaptation coefficients, χ t is the road saturation at time t. represents the β-power of road saturation at time t, represents the a3 power of the road saturation at time t;

[0013] The power consumption per kilometer of an EV with air conditioning turned on is calculated as follows:

[0014]

[0015] Where: E t i,j (x) is the power consumption per kilometer of the EV without air conditioning on;

[0016] The actual temperature change can affect the air conditioning activation rate of EV users, which can be expressed as:

[0017]

[0018] Where: K temp is the energy proportionality coefficient. Assume that a random number r with uniform temperature distribution is generated. If r<K pect , then turn on the air conditioner. q1~q4 and c1~c3 are fitting parameters, b represents temperature, K pect Indicates the air conditioning opening rate at different temperatures.

[0019] Combining equations (2) and (3), the actual power consumption per kilometer of EV is expressed as:

[0020]

[0021] represents the energy proportional coefficient at time t, It indicates the power consumption per kilometer of the EV without turning on the air conditioner.

[0022] Therefore, the calculation formula for the charge state of the EV at location x is:

[0023]

[0024] SOC(x) is the calculation formula for the state of charge of the EV at location x. init Indicates the initial charge state of EV driving, E t (x) represents the actual power consumption per kilometer of the EV, C EV Indicates battery capacity, D X Indicates the distance from the starting point to location x.

[0025] In step 2, based on the trip chain, Monte Carlo sampling and Markov decision theory, the EV charging load spatiotemporal distribution prediction model considering the subjective will of the car owner is applied to the charging load prediction in different scenarios, as follows:

[0026] Remaining power and next driving distance D next The smaller the ratio, the stronger the willingness to charge. Using fuzzy theory, according to the user's subjective willingness sat To describe the subjective charging willingness of EV users:

[0027]

[0028] SOC(x) represents the charge state of the EV at location x. EV Indicates battery capacity, D next Indicates the next driving distance, E t (x) represents the actual power consumption per kilometer of the EV.

[0029] Introducing F fuz The fuzzy set representing charging willingness, whose membership function F fuz (uu sat ), expressed by the following formula (7):

[0030]

[0031] In the formula: uu max sat 、uu min sat uu sat The upper and lower limits of uu sat represents the user's subjective intention, and τ represents the probability of charging the EV out of subjective intention.

[0032] EV has two charging modes. Considering the charging cost and battery loss, the default is slow charging. If the parking is over, the slow charging cannot meet the expected state of charge SOC. exp , that is, fast charging is selected when formula (8) is satisfied:

[0033]

[0034] C EV Indicates battery capacity, SOCexp Indicates the desired state of charge;

[0035] Where: P slow is the rated slow charging power of EV, T i p , SOC i init are the parking time and initial state of charge of the EV at parking point i respectively. If the SOC of the EV is lower than SOC thr , it will be charged midway, the charging location is L mid par is a charging station close to the destination, and the charging time is T char The calculation formula is as follows:

[0036]

[0037] D thr It is expressed as EV's initial state of charge SOC init To critical state of charge SOC thr Driving distance, SOC thr Indicates the critical state of charge, is the charging location of the charging station close to the destination, c Indicates that the SOC of the EV drops to SOC thr When the number of charging stations passed, ∑D r represents the distance between the starting point of the EV trip and the rth charging station, J represents the Jth location;

[0038]

[0039] Where: D thr From SOC to EV init To critical state of charge SOC thr The driving distance; ∑D r represents the distance between the starting point of the EV trip and the rth charging station; r c Indicates that the SOC of the EV drops to SOC thr When the number of charging stations passed; f(r c ) represents r c The corresponding traffic network, P fast and P slow is the rated fast charging power and slow charging power. char Indicates charging time, D mid represents the distance from the starting point to the charging station where the electric vehicle will reach when the SOC is about to reach 0, ∑D r represents the distance between the starting point of the EV trip and the rth charging station;

[0040] According to the coupling relationship between traffic nodes and distribution networks, the time and space loads of each distribution node are counted, and the charging load P at distribution node z is z (t) and the total charging load of the distribution network P DN (t) is expressed as:

[0041]

[0042] represents the charging power of the nth electric vehicle at time t, and n represents the number of electric vehicles charged at node z.

[0043]

[0044] P z (t) represents the charging load at the distribution node z;

[0045] Where: N EV and N w are the number of charging nodes EV and charging nodes at time t, and P DN (t)Stored in N w In the matrix W of ×24, when the number of simulations y1 reaches the maximum value z1 or satisfies formula (13), the simulation is terminated, otherwise the charging load prediction is repeated.

[0046]

[0047] Where: W b represents the b-th column vector in W; A n Wb represents the W of the nth simulation b is the average value in; ξ1 is the convergence accuracy. represents the W of the n-1th simulation b The average value in

[0048] In step 3, considering the comprehensive operating cost and voltage deviation index, the objective function of the mathematical model of ADN dynamic reconstruction and active and reactive joint optimization is:

[0049] Min{μ1(f L +f DG +f O )+μ2C V f V} (14);

[0050] Where: f L 、f DG and f O are respectively the cost of active network loss, wind and solar power abandonment and switch action, f V is the voltage deviation; μ1 and μ2 are weight coefficients, and μ1+μ2=1. C Vrepresents the cost coefficient;

[0051]

[0052]

[0053]

[0054]

[0055] Where, T is the total dispatch time 24h; Ω E and Ω DG are the set of power grid branches and the set of nodes where DG is located; C L , C D and C V is different cost coefficient; V th It is the upper and lower thresholds of the ideal operating voltage. ij,t Represents the current flowing through branch ij at time t. ij represents the resistance of branch ij, Δt represents a single scheduling time period, ΔZ ij Represents a Boolean variable. When its value is 0 / 1, it means that the branch ij is in the open / closed state. DG Indicates the active power of the node where DG is located, Represents the active power of node i in the DG node at time t. ΔZ ij 0 Represents the initial state of the branch ij switch.

[0056] In step 4, the constraints are adjusted taking into account active management techniques and EV spatiotemporal distribution:

[0057] 1) System power constraints:

[0058]

[0059]

[0060]

[0061]

[0062] Where: ΔZ ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state; P G,i,t , Q G,i,t are the active and reactive power injected into node i at time t; P ij,t , Q ij,t is the active and reactive power flowing through branch ij at time t, and both are positive if they flow out from node i; g ij and b ijare the conductance and susceptance of branch ij respectively; V i,t 、V j,t and θ ij,t are the voltage amplitudes of i and j at time t and the voltage phase difference between i and j. ΔZ ij It represents a Boolean variable. When its value is 0 / 1, it means that the branch ij is in an open / closed state. N(i) represents the set of nodes connected to node i.

[0063] 2) Branch current carrying capacity constraints:

[0064]

[0065] Where: I ij,max is the maximum value of the current amplitude. ij Indicates branch connection status;

[0066] 3) Node voltage constraints:

[0067] V i,min ≤V i,t ≤V i,max (twenty four);

[0068] Where: V i,min and V i,max are the lower and upper limits of the voltage at node i.

[0069] 4) Radial network constraints:

[0070]

[0071] Where: E ij 、E ji All variables are 0 or 1, E ij =1 indicates that the branch flow direction is from node j to node i, and node j is the parent node of node i; N(i) indicates the set of nodes connected to node i; Y mk,e , Y km,e It is also a 0 / 1 variable. ij 、E ji All are Boolean variables: E ij Indicates the flow direction variable of branch ij, E ji Indicates the flow direction variable of branch ji, Y mk,e , Y km,e All are 0 / 1 variables. 1j = 0 limits the flow to flow only from node 1, which is the substation node, m and k are the nodes of the dual graph, e is the branch connected by the dual graph node, M K , S K are m and k node sets respectively.

[0072] The specific method is as follows:

[0073] Existing literature on distribution network reconstruction often does not involve dual graphs. Based on graph theory, an algorithm for solving the dual graph of distribution networks is proposed.

[0074] Any graph G on a plane satisfies the following conditions:

[0075] ① On each face F of G i Select a point V i As the apex;

[0076] ②For each edge e in G, draw a line e' that only intersects e and connects the faces F on both sides of e i The vertex V in i 'As an edge.

[0077] This graph is called the dual graph G' of graph G.

[0078] The specific steps to solve the dual graph are as follows:

[0079] Step ①: After closing all switches in the distribution network, convert it into an undirected planar graph G, generate a network adjacency matrix, determine the maximum node number k, and set m = 1.

[0080] Step ②: Determine whether there are two different nodes i and j connected to node k. If not, the branch composed of the nodes connected to node k has no corresponding branch in the dual graph.

[0081] Step 3: Use Floyd's algorithm to find the shortest path between nodes i and j where all node numbers are not greater than k. This path and paths ik and jk are combined to form the mth minimum ring and stored.

[0082] Step ④: Determine whether k is less than 1. If not, set k = k-1, m = m+1, and return to step 2. If satisfied, select a point from each minimum ring as the dual node, and record the number of nodes m in the dual graph.

[0083] Step ⑤: Search and record the connection status between the corresponding dual nodes of each branch in the m smallest rings in the original graph G.

[0084] 5) CB constraints:

[0085] Capacitor Banks (CB), whose constraints are:

[0086]

[0087] Where: y j,t CB is the number of capacitor groups put into operation during period t, which is an integer variable, Q j CB,stepis the reactive output of each group of capacitors.

[0088] Indicates the reactive output of the capacitor put into operation during period t, Indicates the maximum number of capacitors put into operation at time t, Represents any one,Ω CB represents the set of nodes where CB is located, and t represents the time period t;

[0089] 6) SVC constraints:

[0090] Static var compensation (SVC), whose constraints are:

[0091]

[0092] Where: Q j SVC,min , Q j SVC,max It is the lower and upper limit of SVC power.

[0093] represents the reactive power compensation power of node j at time t, Ω SVC Indicates the node set where the SVC is located;

[0094] 7) OLTC operation constraints:

[0095] On Load Tap Changer (OLTC), its constraints are:

[0096]

[0097] Where: r j,t represents the square of OLTC transformation ratio, r j,s Indicates the square difference of the transformation ratio between OLTC gear s and gear s-1, V j,min 、V j,max Respectively represent the minimum / maximum range of the adjustable voltage of node j. Ω Sub is the substation node; V j,t Base is the voltage value of the high voltage side of the transformer; r j,t is the square of OLTC transformation ratio, r j max and r j min For r j,t Upper and lower limits; r j,s is the square difference of the transformation ratio between OLTC gears s and s-1; δ j,s,t OLTC It is a 0 / 1 variable, and the adjustment times are constrained as follows:

[0098]

[0099] Where: δ j,t OLTC,IN and δ j,t OLTC,DE is the OLTC gear adjustment change, t and t-1 represent time; SR j is the maximum range of gear change; j OLTC,max It is the maximum number of times the OLTC gear can be operated within a day. It is a 0 / 1 variable, indicating the gear adjustment change, ensuring that the gears move in sequence 11 1 0 0 0.

[0100] 8) ESS constraints:

[0101]

[0102] Where: y j,t ch and j,t dis Indicates the charge and discharge status, which is a variable of 0 or 1; P j max 、E j,t ESS , α, and β are the maximum charging power of ESS, the power in period t, and the charging and discharging efficiency, respectively. represents the power consumption of ESS in the period t+1, represents the power consumption of ESS in period t, Indicates the maximum power, Indicates the power in the discharge state, Indicates the power in the charging state.

[0103] 9) DG output constraints

[0104] The active power and reactive power constraints of distributed generation (DG) access are as follows:

[0105]

[0106] Where: P DG,j,t and Q DG,j,t are the actual active and reactive power at the DG node at time t respectively; and It is the output prediction value and the amount of DG abandoned at that moment; and the DG is operated within a limited power factor range. DG,j,t , Q DG,j,tThey represent the active and reactive power at the DG node at time t, represents the output forecast value at time t, Indicates the active power angle of DG.

[0107] 10) Switching action times constraints:

[0108]

[0109] Where: ΔZ ij 0 Indicates the initial state of the switch of branch ij; SWmax is the maximum number of actions. ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state, Ω E Represents a collection of power grid branches;

[0110] In step 4, the non-convex power flow constraint is convexified by using the second-order cone relaxation method, as follows:

[0111] In order to ensure the optimality and approximate accuracy of the solution, the variables of the power flow equation are first replaced and defined as:

[0112]

[0113] U i,t , U j,t Corresponding to the "branch voltage" variable and W related to branch "ij" respectively ij,t and T ij,t is the variable associated with line "ij", V i,t 、V j,t ,θ ij,t are the voltage amplitudes of i and j at time t and the voltage phase difference between i and j respectively.

[0114] In the above formula, there is a multiplication of discrete variables and continuous variables. The branch virtual terminal voltage variable determined by the branch on-off state and branch voltage is introduced. Decouple and add the following constraints:

[0115]

[0116] are the node voltage variables related to nodes i and j, U i,t , U j,t Indicates the "branch voltage" variable corresponding to branch "ij", ΔZ ij Represents a Boolean variable. When its value is 0 / 1, it means that the branch ij is in the open / closed state. i,max 、V j,max Respectively represent the minimum / maximum range of the adjustable voltage of nodes i and j.

[0117] And the formula shows that when ΔZ ij = 0, U i,t ij =U j,t ij =0; when ΔZ ij =1, U i,t ij =U i,t , U j,t ij =U j,t , after variable substitution, the power flow equation is transformed into:

[0118]

[0119] are the node voltage variables related to nodes i and j, respectively, ij,t , Q ij,t is the active and reactive power flowing through branches ij and at time t, both of which are positive if they flow out from node i; P ji,t , Q ji,t is the active and reactive power flowing through branches ji and at time t, both of which are positive if they flow out from node j; g ij and b ij are the conductance and susceptance of branch ij respectively; W ij,t and T ij,t is the variable associated with line “ij”;

[0120]

[0121] are the node voltage variables related to nodes i and j, W ij,t and T ij,t are variables associated with line “ij”.

[0122] The second-order cone relaxation of equation (36) is:

[0123]

[0124] are the node voltage variables related to nodes i and j, W ij,t and T ij,t are variables associated with line “ij”.

[0125] In step 4, the branch virtual terminal voltage variable is introduced, and the non-convex variable product term is subjected to variable separation linearization, as follows:

[0126] Eliminate the absolute value nonlinear constraint contained in the switching number constraint and introduce auxiliary variables for linearization:

[0127]

[0128] And Y in the formula ij is an auxiliary variable, Y ij =1 indicates that the switch state changes; otherwise, it remains unchanged.

[0129] Indicates the initial state of the switch of branch ij; ΔZ ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state;

[0130] The absolute value of voltage deviation in the objective function of the mathematical model of ADN dynamic reconstruction and reactive and active power joint optimization is calculated by introducing the auxiliary variable U i,t Aux To perform linearization:

[0131]

[0132] f V is the power deviation, Represents auxiliary variables, U i,t , U j,t Indicates the "branch voltage" variable corresponding to branch "ij", V th The thresholds for the upper and lower limits of the ideal operating voltage.

[0133] Through transformation, the joint optimization mathematical model is converted into a mixed integer second-order cone programming problem.

[0134] The method further includes step 6, wherein a simulation verification is performed based on a modified IEEE 33-node system example:

[0135] 1) Based on different input parameters, four different scenarios are modeled and analyzed: weekdays, weekends, high temperature days, and congested days.

[0136] 2) Charging loads in different scenarios fluctuate greatly in time and space. The optimization results under different modes are compared and analyzed to ensure safe and stable operation of EV time and space load access to ADN. The congested day with the largest load fluctuation is selected to analyze the optimization results under four different modes:

[0137] Mode 1: Only ESS and DG active power optimization is considered.

[0138] Mode 2: Based on Mode 1, voltage deviation is taken into account.

[0139] Mode 3: Based on Mode 2, OLTC, CB, and SVC are added to perform active-reactive power joint optimization.

[0140] Mode 4: Add dynamic network reconstruction based on the joint optimization of active and reactive power in Mode 3.

[0141] The present invention provides a charging load prediction method based on ADN dynamic reconstruction and active and reactive joint optimization model, and the technical effects are as follows:

[0142] 1) The spatiotemporal distribution prediction model of EV charging load taking into account temperature, traffic conditions and subjective willingness of the car owner in the present invention takes into account the influence of weather, people and roads on EV travel paths, and can truly reflect the randomness of EV travel paths and the coupling between various factors.

[0143] 2) The ADN dynamic reconstruction and active and reactive power joint optimization model taking into account active management technology and economic and technical indicators in the present invention can significantly reduce network losses, promote DG consumption and reduce system voltage fluctuations; at the same time, the preferred ADN operation plan can be obtained by changing the weights of economic and technical indicators.

[0144] 3) The present invention uses second-order cone relaxation and variable product linearization to effectively convexify the joint optimization model, which reduces the complexity of the solution while meeting the engineering solution accuracy, and significantly improves the solution efficiency of the model. Energy storage has a significant effect in peak-shaving and valley-filling. The reasonable configuration of energy storage systems at EV charging stations will further reduce the impact of EV temporal and spatial load randomness on the power grid and improve the economy and reliability of ADN operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0145] Figure 1 Schematic diagram of the transportation network based on Markov decision theory.

[0146] Figure 2 Flowchart for electric vehicle charging load calculation

[0147] Figure 3 This is a schematic diagram of the road-network coupling topology in a certain urban area;

[0148] Figure 4(1) is a schematic diagram of 24-hour dynamic load data parameters for industrial areas, commercial areas, and residential areas.

[0149] Figure 4(2) 24-hour time-of-use electricity price parameter diagram.

[0150] Figure 5 (1) shows the spatiotemporal distribution of charging load on weekdays;

[0151] Figure 5(2) shows the spatiotemporal distribution of charging load on weekends.

[0152] Figure 6 (1) shows the spatiotemporal distribution of charging load on a high temperature day;

[0153] Figure 6 (2) shows the spatiotemporal distribution of charging load on a congested day;

[0154] Figure 7 (1) is a 24h network node voltage distribution diagram under mode 1;

[0155] Figure 7(2) is the 24h network node voltage distribution diagram under mode 2;

[0156] Figure 7(3) is the 24h network node voltage distribution diagram under mode 3;

[0157] Figure 7(4) is the 24h network node voltage distribution diagram under mode 4.

[0158] FIG8 (1) is a schematic diagram of the output of reactive power compensation of each device node;

[0159] Figure 8(2) is a schematic diagram of the energy storage charging and discharging status of each device node.

[0160] Fig. 9 This is the second-order relaxation error analysis diagram of the distribution network. DETAILED DESCRIPTION

[0161] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0162] The charging load prediction method based on ADN dynamic reconstruction and active and reactive joint optimization model includes the following steps:

[0163] 1. Considering the comprehensive operating cost and voltage deviation index, the mathematical model of dynamic reconstruction of active distribution network and joint optimization of reactive power and active power is as follows:

[0164] Min{μ1(f L +f DG +f O )+μ2C V f V}

[0165] Where: f L 、f DG and f O are respectively the cost of active network loss, wind and solar power abandonment and switch action, f V is the voltage deviation; μ1 and μ2 are weight coefficients, and μ1+μ2=1.

[0166] 2. EV charging load prediction model:

[0167] Considering the speed-flow model based on transportation capacity, the EV unit mileage energy consumption model considering the actual temperature, transportation conditions, and the EV state of charge:

[0168]

[0169] Where: ν ti,j (x) is the speed of the EV at x on the road (i, j), i and j are the two sides of the road, d t i,j (x) represents the traffic volume of road (i, j), Tr i,j represents the transportation capacity, a1, a2, a3 are road adaptive coefficients, χ t is the road saturation at time t.

[0170] The power consumption per kilometer of an EV with air conditioning turned on is calculated as follows:

[0171]

[0172] Where: E t i,j (x) is the EV power consumption per kilometer without air conditioning on.

[0173] The change of actual temperature can affect the air conditioning activation rate of EV users, which can be expressed as:

[0174]

[0175] Where: K temp is the energy proportionality coefficient. Assume that a random number r with uniform temperature distribution is generated. If r<K pect , then turn on the air conditioner. q1~q4 and c1~c3 are fitting parameters, and b represents temperature.

