Day-ahead response capability evaluation method, scheduling optimization method and equipment of electric vehicle virtual power plant

Abstract

The invention discloses a day-ahead response capability evaluation method, a scheduling optimization method and equipment for an electric vehicle virtual power plant, and relates to the technical field of smart power grids. The method is based on a data driving thought, adopts an LSTM-GA combination method, captures a time dependency relationship in massive historical operation data of a virtual power plant, realizes accurate identification of non-fixed characteristic parameters of an EV monomer, and comprises the following steps: firstly, taking an EV group (namely an electric vehicle virtual power plant) as an evaluation object, and evaluating the day-ahead response capability of the electric vehicle virtual power plant; the accuracy of day-ahead response capability evaluation is improved, and then the accuracy of optimization scheduling is improved.

Description

Technical Field

[0001] The present application relates to the field of smart grid technology, and in particular to a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant, a scheduling optimization method and equipment. Background Art

[0002] Electric vehicles (EVs) have developed rapidly in recent years due to their advantages of energy saving and zero emissions. EV batteries have flexible bidirectional power regulation characteristics, can accept dispatch control to change their own charging and discharging states, and participate in the demand response of the power system. Electric vehicle virtual power plants (EVPPs) are responsible for aggregating EVs in a certain area and managing them in a unified manner, thus providing a way for decentralized and small-capacity EV units to participate in the optimal dispatch of the power system. The common way for EVPPs to participate in regulation is to respond to system dispatch requirements and provide power support. Accurately evaluating the day-ahead response capability of EVPPs has become the key to serving the power grid and EV users. Existing studies generally give the probability distribution of EV unit response characteristic parameters based on statistical experience, and then use the Monte Carlo method for random sampling. The response capabilities of the sampled EV units are calculated one by one and then added together to obtain the final EVPP response capability. However, the probability distribution describes the generality of the characteristic parameters, and cannot truly reflect the differences in the parameters of the EV group in actual scenarios, making it difficult to accurately calculate the response capability. Summary of the invention

[0003] The purpose of this application is to provide a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant, a scheduling optimization method and equipment, so as to improve the accuracy of the day-ahead response capability evaluation and thereby improve the accuracy of the optimized scheduling.

[0004] To achieve the above objectives, this application provides the following solutions.

[0005] In a first aspect, the present application provides a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant, comprising:

[0006] Based on the historical operating data of each EV unit in the electric vehicle virtual power plant, the average calculation method is adopted to obtain the predicted value of the fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day; the fixed characteristic parameters include: the allowable upper limit of the battery state of charge, the allowable lower limit of the battery state of charge and the battery capacity.

[0007] Based on the historical operation data of each EV unit in the electric vehicle virtual power plant, the LSTM-GA combination method is adopted to obtain the predicted values ​​of the non-fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day; the non-fixed characteristic parameters include: grid-connected charge state, minimum off-grid power, grid-connected time and off-grid time.

[0008] According to the predicted values ​​of various characteristic parameters of each EV unit in the electric vehicle virtual power plant on the prediction day, each EV unit in the electric vehicle virtual power plant is clustered to obtain multiple clusters; each characteristic parameter includes various fixed characteristic parameters and various non-fixed characteristic parameters.

[0009] The aggregate expression of each characteristic parameter in each cluster is calculated separately.

[0010] According to the aggregated expression of each characteristic parameter in each cluster, the upper and lower limits of the day-ahead response capability of the EV virtual power plant are calculated.

[0011] In a second aspect, the present application provides a scheduling optimization method for an electric vehicle virtual power plant, comprising:

[0012] An optimization model is constructed with the goal of optimizing the tracking effect of the actual response power on the dispatch demand signal and minimizing the amount of wind and solar power curtailment.

[0013] Based on the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant, the constraints of the optimization model are determined; the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant are obtained using the above-mentioned day-ahead response capability evaluation method of the electric vehicle virtual power plant.

[0014] Based on the constraints, the optimization model is solved to obtain the optimal scheduling solution.

[0015] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for evaluating the day-ahead response capability of an electric vehicle virtual power plant, or the above-mentioned method for optimizing the scheduling of an electric vehicle virtual power plant.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects.

[0017] The present application provides a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant, a scheduling optimization method and a device. The present application is based on the data-driven concept and adopts the LSTM-GA combination method to capture the time dependency in the massive historical operation data of the virtual power plant and realize the accurate identification of the non-fixed characteristic parameters of EV monomers. Firstly, the EV group (i.e., the electric vehicle virtual power plant) is taken as the evaluation object to evaluate the day-ahead response capability of the electric vehicle virtual power plant, thereby improving the accuracy of the day-ahead response capability evaluation and thus improving the accuracy of the optimized scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flow chart of a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant is provided for one embodiment of the present application.

[0020] Figure 2 A schematic diagram of the maximum controllable area for EV cell access to the network provided in one embodiment of the present application.

[0021] Figure 3 A curve diagram showing the relationship between the number of cluster groups and SSE provided in an embodiment of the present application.

[0022] Figure 4 This is a diagram of the internal structure of the LSTM provided in one embodiment of the present application.

[0023] Figure 5 This is an overall structural diagram of the LSTM-GA combination method provided in one embodiment of the present application.

[0024] Figure 6 A flowchart of a scheduling optimization method for an electric vehicle virtual power plant provided in one embodiment of the present application.

[0025] Figure 7 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0027] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0028] In an exemplary embodiment, Figure 1 As shown, a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant is provided, including the following steps 101 to 105.

[0029] Step 101, based on the historical operation data of each EV unit in the electric vehicle virtual power plant, adopt the mean calculation method to obtain the predicted value of the fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day; the fixed characteristic parameters include: the allowable upper limit of the battery state of charge, the allowable lower limit of the battery state of charge and the battery capacity.

[0030] Step 102, based on the historical operation data of each EV unit in the electric vehicle virtual power plant, adopt the LSTM-GA combination method to obtain the predicted value of the non-fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day; the non-fixed characteristic parameters include: grid-connected charge state, minimum off-grid power, grid-connected time and off-grid time.

[0031] Step 103, clustering each EV unit in the electric vehicle virtual power plant according to the predicted value of each characteristic parameter of each EV unit in the electric vehicle virtual power plant on the predicted day to obtain multiple clusters; each characteristic parameter includes each fixed characteristic parameter and each non-fixed characteristic parameter.

[0032] Step 104, respectively calculating the aggregate expression of each characteristic parameter in each cluster.

[0033] Step 105 , calculating the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant according to the aggregated expression of each characteristic parameter in each cluster.

[0034] By implementing the above steps 101 to 105 , the day-ahead response capability of the electric vehicle virtual power plant can be evaluated.

[0035] In another exemplary embodiment of the present application, step 105 of the present application is implemented by the day-ahead response capability evaluation model constructed by the present application. The present application first models the grid-connected response characteristics of EV monomers, obtains the day-ahead response capabilities of EV monomers under different response conditions, determines the characteristic parameters that affect the grid-connected response capabilities of EV monomers, and then uses the characteristic parameters as clustering indicators to group EVs, and expresses the characteristic parameters in an aggregated manner, and finally constructs a day-ahead response capability evaluation model for EVPP with EV groups as evaluation objects, specifically including the following steps 201-203.

[0036] