Charging and swapping station energy scheduling method considering seasonal difference

Abstract

The invention discloses a charging and swapping station energy scheduling method considering seasonal difference, and the method comprises the steps: building an EV energy consumption space-time distribution model considering seasonal difference according to EV grid-connected load space-time distribution in multiple scenes, and then building an improved road network model based on a charging and swapping station virtual node; the method comprises the following steps: respectively modeling schedulable domains which can be provided after a battery swap type EV (BSEV) and a plug-in type EV (PEV) enter a station, and establishing a switching model based on the incomplete homogeneity of battery loss of different types of electric vehicles, thereby respectively depicting the schedulable capacity of the battery swap type EV under the condition that the battery swap type EV meets the battery swap service continuous operation constraint and the schedulable capacity of the plug-in type EV; and finally, through a reinforcement learning algorithm, solving the EVA charging service adjustable capability complementary model considering the large-scale EV space-time distribution uncertainty in a multi-dimensional state space to realize the energy optimization scheduling of the charging and charging station.

Description

Technical Field

[0001] The present invention relates to a technology in the field of charging stations, and specifically to an energy scheduling method for charging and swapping stations taking seasonal differences into consideration. Background Art

[0002] Existing research on the adjustability of electric vehicles lacks consideration of the impact of the environment and transportation network on the adjustable capacity of electric vehicles, and does not consider the differences in the dispatchability of different types of electric vehicles, namely plug-in and battery-swap electric vehicles. As a result, there is a lack of practicality of energy scheduling methods and corresponding in-depth analysis and data support. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention proposes an energy scheduling method for charging and swapping stations taking into account seasonal differences. By constructing a method for simulating the spatiotemporal distribution of EV energy consumption taking into account seasonal differences, the influence of environmental and traffic characteristics on the adjustable capacity is taken into consideration, and based on the non-completeness of battery loss of different types of electric vehicles, an electric vehicle responsive switching model is established. A reinforcement learning algorithm is used to construct an EVA charging and swapping service adjustable capacity complementary scheduling model for large-scale EVs with uncertain spatiotemporal distribution, so as to realize EVA scheduling optimization in a multi-dimensional continuous space.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an energy scheduling method for a charging and swapping station considering seasonal differences. After constructing an EV energy consumption spatiotemporal distribution model considering seasonal differences according to the spatiotemporal distribution of EV grid-connected loads in multiple scenarios and establishing an improved road network model based on virtual nodes of charging and swapping stations, the dispatchable domains that can be provided by battery-swap EVs (BSEVs) and plug-in EVs (PEVs) after entering the station are modeled respectively, and a switching model is established based on the incompleteness of battery losses of different types of electric vehicles, thereby respectively characterizing the dispatchable capacity of battery-swap EVs under the constraints of continuous operation of battery-swap services and the dispatchable capacity of plug-in EVs. Finally, a reinforcement learning algorithm is used to solve an EVA charging and swapping service adjustable capacity complementary model considering the uncertainty of spatiotemporal distribution of large-scale EVs in a multi-dimensional state space, thereby realizing energy optimization scheduling of charging and swapping stations.

[0006] The EV energy consumption spatiotemporal distribution model considering seasonal differences is implemented by the following method:

[0007] Step 1.1: Establish road network models for plug-in electric vehicles and battery swap electric vehicles, considering plug-in EV (PEV) and battery swap EV (BSEV) respectively. The two are equal in the travel model, that is, their path finding methods on the road are the same and there is no need to distinguish them during driving simulation, but they are different in the charging / battery swap station: PEV is charged in the station, and its charging amount is continuous and within a certain range, that is, it does not exceed its maximum chargeable amount, but not less than the minimum required amount for travel, and its charging rate is limited, so the charging time increases with the increase of charging amount; corresponding to BSEV, its battery swap amount is a fixed value in the station, so it is discrete and the battery swap time is shorter than that of PEV.

[0008] The traditional modeling does not consider the difference between PEV and BSEV. They are all regarded as BSEV, that is, charging time is not taken into account, so there is no distinction on the road. The traditional road model contains 5 nodes, and multiple road section intersections form a traffic node set V = {1, 2, 3, 4, 5}, and the traffic network contains E = {d 12 ,d 14 ,d 24 ,d 23 ,d 34 ,d 25 ,d 35}Five connected road sections, d corresponds to the energy consumption generated by the path, and its subscript represents the node connected by the road, then the corresponding adjacency matrix This matrix represents the connectivity of each road and corresponds to a shortest path problem.

[0009] Since the present invention distinguishes it into two different EV models, among which: PEV will introduce charging time, resulting in the shortest path problem cannot be solved directly from the adjacency matrix. Considering that the battery swapping and charging station (BSCS) is located at node 5, a virtual node 5' is added. The road connecting the node is one-way, that is, only the value d 5′3 ,d 5′2 , and d 55' is a variable parameter, which makes the difference between the entry of BSEV and PEV reflected here. From then on, the adjacency matrix of the traditional road is transformed into the energy consumption adjacency matrix The matrix corresponds to a minimum energy consumption problem. So far, the traditional shortest path problem is transformed into a minimum energy consumption problem by adding virtual nodes.

[0010] Step 1.2: Consider the impact of seasonal differences on EV energy consumption, and then construct a seasonal energy consumption adjacency matrix: First, consider the direct effect of seasonal differences on EV energy consumption. Since the on-board electrical equipment is affected by the season, its energy consumption under different conditions in the four seasons will be different. Based on this, the impact of seasonal changes on vehicle driving energy consumption is depicted as follows: Where: p is the probability matrix of vehicle travel. The road network obtained in step 1.1 is further divided into four areas: H, S, PS, and W. H represents the civil area, S represents the business area, PS represents the public area, and W represents the industrial area. 0≤p ij ≤1, Furthermore, using the normal distribution to express whether the air conditioning service is turned on, with temperature as the main variable, we have in: The air conditioning energy consumption at ambient temperature t is f3(t) = a4t 4 +a3t 3 +a2t 2 +a1t+a0, where: a4=-0.000117, a3=0.004853, a2=0.195181, a1=-8.799674, a0=85.353171. At this point, the energy consumption adjacency matrix is ​​updated to D”(G), which is specifically expressed as: D”(G)=D'(G)*f2(t)*f3(t).

[0011] In addition to having a direct impact on whether the on-board electrical equipment is turned on or not, seasonal differences also have a greater impact on battery capacity, which in turn indirectly affects the charging / replacing of EVs and driving models. The present invention refers to actual rules and models it as follows: when the ambient temperature is high, the impact on battery capacity is so small that it can be ignored, and when the ambient temperature drops to a certain value, the battery is greatly affected by the ambient temperature, which is reflected in the lower actual usable capacity, faster battery power consumption, and more frequent charging / replacing. Using a polynomial fitting method, the relationship between temperature and battery is specifically modeled as follows: the battery capacity ratio at temperature t is f1(t) = a7x 7 +a6x 6 +a5x 5 +a4x 4 +a3x 3 +a2x 2 +a1x+a0, where: a0=79.6359, a1=1.3276, a2=-5.708×10 -3 , a3=-7.6530×10 -4 , a4=-2.5938×10 -5 , a5=1.9589×10 -6 , a6=-3.4166×10 -8, a7=1.8956×10 -10 .

[0012] The energy consumption adjacency matrix that finally takes into account seasonal factors is D"'(G), which can be specifically expressed as: D"'(G)=D"(G)*f1(t).

[0013] Step 1.3: Considering the impact of seasonal factors on the traffic network, a time-varying traffic flow statistical speed-flow model based on seasonal changes is constructed, specifically: Where: v ij,0 is the rated speed of R(i,j), that is, the driving speed under zero traffic flow; q C is the road saturation, that is, the percentage of the road's real-time traffic volume to its maximum traffic volume; β is the statistical coefficient, which can also be obtained from the expert database; a, b, and c are the statistical coefficients corresponding to different road characteristics. tp is the air conditioning power consumption of EV passing through road R(i,j) at temperature t; l ij is the average travel length of road R(i,j); E cost (i,j,w) is the total power consumption of EV passing through road R(i,j) at temperature t, E ct is the power consumption per unit mileage of EV at temperature t.

[0014] Step 1.4: Solve the EV traffic network driving path considering the influence of seasonal factors, specifically including: generating basic vehicle parameters of different types of EVs, including vehicle battery capacity information, battery safe working area, initial power information, travel time, travel section, etc.; randomly extracting road network nodes within the study area according to the traffic network information, generating an adjacency matrix based on different EV vehicle information, and generating the corresponding energy consumption adjacency matrix according to the above steps 1.1-1.2, converting the original shortest path problem into a minimum energy consumption path problem, using the total energy consumption obtained in step 1.3, and using the Dijkstra algorithm to find the minimum energy consumption path in the study scenario, and based on this minimum energy consumption path, calculate the total power consumption and update the battery power information; when the path passes through BSCS, the travel time is not included for BSEV, and is included for PEV, the travel time on the path is calculated by the travel speed obtained in step 1.3, the travel time is added, and finally the entry time t is calculated according to the path. start , departure time t end , so far we have obtained the complete EV travel situation.

[0015] The characterization specifically includes:

[0016] Step 2.1: Model the energy dispatch boundary of PEV, starting from the energy E when PEV enters the station start , average power when leaving the station E exp、The minimum amount of electricity required for the next travel demand E ms , Entry time t start , average exit time t end These five dimensions are considered to construct the state vector S evcs =[E start ,E exp ,E ms ,t start ,t end ], and based on this, the PEV energy dispatch boundary is established. Considering that different PEVs enter the station, due to differences in their own properties such as battery energy efficiency, battery charging and discharging capabilities, and driving time, the energy dispatch boundary changes, and they are divided into three modes: Mode I, Mode II, and Mode III. The specific modeling is as follows.

[0017] PEV Mode I: After entering the charging / swapping station, PEV does not participate in discharge scheduling, that is, general charging. When scheduling its energy, its charging time position can only be changed within a certain time range. Furthermore, its charging power cannot be changed during scheduling. Therefore, its boundary during energy scheduling is modeled as: in: is the charging power of vehicle j at time t; is a 0-1 variable representing whether the PEV is in a charging state; The upper limit of the PEV's own charging power; is the battery capacity of PEV at time t; ab is the time position of the controllable charging of the PEV; η L is the charging efficiency; the charging process is discretized and the charging period is divided into two parts according to the time distance of Δt. ms How long does it take to charge a PEV?

[0018] PEV Mode II: PEV does not discharge after entering the charging / swapping station, but its charging power can be adjusted during energy scheduling, and the time t for it to leave the charging / swapping station can be set. end , so its boundary during energy scheduling is modeled as: in: is the intermediate variable, and the remaining variables are the same as in the above-mentioned mode I.

[0019] PEV Mode III: After the PEV enters the charging / swapping station, its charging / discharging state can be set during energy scheduling, and its charging / discharging specific power can be set, and the time t for it to leave the charging / swapping station can be setend , so its boundary during energy scheduling is modeled as: The variables are the same as in Mode II above.

[0020] Step 2.2: Based on the three PEV modes divided above, solve the overall dispatchability of the cluster PEV. Preferably, considering that the battery life is related to the number of battery charge and discharge times, the number of charge times N is usually calculated in the application scenario. ch,j and discharge times N dis,j

[0021] Restrictions are imposed, specifically: Where: ch,j (t) = 0 means that the EV is in a non-charging state, Γ ch,j (t) = 1, indicating that the EV is in the charging state; N ch,j ,N dis,j is the number of charge / discharge times, and the specific modeling of mode i is as described above.

[0022] Solve the charging power of PEV in different modes after entering the charging / swapping station. Generally, when PEV enters the charging / swapping station for charging, the minimum charging cost will be the main trend of its charging situation change, that is, when it is not dispatched, the charging method with the minimum charging cost will be given priority as the change mode of its charging situation, including charging power, specifically: in: is the charging power of PEV in mode i; α c (t) is the charging consumption at time t; represents the volatility statistical constant under mode i; The consumption of battery charging and discharging. For the three modes divided above, the battery loss in different modes is different, so there is Γ ch,j (t) = 0 means that the PEV is in a non-charging state, Γ ch,j (t) = 1, indicating that the PEV is in a charging state; Γ dis,j (t) = 0 means that the PEV is in a non-discharging state, Γ dis,j (t) = 1, indicating that the PEV is in a discharge state.

[0023] For the PEV cluster, the mode it is in after entering the station is not fixed, and there is mode switching. The present invention describes the mode switching probability based on the Logistic function, considering that the three modes can be converted to each other, specifically: ; in: represents the probability that mode i does not switch modes; is the variable parameter of the Logistic function, a sw is the probability parameter of PEV switching mode in three modes, For variable parameters, Same as above.

[0024] Solving the dispatchability of cluster PEVs Specifically: in: is the maximum discharge capacity in the tth period, It is the normalized index of the discharge capacity of EV under the current working condition, and the other parameters are obtained by the above solutions.

[0025] Step 2.3: Model the BSEV energy dispatch boundary. Different from PEV, the energy E start , Average power consumption when leaving the station E exp , Entry time t start , average exit time t end Consider it in four dimensions and construct the state vector S evcs =[E start ,E exp ,E ms ,t start ,t end ], and based on this, the PEV energy scheduling boundary is established.

[0026] Based on the characteristics of BSEV, it can be known that after entering the charging / swapping station, the battery will only be fully charged, but the starting time of the battery swap can be scheduled, and the battery swap time is approximately negligible. Therefore, the BSEV battery swap state constraints can be further established, which are as follows: 0≤N m,c,t ≤N m,s,t-1 ,m∈1,2,...,M-1, Where: N all,c,t N represents the total number of batteries that need to be replaced in the station at time t. max,sw Indicates the number of battery swapping interfaces in the station, ΔS OC,max The difference between the battery capacity after replacement and the original capacity. 1,s,t, N 2,s,t ...N M,s,t The corresponding time periods are in the interval [S OC,0 ,S OC,1 ]; [S OC,1 ,S OC,2 ]……[S OC,M-1 ,S OC,M ] is the number of batteries in the system. Where: N 1,c,t ,N 2,c,t ...N M,c,tis the number of batteries in the charging state in the corresponding power interval, N 1,d,t ,N 2,d,t ...N M,d,t It is the number of batteries in the discharge state in the corresponding power range.

[0027] Step 2.4: Solve the overall dispatchable energy of the BSEV cluster, that is, solve its overall operating power boundary under the BSEV operation constraints.

[0028] First, the battery replacement situation of BSEV after entering the station is solved, that is, the number of battery replacement batteries, with the minimum number of remaining replaceable spare batteries as the solution target, specifically: 0≤x(t)≤N max,sw , 0≤x′(t)≤D(t), 0≤y(t)≤y(0), y(t)=y(0)-a(t)-b(t),t∈(0,T], Where: x(t) is the number of batteries that have been charged since time t; x'(t) is the number of batteries that have just been fully charged at time t, T need is the full charge time; y(t) is the number of remaining replaceable spare batteries; a(t) is the cumulative number of batteries in charge; b(t) is the number of delayed charging batteries; D(t) is the amount of replacement. Based on the above solution, the scheduling boundary of BSEV can be further obtained, which is:

[0029] Then the energy scheduling boundary of the BSEV cluster is obtained: Specifically:

[0030] The reinforcement learning algorithm is: according to the EV spatiotemporal distribution considering seasonal effects and the energy scheduling boundaries of different types of EV clusters, the optimal energy scheduling method is solved using the deep deterministic policy gradient network (DDPG) algorithm, whose state space is obtained by the EV spatiotemporal distribution, and the action set is obtained by the energy scheduling boundary. The reward is set as the charging / swapping station income, and its different policy gradient update method is: Where: Q μ (s,a) is the action value function during training, whose input is the information of action and state, and output is the Q value under the corresponding strategy θ; an independent network is established to update the loss, and the loss is calculated as: ,y k =r k +γQ′(s k+1 ,μ′(s k+1 ∣θ μ′ )|θ Q′ ), where: t For status, a t For action, r tFind the best action for the return Technical Effects

[0031] The present invention incorporates the impact of environmental factors on user travel into the road network model by simulating the spatiotemporal distribution of EV energy consumption with seasonal characteristics, and establishes an improved road network model based on the "virtual nodes" of battery swapping and charging stations, converting the shortest path problem into the minimum energy consumption path problem, and simulating the travel of battery swapping EVs and plug-in EVs. The dispatchable capacity of large-scale plug-in EVs in BSCS is characterized to obtain the dispatchable capacity provided by the charging / discharging service for EVA under seasonal differences. Based on the different battery states when the battery swapping EV enters the station to replace the battery, the inbound battery intervals are divided, and a method for characterizing the dispatchable capacity under the continuous operation constraints of the battery swapping service is established. Compared with the prior art, the present invention can accurately capture the dynamic characteristics of the interaction between the battery swapping station and the power grid. The dispatching capacity of the battery swapping station with significant seasonal and locational differences is obtained, and the seasonal complementarity of the dispatching capacity of the battery swapping station is confirmed through data analysis. This discovery is of great significance for optimizing the optimal strategy for battery swapping stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the present invention;

[0033] Figure 2 It is a schematic diagram of the energy consumption adjacency matrix;

[0034] Figure 3 This is a schematic diagram of the relationship between the probability of air conditioning being turned on and its energy consumption;

[0035] Figure 4 Schematic diagram of the dispatchable capacity of PSEV under various grid access modes;

[0036] Figure 5 This is a schematic diagram of BSEV dispatchability;

[0037] Figure 6 This is the DDPG flow chart;

[0038] Figure 7 This is a schematic diagram of a 29-node road network model;

[0039] Figure 8 This is a schematic diagram of the number of vehicles entering each charging station;

[0040] Fig. 9 This is a schematic diagram of the inbound power distribution at stations #26 and #6;

[0041] Fig.10 Schematic diagram of dispatchable capacity for charging / discharging services of stations #26 and #6 in spring and summer;

[0042] Fig.11This is a schematic diagram of the dispatchable capacity of the power swapping service of stations #26 and #6 in spring and summer;

[0043] Fig.12 Schematic diagram of the bidding capacity of EVA in the electric energy market for spring and summer operation;

[0044] Fig.13 This is a schematic diagram of EVA's winning bid capacity in the reserve capacity market during spring and summer operations;

[0045] Fig.14 Schematic diagram of the air conditioner turning on probability curve. DETAILED DESCRIPTION

[0046] like Figure 1 As shown, a charging and swapping station energy scheduling method considering seasonal differences involved in this embodiment is constructed. According to the spatiotemporal distribution of EV grid-connected load in multiple scenarios, a spatiotemporal distribution model of EV energy consumption considering seasonal differences is constructed, and an improved road network model based on virtual nodes of charging and swapping stations is established based on this. Then, the dispatchable domains that can be provided by battery-swap EV (BSEV) and plug-in EV (PEV) after entering the station are modeled respectively, and a switching model is established based on the incompleteness of battery loss of different types of electric vehicles, so as to characterize the dispatchable capacity of battery-swap EV under the continuous operation constraint of battery-swap service and the dispatchable capacity of plug-in EV respectively. Finally, through the reinforcement learning algorithm, the EVA charging service adjustable capacity complementary model considering the uncertainty of spatiotemporal distribution of large-scale EV is solved in the multi-dimensional state space to realize the energy optimization scheduling of charging and swapping stations.

[0047] After specific experiments, based on the relevant road network data provided by OpenStreetMap, Shanghai was selected as the research background, and a certain area in Shanghai was selected as the simulation platform for EV travel. The road network model has 118 nodes connected by 9 charging stations. Different nodes correspond to different functional areas, among which: nodes 1-11 are H areas, 12-16 are PS areas, 17-20 are W areas, and 21-29 are S areas. Their distribution is as follows Figure 7 shown.

[0048] The proposed EVA operation model and its solution algorithm are analyzed by using the IEEE-118 node standard test system. Three units are used as other dispatching entities. The overall system reserve capacity is 10% of the overall system load. The dispatch ratio of positive and negative reserve capacity is characterized by a random value falling within [-1,1] to characterize its uncertainty. Negative values ​​are used to call negative reserve capacity, and positive values ​​are used to call positive reserve capacity. The EVA operation is analyzed with the background that EVA and traditional units participate in the dispatch balance on an equal footing.

[0049] This embodiment selects the driving probability matrix, ambient temperature and other data of four typical days in spring, summer, autumn and winter. The total number of simulated vehicles is N EV=2000, of which: NIO ET5 is BSEV accounting for 15% of the total number of vehicles, and the remaining 85% are PEVs, among which the average vehicle power when leaving the station is E exp The values ​​are 25%, 50%, and 75%, with a ratio of 5:3:2; the initial power of the vehicle is a random number between [0.2, 0.85]; the vehicle is charged and swapped at the charging and swapping station when the power is between [0.2, 0.4].

[0050] like Figure 8 As shown in the figure, the inbound vehicle data of each battery swap station on a typical day in different seasons is analyzed, and the figure shows that there are significant differences in the traffic volume between each battery swap station. At the same time, the overall difference in the number of inbound vehicles at most battery swap stations with seasonal changes is not significant. Considering the influence of geographical factors, the number of inbound vehicles at battery swap stations #24 and #26 located in the downtown business district and #13 and #19 in the production area is greater than that at charging stations in other areas. That is, different locations in different geographical locations will affect the difference in traffic volume, and this difference also represents the adjustable potential of each battery swap station.

[0051] Select the inbound data of two charging stations: Fig. 9 As shown in the figure, the battery level of vehicles entering the station #26 in the commercial area is densely distributed between [0.2, 0.5] in the four seasons. The difference across seasons is not obvious, and the data change trend is the same. The stability of the cross-seasonal distribution of incoming power also means the stability of its dispatching ability. The data of station #6 located in the residential area has obvious seasonal fluctuations. In summer and winter, when the temperature has the greatest impact on the battery, the incoming power peaks at [0, 0.1]. The lower the incoming power, the higher the charging power will be, and the corresponding adjustable capacity will be stronger. Therefore, the adjustable capacity of station #6 in winter and summer is more abundant than that in the other two seasons.

[0052] like Fig.10As shown in the figure, the power swapping capacity of station #26 and station #6 is concentrated in working hours, while the power swapping capacity of station #6 located in the residential area is basically less than that of station #26, and is evenly distributed during the day. The daily power swapping capacity of station #26 has a significant shift from spring to summer, with the peak value increasing backwards, while winter is similar to spring, with the peak value shifting forward. Compared with the power when EV enters the station, the time of EV entry has a greater impact on the adjustable capacity. The power swapping capacity of station #26 varies significantly across seasons, while the seasonal variation of station #6 is not significant. Therefore, the cross-seasonal stability of the adjustable capacity of direct power swapping between #26 and #6 is different. In each season, how to make full use of the adjustable capacity advantages of charging stations in different regions is a key factor that EVA must pay attention to when performing energy scheduling. Based on this, the proposed electric vehicle charging and swapping capacity model can accurately simulate its distribution characteristics while taking into account the impact of the changing seasons on the uncertainty of the spatiotemporal distribution of electric vehicles, thereby providing solid data support for evaluating the scheduling capacity of each power swap station.

[0053] Based on the simulation data of the previous stage, the dispatchable capacity of each BSEV is calculated. Considering three modes of charging / discharging service They are [5, 10, 15] respectively, corresponding to the battery loss of [0.08, 0.088, 0.094] yuan / time. The initial mode of PEV access to the network is considered to be random and equal, and the ratio of the number of vehicles in the three modes is 1:1:1; the charging / discharging price refers to the data of the charging station, and the price of battery replacement is 1.5￥ / kWh, Take 0.5; the total number of spare batteries in the station is 100, and the number of battery swapping interfaces is 30, that is, a maximum of 30 BSEVs can be swapped at the same time. Considering the scheduling continuity, the total number of spare batteries available in the day is not less than 75. The sampling frequency of inbound data is 15 minutes. OC,1 ,S OC，2 ] Charge to the next interval [S OC，2 ,S OC，3 ]The time required T need =45min. Negative reserve is represented by negative value.