[0176] Combining equations (2) and (3), the actual power consumption per kilometer of EV can be expressed as:

[0177]

[0178] Therefore, the calculation formula for the charge state of the EV at location x is:

[0179]

[0180] The charging behavior of electric vehicles can be divided into two categories: charging when the state of charge is insufficient and charging at the end of the journey due to subjective mileage anxiety: 1) When the SOC of the electric vehicle is not enough to reach the next destination or charging station, the user will choose to charge at the current location. 2) The electric vehicle is not enough to meet the next journey, and the user will charge the electric vehicle out of subjective will. The smaller the ratio of the remaining power to the next driving distance, the stronger the willingness to charge. Based on the above two situations, fuzzy theory is used to calculate the user satisfaction index uu sat To describe the subjective charging willingness of electric vehicle users:

[0181]

[0182] Introducing Ffuz The fuzzy set representing charging willingness, whose membership function F fuz (uu sat ) can be expressed by the following formula (7):

[0183]

[0184] In the formula: uu max sat 、uu min sat uu sat The upper and lower limits of .

[0185] EV has two charging modes. Considering the charging cost and battery loss, the default is slow charging. If the parking is over, the slow charging cannot meet the expected state of charge SOC. exp , that is, fast charging is selected when formula (8) is satisfied:

[0186]

[0187] Where: P slow is the rated slow charging power of EV, T i p , SOC i init are the parking time and initial state of charge of the EV at parking point i respectively. If the SOC of the EV is lower than SOC thr , it will be charged midway, the charging location is L mid par is a charging station close to the destination, and the charging time is T char The calculation formula is as follows:

[0188]

[0189]

[0190] Where: D thr From SOC to EV init To critical state of charge SOC thr The driving distance; ∑D r represents the distance between the starting point of the EV trip and the rth charging station; r c Indicates that the SOC of the EV drops to SOC thr When the number of charging stations passed; f(r c ) represents r c The corresponding traffic network, P fast and P slow It is the rated fast charging power and slow charging power.

[0191] According to the coupling relationship between traffic nodes and distribution networks, the time and space loads of each distribution node are counted, and the charging load P at distribution node z is z (t) and the total charging load of the distribution network P DN (t) can be expressed as:

[0192]

[0193]

[0194] Where: N EV and N w are the number of charging nodes EV and charging nodes at time t, and P DN (t)Stored in N w In the matrix W of ×24, when the number of simulations y1 reaches the maximum value z1 or satisfies formula (14), the simulation is terminated, otherwise the charging load prediction is repeated.

[0195]

[0196] Where: W b represents the b-th column vector in W; A n Wb represents the W of the nth simulation b The specific flow chart of time-space load calculation and prediction is as follows: Figure 2 shown.

[0197] 3. Constraints of the joint optimization mathematical model:

[0198] 1) System power constraints:

[0199]

[0200]

[0201]

[0202]

[0203] Where: ΔZ ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state; P G,i,t , Q G,i,t are the active and reactive power injected into node i at time t; P ij,t , Q ij,t is the active and reactive power flowing through branch ij at time t, and both are positive if they flow out from node i; g ij and b ij are the conductance and susceptance of branch ij respectively; V i,t 、V j,t and θij,t are the voltage amplitudes of i and j at time t and the voltage phase difference between i and j respectively.

[0204] 2) Branch current carrying capacity constraints:

[0205]

[0206] Where: I ij,max is the maximum value of the current amplitude.

[0207] 3) Node voltage constraints:

[0208] V i,min ≤V i,t ≤V i,max (twenty four);

[0209] Where: V i,min and V i,max are the lower and upper limits of the voltage at node i.

[0210] 4) Radial network constraints:

[0211]

[0212] Where: E ij 、E ji All variables are 0 or 1, E ij =1 indicates that the branch flow direction is from node j to node i, and node j is the parent node of node i; N(i) indicates the set of nodes connected to node i; Y mk,e , Y km,e Also a 0 / 1 variable.

[0213] Distribution networks are generally designed as closed-loop networks and operated in open-loop networks. The minimum spanning tree method shown in the following steps can ensure the radiation structure of the distribution network during the reconstruction process. Existing literature related to distribution network reconstruction often does not involve dual graphs. Based on graph theory, an algorithm for solving the dual graph of distribution networks is proposed. For any graph G on a plane, if the following conditions are met: (1) a point Vi is selected as a vertex in each face Fi of G; (2) for each edge e in G, a line e' is drawn, and this line only intersects with e, and it connects the vertices Vi' in the faces Fi on both sides of e as edges. Then this graph is called the dual graph G' of graph G.

[0214] The specific steps to solve the dual graph are as follows:

[0215] Step 1: After closing all switches in the distribution network, convert it into an undirected planar graph G, generate a network adjacency matrix, determine the maximum node number k, and set m=1.

[0216] Step 2: Determine whether there are two different nodes i and j connected to node k. If not, the branch composed of the nodes connected to node k has no corresponding branch in the dual graph.

[0217] Step 3: Use Floyd's algorithm to find the shortest path between nodes i and j where all node numbers are not greater than k. This path and paths ik and jk are combined to form the mth minimum ring and stored.

[0218] Step 4: Determine whether k is less than 1. If not, set k = k-1, m = m+1, and return to step 2. If satisfied, select a point from each minimum ring as the dual node, and record the number of nodes m in the dual graph.

[0219] Step 5: Search and record the connection status between the dual nodes corresponding to each branch in the m smallest rings in the original graph G.

[0220] 5) CB constraints:

[0221] CB is a discrete reactive power compensation, and its constraints are:

[0222]

[0223] Where: y j,t CB is the number of capacitor groups put into operation during period t, which is an integer variable, Q j CB,step is the reactive output of each group of capacitors.

[0224] 6) SVC constraints:

[0225] SVC is a continuous reactive power regulation device, and its constraints are:

[0226]

[0227] Where: Q j SVC,min , Q j SVC,max It is the lower and upper limit of SVC power.

[0228] 7) OLTC operation constraints:

[0229] OLTC is an on-load tap changer, and its constraints are:

[0230]

[0231] Where: Ω Sub is the substation node; V j,t Base is the voltage value of the high voltage side of the transformer; r j,t is the square of the ratio, r j max and rj min For r j,t Upper and lower limits; r j,s is the square difference of the transformation ratio between OLTC gears s and s-1; δ j,s,t OLTC It is a 0 / 1 variable, and the adjustment times are constrained as follows:

[0232]

[0233] Where: δ j,t OLTC,IN and δ j,t OLTC,DE OLTC gear adjustment changes; SR j OLTC is the maximum range of gear change; j OLTC,max It is the maximum number of times the OLTC gear can be operated within a day.

[0234] 8) ESS constraints:

[0235]

[0236] Where: y j,t ch and j,t dis Indicates the charge and discharge status, which is a variable of 0 or 1; P j max 、E j,t ESS , α, and β are the maximum charging power of ESS, the power in period t, and the charging and discharging efficiency, respectively.