[0054] like Fig.11 As shown in the figure, the change in the dispatchable capacity that stations #26 and #6 can provide for their charging / discharging services as the seasons rotate. Considering the influence of ambient temperature, the dispatchable domain of station #26 changes little, and the total amplitude decreases; while the opposite is true for station #6, the shape of the dispatchable domain changes greatly, but the average remains unchanged. In addition, there is also the influence of geographical factors. Compared with station #6 in the residential area, station #26 in the commercial area has a more abundant adjustable domain in the second half of the day, while station #6 has a more abundant adjustable domain in the early morning, which is consistent with the actual distribution.

[0055] like Fig.12As shown in the figure, the swapping dispatchable capacity of stations #26 and #6 is different from that of charging / discharging. The swapping dispatchable capacity of station #26 is less affected by the season, and the corresponding dispatchable domain shape and amplitude remain basically the same; while the dispatchable capacity amplitude of station #6 is larger overall. This corresponds to the complementary fluctuation of the adjustable capacity caused by the season in charging / discharging, which can smooth the seasonal uncertainty of EV.

[0056] In this embodiment, the maximum number of iterations in each period is 100, and the neural network parameters are updated every 5 iterations. The batch_size of the experience replay set is 10. The discount factor γ is 0.9. The cross entropy weight is 0.01. The actor network learning rate is 0.009, and the critic network learning rate is 0.02. The simulation environment is: Intel Core i5-8600K CPU @ 4.30 GHz, running memory (RAM) 32 GB, software configuration is python 3.8.7, and the neural network is built using pytorch.

[0057] Considering the timeliness of EVA scheduling in actual application, based on the above adjustable capacity quantification results, a total of 96 time nodes in 24 hours a day are simulated to analyze the optimal energy scheduling strategy of EVA.

[0058] like Fig.13 As shown in the figure, the winning bid power and load power of EVA. In the power balance in spring and summer, traditional units undertake the main load balancing tasks. The total output of the three traditional units in spring and summer accounts for 63.37% and 67.51% of the total. However, EVA accounts for a small proportion in load balancing. The concentrated output of the nine charging stations in the season accounts for the largest proportion of 37.63% in spring, indicating that the adjustable capacity of the charging stations is higher in spring.

[0059] In addition, EVs in the commercial area provide relatively stable dispatchable capabilities in spring and autumn. EVA selects stations #23, #24, #21 and #26 in the commercial area for priority output, accounting for 58.25% and 65.06% in spring and autumn respectively; while in summer and winter, BSEVs in other areas provide more stable dispatchable capabilities, accounting for 63.51% and 61.38% of the output respectively.

[0060] Considering the seasonal response capability, the output of EVA in each period is significantly higher than that of traditional units, such as Fig.14 As shown in the figure, the output of EVA in spring and summer accounts for 58.32% and 69.60% of the total respectively, which shows that responsiveness is an important attribute. Given that the dispatching capacity of battery swap stations in commercial areas is more stable during spring and autumn, while the dispatching capacity of battery swap stations in other areas is insufficient to cope with the peak of the first backup demand. Correspondingly, in summer and winter, EVA energy dispatch is mainly concentrated in the afternoon period with high dispatch margin.

[0061] The operating statistics of station #23 in the commercial area in EVA in four seasons are selected to analyze its cost / benefit situation. The cross-seasonal station #23 has increased its revenue when participating in the market, but due to the influence of distribution and mode switching uncertainty, the adjustable capacity it provides to EVA in summer is 43.34% lower than that in spring, and the revenue from participating in the market in summer is significantly lower than that in other seasons; while station #17 in the working area has relatively stable charging capacity over a long period of time, which means that the total adjustable capacity in each season is also stable. The operation of this station is analyzed, and the complementary adjustable capacity can better smooth the seasonal risk.

[0062] Compared with the prior art, the present invention comprehensively considers the seasonal characteristics of the difference in the temporal and spatial distribution simulation method of electric vehicle energy consumption. The method incorporates the influence of environmental factors into the road network model, and innovatively improves the road network model based on the "virtual node" of the battery swap station. Through this improvement, the traditional shortest path problem is transformed into the minimum energy consumption path problem, so that the travel mode of battery swap and plug-in electric vehicles can be accurately simulated. According to the influence of battery loss and air conditioning temperature, a travel mode switching model is constructed to characterize the dispatchable capacity of large-scale plug-in EVs in BSCS. By considering seasonal changes, the dispatchable capacity provided by charging / discharging services to electric vehicle aggregators is determined, thereby providing electric vehicle aggregators with a more accurate dispatchable capacity forecast. A complementary participation model of the adjustable capacity of electric vehicle aggregators' charging and swap services in the market is constructed considering the uncertainty of the spatiotemporal distribution of large-scale electric vehicles. The model is solved in the multi-dimensional continuous quotation and market state space to realize the optimization strategy of electric vehicle aggregators in the multi-dimensional continuous strategy space, and achieve more efficient resource allocation and scheduling optimization.