[0237] 9) DC output constraints

[0238] The active power and reactive power constraints of DG access are as follows:

[0239]

[0240] Where: P DG,j,t and Q DG,j,t are the actual active and reactive power at the DG node at time t; P f DG,j,t and P curt DG,j,t It is the output forecast value and the amount of DG abandoned at that moment; and the DG is operated within a limited power factor range.

[0241] 10) Switching action times constraints:

[0242]

[0243] Where: ΔZ ij 0 Indicates the initial state of the switch of branch ij; SWmax is the maximum number of actions.

[0244] 4. The non-convex form of the power flow constraint is convexified by using the second-order cone relaxation method. To ensure the optimality and approximate accuracy of the solution, the power flow equation is first replaced with variables and defined as:

[0245]

[0246] In the above formula, there is a multiplication of discrete variables and continuous variables. The virtual terminal voltage variable U of the branch determined by the branch on-off state and the branch voltage is introduced. i ij , U j ij Decouple and add the following constraints:

[0247]

[0248] From the formula, we know that when ΔZ ij = 0, U i,t ij =U j,t ij =0; when ΔZ ij =1, U i,t ij =U i,t , U j,t ij =U j,t , after variable substitution, the power flow equation is transformed into:

[0249]

[0250]

[0251] The second-order cone relaxation of equation (32) is:

[0252]

[0253] The branch virtual terminal voltage variable is introduced to perform variable separation linearization on the non-convex variable product term, eliminate the absolute value nonlinear constraint contained in the switching number constraint condition, and introduce auxiliary variables for linearization:

[0254]

[0255] And Y in the formula ij is an auxiliary variable, Y ij =1 indicates that the switch state changes; otherwise, it remains unchanged.

[0256] The absolute value of voltage deviation in the objective function of the mathematical model of ADN dynamic reconstruction and reactive and active power joint optimization is calculated by introducing the auxiliary variable U i,t Aux To perform linearization:

[0257]

[0258] Through transformation, the joint optimization mathematical model is converted into a mixed integer second-order cone programming problem, and the solution efficiency is greatly improved.

[0259] 5. Example analysis:

[0260] The EV travel path based on Markov decision theory includes five parameters: the decision time set, spatial state set, action set, transition probability matrix set and reward value of the travel chain, which are denoted as {T, S, A, P, R} respectively. The traffic network diagram is shown in Figure 1 As shown in the figure, the flow chart of charging load calculation and prediction based on travel path and charging time and space information is as follows Figure 2 Taking the modified IEEE33 node system and the urban road network coupling network as an example, the road-network coupling topology diagram is shown in Figure 3 (1), the coupling relationship of the distribution network nodes, the saturation of the network roads in each period, and the specific distance of each road are shown in Tables 1, 2, and 3:

[0261] Table 1 Road-network node coupling information parameters

[0262]

[0263] Table 2 Traffic road saturation parameters

[0264]

[0265] Table 3 Traffic road length parameters

[0266]

[0267] The travel chains, proportions and specific parameters of weekdays and weekends in this area are shown in Table 4, and the probability distribution of different travel chains is shown in Table 5.

[0268] Table 4. Travel chain structure and composition ratio parameters on weekdays and weekends

[0269]

[0270] Table 5. Travel chain structure and composition ratio parameters on weekdays and weekends

[0271]

[0272] According to the temperature energy consumption model and the user's subjective charging willingness model, the charging load distribution diagram on weekdays and weekends is obtained as follows: Figure 5(1) to Figure 5(2) As shown in the figure, and the schematic diagram of charging load distribution on high temperature days and congested days is as follows Figure 6(1) to Figure 6(2) The 24-hour dynamic load data parameters and time-of-use electricity price parameters of industrial areas, commercial areas and residential areas are shown in Figure 6. The congestion day with the largest load fluctuation is selected, and the optimization results under the four modes are shown in Table 6.

[0273] Table 6 Optimization results in four modes

[0274]

[0275] From Table 6, it can be seen that due to the large fluctuation of charging load, when ESS and DG are not considered, the voltage will be lower than 0.95pu, and the voltage is lower than the lower limit, so no comparative analysis is set. Based on the consideration of active optimization, the optimization results of the four modes are shown in Table 1. Compared with Mode 1 and Mode 2, it can be seen that the operating cost is slightly increased, but the risk of voltage lower limit is significantly reduced; compared with Mode 2 and Mode 3, it can be seen that the active-reactive joint optimization significantly reduces the line loss cost and active power purchase cost, which are reduced by 49.4% and 8.9% respectively, and at the same time greatly reduces the wind and light abandonment of DG, which is reduced by 81.1%; compared with Mode 3 and Mode 4, it can be seen that considering dynamic network reconstruction can further reduce network loss and active power purchase cost, which are reduced by 33.8% and 4.4% respectively, and at the same time promote the complete absorption of DG, indicating that dynamic reconstruction and active-reactive joint optimization have good effects on improving the economic operation of distribution network after EV time and space load access.

[0276] The 24h network node voltage distribution in four modes is as follows: Figure 7(1) to Figure 7(4) shown.

[0277] The reactive power compensation and energy storage conditions of SVC, CB and OLTC are shown in Figure 8(1) and Figure 8(2) respectively.

[0278] The specific parameters of these devices are shown in Table 7. When performing error analysis, the second-order cone relaxation error is as follows Fig. 9 As shown. Fig. 9 It can be seen that the second-order cone error is 10 -8 The order of magnitude, the second-order cone meets the engineering requirements. It can be seen from the figure that there are breakpoints because the distribution network operates in an open loop, there are disconnected branches in the network, and the error of disconnected branches is 0.

[0279] Table 7 WT, PV, ESS, CB and SVC equipment parameters

[0280]

[0281] To verify the superiority of the model, the solution speed before and after the model conversion is compared. The optimization model before conversion is MINLP. For large-scale nonlinear programming problems, the existing solution methods include intelligent algorithms and traditional numerical methods. This paper will compare the particle swarm algorithm and Knitro algorithm package as examples. As shown in Table 8, it can be seen that the direct solution takes a long time and is difficult to solve. After the model is converted to MISOCP, the solution efficiency is significantly improved.

[0282] Table 8 Comparison of solution speed of different methods

[0283] Table 8 Different methods to solve the speed

[0284]

[0285]

[0286] 1) According to the known parameters and the established mathematical model, the temperature energy consumption model and the user's subjective charging willingness model make the charging time and space distribution more realistic. Based on the above considerations, four different scenarios, namely weekdays, weekends, high temperature days and congested days, are modeled and analyzed according to different input parameters.

[0287] 2) On weekdays, the peak hours of electricity consumption in the work area are mainly concentrated between 9:00 and 12:00, and the peak hours of electricity consumption in residential and commercial areas are mainly concentrated between 18:00 and 22:00. The charging load at all locations is almost zero during the rest of the time. On weekends, when people are not working, the charging load in the work area is basically zero, and the charging load in residential and commercial areas is evenly distributed from 09:00 to 24:00. This spatiotemporal distribution is mainly determined by the travel chain structure and departure time of the itinerary. As shown in Figure 5(1) and Figure 5(2), the load peak periods of each area on weekdays and weekends are delayed by 1-3 hours compared to the departure start period. This time difference is caused by traffic travel. In addition, the load peak on weekdays is about 1.5 times the load peak on weekends. This is because 30% of the number of vehicles traveling on weekends are idle, thus generating no additional charging load.

[0288] 3) The impact of temperature and traffic conditions on charging demand was studied, and the charging load prediction for high temperature days and congested days was carried out. The results are shown in Figure 6 (1) and Figure 6 (2). It should be noted that the input parameters of high temperature days and congested days are the same as those of weekdays, except for temperature and electric vehicle driving speed. It can be seen that the temporal and spatial distribution of high temperature days is basically the same as that of weekdays, but the load peak of the former increases sharply, and the increased load accounts for 80.1% of the load on weekdays. This is because on high temperature days, EV's electricity consumption is not only used for transportation, but also for air conditioning and refrigeration. For congested days, the charging load amplitude increased by 52.3% compared with the load on weekdays, and the overall load was delayed in the time dimension, especially in residential areas, which gradually entered the load peak after 20:00. This is because the congestion of EVs increases the traffic time, resulting in an overall delay in the access period.

[0289] 4) After analyzing the charging load in different scenarios, it can be found that the charging load has large fluctuations in time and space. In order to ensure the safe and stable operation of EV time and space load access to ADN, this paper conducts a comparative analysis.

[0290] Mode 1: Only ESS and DG active power optimization is considered.

[0291] Mode 2: Based on Mode 1, voltage deviation is taken into account.

[0292] Mode 3: Based on Mode 2, OLTC, CB, and SVC are added to perform active-reactive power joint optimization.

[0293] Mode 4: Add dynamic network reconstruction based on the joint optimization of active and reactive power in Mode 3.

[0294] 5) Due to the large fluctuation of charging load, when ESS and DG are not considered, the voltage will be lower than 0.95pu, and the voltage will be lower than the lower limit, so no comparative analysis is set. Based on the consideration of active power optimization, the optimization results of the four modes are shown in Table 6. Comparing Mode 1 and Mode 2, it can be seen that the operating cost increases slightly, but the risk of voltage lowering is significantly reduced. Figure 7(1) to Figure 7(4) Analysis shows that; comparing mode 2 and mode 3, it can be seen that the active-reactive joint optimization significantly reduces the line loss cost and the active power purchase cost, which are reduced by 49.4% and 8.9% respectively, and greatly reduces the wind and light abandonment of DG, which is reduced by 81.1%; comparing mode 3 and mode 4, it can be seen that considering dynamic network reconstruction can further reduce network loss and active power purchase costs, which are reduced by 33.8% and 4.4% respectively, and promote the complete absorption of DG, indicating that dynamic reconstruction and active-reactive joint optimization after EV time-space load access have a good effect on improving the economic operation of the distribution network.

[0295] 6) The network node voltage distribution in four modes is as follows Figure 7(1) to Figure 7(4) As shown. Figure 7(1) to Figure 7(4) It can be found that due to the large load connected in Mode 1 and Mode 2, and the fact that OLTC, CB, and SVC are not considered, the reactive voltage compensation of the distribution network is insufficient, and the voltage level is generally low. However, Mode 2 takes the voltage deviation into consideration, and the voltage around 0.95pu (the blue part in the figure) is significantly reduced; after Mode 3 adopts active and reactive joint optimization, the voltage level of each node is significantly improved, but the voltage curve changes greatly, which is caused by the large change of charging load over time; and dynamic reconstruction can improve voltage fluctuations. Compared with Mode 2 and Mode 3, the dynamic reconstruction and active-reactive joint optimization of Mode 4 make the node voltage more stable, all between 0.97pu and 1.03pu. It can be seen that by changing the network topology, the voltage fluctuation caused by spatiotemporal load can be effectively smoothed and the voltage quality of the distribution network can be improved. It also shows that when EV spatiotemporal load is connected, dynamic reconstruction and various active management measures can ensure the safe and economic operation of the distribution network. In addition, after adding the voltage offset threshold, the operating cost only increases slightly. Operators can reduce the risk of voltage over-limit by sacrificing a certain degree of economy and changing the weight of voltage offset.

[0296] 7) The output of reactive power compensation is shown in Figure 8 (1). The output of CB and SVC equipment at node 5 gradually increases after 6:00 and reaches a peak at 10:00. This is because the area corresponding to node 25 connected to node 5 is an industrial area, and its peak power consumption is mainly concentrated in 09:00-12:00. At the same time, the total load of the system is also the largest during this period, so the compensation value of node 5 reaches the maximum during this period, and the compensation values ​​of other nodes also increase. The charging and discharging of energy storage is determined by the electricity price and load. As shown in Figure 8 (2), the electricity prices are at the lowest and highest levels at 0:00-7:00 and 9:00-12:00 respectively. The energy storage is affected by the time-of-use electricity price and is charged at 0:00-7:00 and discharged at 9:00-12:00. The residential and commercial areas to which nodes 15 and 32 belong do not enter the peak load until after 20:00. To reduce the impact of high load on the power grid, the energy storage is charged from 13:00 to 19:00 and discharged from 20:00 to 24:00. It can be seen that the energy storage plays a good role in peak load reduction and valley filling after EV access. In addition, the charging and discharging trends of the energy storage at the two nodes are roughly the same, indicating that electricity price is the main reason affecting the output of energy storage.

[0297] 8) By Fig. 9 It can be seen that the second-order cone error is 10 -8 The order of magnitude, the second-order cone meets the engineering requirements. It can be seen from the figure that there are breakpoints because the distribution network operates in an open loop, there are disconnected branches in the network, and the error of disconnected branches is 0.