[0063] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for energy scheduling of charging and swapping stations considering seasonal differences, characterized in that: According to the spatiotemporal distribution of EV grid-connected load in multiple scenarios, an EV energy consumption spatiotemporal distribution model considering seasonal differences is constructed, and based on this, an improved road network model based on virtual nodes of charging and swapping stations is established. Then, the dispatchable domains that can be provided by battery-swap EVs (BSEVs) and plug-in EVs (PEVs) after entering the station are modeled respectively, and a switching model is established based on the incompleteness of battery loss of different types of electric vehicles. In this way, the dispatchable capacity of battery-swap EVs under the continuous operation constraints of battery-swap services and the dispatchable capacity of plug-in EVs are characterized respectively. Finally, through the reinforcement learning algorithm, the EVA charging service adjustable capacity complementary model considering the uncertainty of spatiotemporal distribution of large-scale EVs is solved in the multi-dimensional state space to achieve energy optimization scheduling of charging and swapping stations.

2. The energy dispatching method of a charging and swapping station considering seasonal differences according to claim 1 is characterized in that: The EV energy consumption spatiotemporal distribution model considering seasonal differences is implemented by the following method: Step 1.1: Establish the road network model for plug-in electric vehicles and battery-swap electric vehicles, specifically: Among them, considering the virtual node parameter d 55' is the variable parameter that differentiates plug-in electric vehicles from battery-swap electric vehicles; Step 1.2: Construct the seasonal difference energy consumption adjacency matrix, specifically: D"'(G) = D" (G) * f1(t), where D" (G) = D'(G) * f2(t) * f3(t), f1(t) = a7x 7 +a6x 6 +a5x 5 +a4x 4 +a3x 3 +a2x 2 +a1x+a0, f3(t)=a4t 4 +a3t 3 +a2t 2 +a1t+a0; Step 1.3: Construct a time-varying traffic flow statistical speed-flow model based on seasonal changes, specifically: Where: v ij,0 is the rated speed of R(i,j), that is, the driving speed under zero traffic flow; q C is the road saturation, that is, the percentage of the road's real-time traffic volume to its maximum traffic volume; β is the statistical coefficient, which can also be obtained from the expert database; a, b, c are the statistical coefficients corresponding to different road characteristics, E tp is the air conditioning power consumption of EV passing through road R(i,j) at temperature t; l ij is the average travel length of road R(i,j); E cost (i,j,w) is the total power consumption of EV passing through road R(i,j) at temperature t, E ct is the power consumption per unit mileage of EV at temperature t; Step 1.4: Solve the EV traffic network driving path considering seasonal factors: Generate vehicle basic parameters of different types of EVs, including vehicle battery capacity information, battery safe working area, initial power information, travel time, travel section, etc.; Randomly extract road network nodes within the study area based on traffic network information, generate an adjacency matrix based on different EV vehicle information, and generate the corresponding energy consumption adjacency matrix by steps 1.1-1.2, transform the original shortest path problem into the minimum energy consumption path problem, and use the Dijkstra algorithm to find the minimum energy consumption path in the study scenario based on the total energy consumption obtained in step 1.3, and calculate the total power consumption based on this minimum energy consumption path, and update the battery power information; When the path passes through BSCS, the travel time is not included for BSEV, and the travel time is included for PEV. The travel time on the path is calculated by the travel speed obtained in step 1.3, and the travel time is added. Finally, the entry time t is calculated according to the path. start , departure time t end , so far we have obtained the complete EV travel situation.