Claims

1. A charging load prediction method based on ADN dynamic reconstruction and active and reactive joint optimization model, characterized by The following steps are involved: Step 1: Establish an EV unit mileage energy consumption model; Step 2: Considering the subjective will of the car owner, a spatiotemporal distribution prediction model of EV charging load is established to predict charging load in different scenarios; Step 3: Taking into account the comprehensive operating cost and voltage deviation index, determine the objective function and establish a mathematical model including ADN dynamic reconstruction and active and reactive joint optimization; Step 4: Based on the constraints, the joint optimization mathematical model in step 3 is transformed using the second-order cone relaxation and variable product linearization method; In step 3, considering the comprehensive operating cost and voltage deviation index, the objective function of the mathematical model of ADN dynamic reconstruction and active and reactive joint optimization is: Min{μ1(f L +f DG +f O )+μ2C V f V }(14); Where: f L 、f DG and f O are respectively the cost of active network loss, wind and solar power abandonment and switch action, f V is the voltage deviation; μ1 and μ2 are weight coefficients, and μ1+μ2=1; C V represents the cost coefficient; Where T is the total scheduling time; Ω E and Ω DG are the set of power grid branches and the set of nodes where DG is located; C L , C D and C V is different cost coefficient; V th The upper and lower thresholds of the ideal operating voltage; I ij,t represents the current flowing through branch ij at time t; R ij represents the resistance of branch ij, △t represents a single scheduling time period, △Z ij Represents a Boolean variable. When its value is 0 / 1, it means that the branch ij is in the open / closed state. DG Indicates the active power of the node where DG is located, represents the active power of node i in the DG node at time t; ΔZ ij 0 Represents the initial state of the branch ij switch.

2. The charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model according to claim 1 is characterized by: In step 1, based on the speed and flow of transportation capacity, the EV unit mileage energy consumption model considering the actual temperature and transportation conditions and the EV charge state are as follows: Where: t i,j (x) is the speed of the EV at x on the road (i, j), where i and j are the two sides of the road respectively. represents the traffic volume of road (i, j), Tr i,j represents the transportation capacity, a1, a2, a3 are the road adaptation coefficients, χ t is the road saturation at time t; represents the β-power of road saturation at time t, represents the a3 power of the road saturation at time t; The power consumption per kilometer of an EV with air conditioning turned on is calculated as follows: Where: E t i,j (x) is the power consumption per kilometer of the EV without air conditioning on; The actual temperature change can affect the air conditioning activation rate of EV users, which can be expressed as: Where: K temp is the energy proportionality coefficient. Assume that a random number r with uniform temperature distribution is generated. If r<K pect , then turn on the air conditioner; q1~q4 and c1~c3 are fitting parameters, b represents temperature, K pect Indicates the air conditioning opening rate at different temperatures; Combining equations (2) and (3), the actual power consumption per kilometer of EV is expressed as: represents the energy proportional coefficient at time t, It indicates the power consumption per kilometer of the EV without the air conditioner turned on; Therefore, the calculation formula for the charge state of the EV at location x is: SOC(x) is the calculation formula for the state of charge of the EV at location x. init Indicates the initial charge state of EV driving, E t (x) Indicates the actual power consumption per kilometer of EV, C EV Indicates battery capacity, D X Indicates the distance from the starting point to location x.

3. The charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model according to claim 1 is characterized by: In step 2, the EV charging load spatiotemporal distribution prediction model considering the subjective will of the car owner is applied to the charging load prediction in different scenarios, as follows: Remaining power and next driving distance D next The smaller the ratio, the stronger the willingness to charge; according to the user's subjective willingness uu sat To describe the subjective charging willingness of EV users: SOC(x) represents the charge state of the EV at location x. EV Indicates battery capacity, D next Indicates the next driving distance, E t (x) represents the actual power consumption per kilometer of the EV; Introducing F fuz The fuzzy set representing charging willingness, whose membership function F fuz (uu sat ), expressed by the following formula (7): In the formula: uu max sat 、uu min sat uu sat The upper and lower limits of sat represents the user's subjective willingness, τ represents the probability of charging the EV out of subjective willingness; EV has two charging modes. Considering the charging cost and battery loss, the default is slow charging. If the parking is over, the slow charging cannot meet the expected state of charge SOC. exp , that is, fast charging is selected when formula (8) is satisfied: C EV Indicates battery capacity, SOC exp Indicates the desired state of charge; Where: P slow is the rated slow charging power of EV, SOC i init are the parking time and initial state of charge of EV at parking point i respectively; If the SOC of the EV is lower than SOC thr , it will be charged midway, the charging location is L mid par is a charging station close to the destination, and the charging time is T char The calculation formula is as follows: D thr It is expressed as EV's initial state of charge SOC init To critical state of charge SOC thr Driving distance, SOC thr Indicates the critical state of charge, is the charging location of the charging station close to the destination, c Indicates that the SOC of the EV drops to SOC thr When the number of charging stations passed, ∑D r represents the distance between the starting point of the EV trip and the rth charging station, J represents the Jth location; Where: D thr From SOC to EV init To critical state of charge SOC thr The driving distance; ∑D r represents the distance between the starting point of the EV trip and the rth charging station; r c Indicates that the SOC of the EV drops to SOC thr When the number of charging stations passed; f(r c ) represents r c The corresponding traffic network, P fast and P slow is the rated fast charging power and slow charging power; T char Indicates charging time, D mid represents the distance from the starting point to the charging station where the electric vehicle will reach when the SOC is about to reach 0, ∑D r represents the distance between the starting point of the EV trip and the rth charging station; According to the coupling relationship between traffic nodes and distribution networks, the time and space loads of each distribution node are counted, and the charging load P at distribution node z is z (t) and the total charging load of the distribution network P DN (t) is expressed as: represents the charging power of the nth electric vehicle at time t, and n represents the number of electric vehicles charged at node z; P z (t) represents the charging load at the distribution node z; Where: N EV and N w are the number of charging nodes EV and charging nodes at time t, and P DN (t)Stored in N w ×24 matrix W, when the number of simulations y1 reaches the maximum value z1 or satisfies formula (13), the simulation terminates, otherwise the charging load prediction is repeated; Where: W b represents the b-th column vector in W; A n Wb represents the W of the nth simulation b The average value in; ξ1 is the convergence accuracy; represents the W of the n-1th simulation b The average value in .