3. The energy dispatching method of charging and swapping stations considering seasonal differences according to claim 1 is characterized in that: The characterization specifically includes: Step 2.1: Model the energy dispatch boundary of PEV, starting from the energy E when PEV enters the station start , Average power consumption when leaving the station E exp 、The minimum amount of electricity required for the next travel demand E ms , Entry time t start , average exit time t end These five dimensions are considered to construct the state vector S evcs =[E start ,E exp ,E ms ,t start ,t end ], based on which the PEV energy dispatch boundary is established; Considering that different PEVs enter the station, their energy scheduling boundaries change due to differences in their own properties such as battery energy efficiency, battery charging and discharging capabilities, and driving time, and they are divided into three PEV modes accordingly; Step 2.2: Solve the overall dispatchability of cluster PEVs based on the PEV model, specifically: Where: ch,j (t) = 0 means that the EV is in a non-charging state, Γ ch,j (t) = 1, indicating that the EV is in the charging state; N ch,j ,N dis,j is the number of charge / discharge times, mode i is the PEV mode; Solve the charging power of PEV in different modes after entering the charging / swapping station, specifically: in: is the charging power of PEV in mode i; α c (t) is the charging consumption at time t; represents the volatility statistical constant under mode i; The consumption of battery charging and discharging. For the three modes divided above, the battery loss in different modes is different, so there is Γ ch,j (t) = 0 means that the PEV is in a non-charging state, Γ ch,j (t) = 1, indicating that the PEV is in a charging state; Γ dis,j (t) = 0 means that the PEV is in a non-discharging state, Γ dis,j (t) = 1, indicating that the PEV is in the discharge state; Step 2.3: Model the BSEV energy dispatch boundary: The energy E when the BSEV enters the station start , Average power consumption when leaving the station E exp , Entry time t start , average exit time t end Consider it in four dimensions and construct the state vector S evcs =[E start ,E exp ,E ms ,t start ,t end ], based on which the PEV energy dispatch boundary is established; further the BSEV battery swap state constraints are established, specifically: , 0≤N m,c,t ≤N m,s,t-1 ,m∈1,2,...,M-1, Where: N all,c,t N represents the total number of batteries that need to be replaced in the station at time t. max,sw Indicates the number of battery swapping interfaces in the station, ΔS OC,max The difference between the battery capacity after replacement and the original capacity, N 1,s,t, N 2,s, t ...N M,s,t The corresponding time periods are in the interval [S OC,0, S OC,1 ]; [S OC,1, S OC,2 ]……[S OC,M-1 ,S OC,M ], where: N 1,c,t ,N 2,c,t ...N M,c,t is the number of batteries in the charging state in the corresponding power interval, N 1,d,t ,N 2,d,t …N M,d,t is the number of batteries in the discharge state in the corresponding power interval; Step 2.4: Solve the overall dispatchable energy of the BSEV cluster, that is, solve its overall operating power boundary under the BSEV operation constraints, specifically including: first solve the battery replacement situation after the BSEV enters the station, that is, the number of battery replacement batteries, and take the minimum remaining number of replaceable spare batteries as the solution target, specifically: min∑ t∈T y(t), 0≤x(t)≤N max,sw , 0≤x′(t)≤D(t), 0≤y(t)≤y(0), y(t)=y(0)-a(t)-b(t),t∈(0,T], Where: x(t) is the number of batteries that have been charged since time t; x'(t) is the number of batteries that have just been fully charged at time t, T need is the full charging time; y(t) is the number of remaining replaceable spare batteries; a(t) is the cumulative number of batteries in charge; b(t) is the number of delayed charging batteries; D(t) is the amount of replacement power; the scheduling boundary of BSEV can be further obtained, which is: Solve to obtain the energy scheduling boundary of the BSEV cluster Specifically:

4. The energy dispatching method for charging and swapping stations considering seasonal differences according to claim 3 is characterized in that: The three PEV modes include: PEV Mode I: After entering the charging / swapping station, PEV does not participate in discharge scheduling, that is, general charging. When scheduling its energy, its charging time position can only be changed within a certain time range. Furthermore, its charging power cannot be changed during scheduling. Therefore, its boundary during energy scheduling is modeled as: in: is the charging power of vehicle j at time t; is a 0-1 variable representing whether the PEV is in a charging state; The upper limit of the PEV's own charging power; is the battery capacity of PEV at time t; ab is the time position of the controllable charging of the PEV; η L is the charging efficiency; the charging process is discretized and the charging period is divided into two parts according to the time distance of Δt. ms How long does it take to charge the PEV? PEV Mode II: PEV does not discharge after entering the charging / swapping station, but adjusts its charging power during energy scheduling and sets the time t for it to leave the charging / swapping station. en d, so its boundary during energy scheduling is modeled as: in: is the intermediate variable, and the other variables are the same as those in the above-mentioned mode I; PEV Mode III: After the PEV enters the charging / swapping station, its charging / discharging state can be set during energy scheduling, and its specific charging / discharging power can be set, and the time t for it to leave the charging / swapping station can be set. en d, so its boundary during energy scheduling is modeled as: , The variables are the same as in Mode II above.

5. The energy dispatching method of a charging and swapping station considering seasonal differences according to claim 3 is characterized in that: For the PEV cluster, the mode it is in after entering the station is not fixed, and there is mode switching. The present invention describes the mode switching probability based on the Logistic function, considering that the three modes can be converted to each other, specifically: , in: represents the probability that mode i does not switch modes; is the variable parameter of the Logistic function, a sw is the probability parameter of PEV switching mode in three modes, For variable parameters, Same as above; Solving the dispatchability of cluster PEVs Specifically: in: is the maximum discharge capacity in the tth period, It is the normalized index of the dischargeable capacity of EV under the current working condition, and the other parameters are obtained by the above solutions.

6. The energy dispatching method of a charging and swapping station considering seasonal differences according to claim 1 is characterized in that: The reinforcement learning algorithm is: according to the solved EV spatiotemporal distribution considering seasonal influence and the energy scheduling boundary of different types of EV clusters, the optimal energy scheduling method is solved using the DDPG algorithm, wherein the state space is obtained by the EV spatiotemporal distribution, the action set is obtained by the energy scheduling boundary, and the reward is set as the charging / swapping station income; The different policy gradient update method is: Where: Q μ (s,a) is the action value function during training, whose input is the information of action and state, and output is the Q value under the corresponding strategy θ; an independent network is established to update the loss, and the loss is calculated as: y k =r k +γQ′(s k+1 ,μ′(s k+1 ∣θ μ′ )|θ Q′ ), where: t For status, a t For action, r t Find the best action for the return

Images