4. The charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model according to claim 1 is characterized by: In step 4, the constraints are adjusted taking into account active management techniques and EV spatiotemporal distribution: 1) System power constraints: Where: ΔZ ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state; P G,i,t , Q G,i,t are the active and reactive power injected into node i at time t; P ij,t , Q ij,t is the active and reactive power flowing through branch ij at time t, and both are positive if they flow out from node i; g ij and b ij are the conductance and susceptance of branch ij respectively; V i,t 、V j,t and θ ij,t are the voltage amplitudes of i and j at time t and the voltage phase difference between i and j; △Z ij represents a Boolean variable, and when its value is 0 / 1, it means that the branch ij is in an open / closed state, and N(i) represents the set of nodes connected to the node i; 2) Branch current carrying capacity constraints: Where: I ij,max is the maximum value of the current amplitude; ij Indicates branch connection status; 3) Node voltage constraints: In i,min ≤V i,t ≤V i,max (24); Where: V i,min and V i,max is the lower and upper limits of the voltage at node i; 4) Radial network constraints: Where: E ij 、E ji All variables are 0 or 1, E ij =1 indicates that the branch flow direction is from node j to node i, and node j is the parent node of node i; N(i) indicates the set of nodes connected to node i; Y mk,e , Y km,e Also a 0 / 1 variable; E ij 、E ji All are Boolean variables: E ij Indicates the flow direction variable of branch ij, E ji Indicates the flow direction variable of branch ji, Y mk,e , Y km,e All are 0 / 1 variables; E 1j = 0 limits the flow to flow only from node 1, is the substation node, m and k are the nodes of the dual graph, e is the branch connected by the dual graph node, M K , S K are m and k node sets respectively; 5) CB constraints: Capacitor Banks (CB), whose constraints are: Where: y j,t CB is the number of capacitor groups put into operation during period t, which is an integer variable, Q j CB,step is the reactive output of each group of capacitors; Indicates the reactive output of the capacitor put into operation during period t, Indicates the maximum number of capacitors put into operation at time t, Represents any one,Ω CB represents the set of nodes where CB is located, and t represents the time period t; 6) SVC constraints: Static var compensation (SVC), whose constraints are: Where: Q j SVC,min , Q j SVC,max are the lower and upper limits of SVC power; represents the reactive power compensation power of node j at time t, Ω SVC Indicates the node set where the SVC is located; 7) OLTC operation constraints: On Load Tap Changer (OLTC), its constraints are: Where: r j,t represents the square of OLTC transformation ratio, r j,s Indicates the square difference of the transformation ratio between OLTC gear s and gear s-1, V j,min 、V j,max Respectively represent the minimum / maximum range of the adjustable voltage of node j; Ω Sub is the substation node; V j,t Base is the voltage value of the high voltage side of the transformer; r j,t is the square of OLTC transformation ratio, r j max and r j min For r j,t Upper and lower limits; r j,s is the square difference of the transformation ratio between OLTC gears s and s-1; δ j,s,t OLTC It is a 0 / 1 variable, and the adjustment times are constrained as follows: Where: δ j,t OLTC,IN and δ j,t OLTC,DE is the OLTC gear adjustment change, t and t-1 represent time; SR j is the maximum range of gear change; j OLTC,max The maximum number of times the OLTC gear can be operated within a day; It is a 0 / 1 variable, indicating the gear adjustment change, ensuring that the gears act in sequence 1 1 1 00 0; 8) ESS constraints: Where: y j,t ch and j,t dis Indicates the charge and discharge status, which is a variable of 0 or 1; P j max 、E j,t ESS , α, β are the maximum charging power of ESS, the power in period t, and the charging and discharging efficiency respectively; represents the power consumption of ESS in the period t+1, represents the power consumption of ESS in period t, Indicates the maximum power, Indicates the power in the discharge state, Indicates the power in charging state; 9) DG output constraints The active power and reactive power constraints of distributed generation (DG) access are as follows: Where: P DG,j,t and Q DG,j,t are the actual active and reactive power at the DG node at time t respectively; and is the output prediction value and the amount of DG abandoned at that moment; and the DG is operated within a limited power factor range; P DG,j,t , Q DG,j,t They represent the active and reactive power at the DG node at time t, represents the output forecast value at time t, represents the active power angle of DG; 10) Switching action times constraints: Where: ΔZ ij 0 Indicates the initial state of the switch of branch ij; SWmax is the maximum number of actions; ΔZ ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state, Ω E Represents a collection of power grid branches.

5. The charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model according to claim 1 is characterized by: In step 4, the non-convex power flow constraint is convexified by using the second-order cone relaxation method, as follows: In order to ensure the optimality and approximate accuracy of the solution, the variables of the power flow equation are first replaced and defined as: U i,t , U j,t Corresponding to the "branch voltage" variable related to branch "ij", W ij,t and T ij,t is the variable associated with line "ij", V i,t 、V j,t ,θ ij,t are the voltage amplitudes of i and j at time t and the voltage phase difference between i and j; In the above formula, there is a multiplication of discrete variables and continuous variables. The virtual terminal voltage variable U of the branch determined by the branch on-off state and the branch voltage is introduced. i ij , U j ij Decouple and add the following constraints: are the node voltage variables related to nodes i and j, U i,t , U j,t Indicates the "branch voltage" variable corresponding to branch "ij", △Z ij Represents a Boolean variable. When its value is 0 / 1, it means that the branch ij is in the open / closed state. i,max 、V j,max Respectively represent the minimum / maximum range of adjustable voltage of nodes i and j; And the formula shows that when ΔZ ij = 0, U i,t ij =U j,t ij =0; when ΔZ ij =1, U i,t ij =U i,t , U j,t ij =U j,t , after variable substitution, the power flow equation is transformed into: are the node voltage variables related to nodes i and j, respectively, ij,t , Q ij,t is the active and reactive power flowing through branches ij and at time t, both of which are positive if they flow out from node i; P ji,t , Q ji,t is the active and reactive power flowing through branches ji and at time t, both of which are positive if they flow out from node j; g ij and b ij are the conductance and susceptance of branch ij respectively; W ij,t and T ij,t is the variable associated with line "ij"; are the node voltage variables related to nodes i and j, W ij,t and T ij,t is the variable associated with line "ij"; The second-order cone relaxation of equation (36) is: are the node voltage variables related to nodes i and j, W ij,t and T ij,t is the variable associated with line "ij".

6. The charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model according to claim 5 is characterized by: In step 4, the branch virtual terminal voltage variable is introduced to perform variable separation linearization on the non-convex variable product term, as follows: Eliminate the absolute value nonlinear constraint contained in the switching number constraint and introduce auxiliary variables for linearization: And Y in the formula ij is an auxiliary variable, Y ij =1 indicates the switch state changes; On the contrary, it remains unchanged; Indicates the initial state of the switch of branch ij; △Z ij is a variable of 0 or 1, indicating that the branch ij is in the open or closed state; The absolute value of voltage deviation in the objective function of the mathematical model of ADN dynamic reconstruction and reactive and active power joint optimization is calculated by introducing the auxiliary variable U i,t Aux To perform linearization: f V is the power deviation, Represents auxiliary variables, U i,t , U j,t Indicates the "branch voltage" variable corresponding to branch "ij", V th The upper and lower thresholds for the ideal operating voltage.

7. The charging load prediction method based on ADN dynamic reconstruction and active and reactive power joint optimization model according to claim 1 is characterized by: The method further includes step 6, wherein a simulation verification is performed based on a modified IEEE 33-node system example: 1) Based on different input parameters, four different scenarios are modeled and analyzed: weekdays, weekends, high temperature days, and congested days; 2) Compare and analyze the optimization results under different modes to ensure the safe and stable operation of EV time-space load access to ADN; select the congestion day with the largest load fluctuation and analyze the optimization results under four different modes: Mode 1: Only ESS and DG active power optimization is considered; Mode 2: Taking voltage deviation into account based on Mode 1; Mode 3: Based on Mode 2, OLTC, CB, and SVC are added to perform active-reactive power joint optimization; Mode 4: Dynamic network reconstruction is added based on the joint optimization of active and reactive power in Mode 3.

Images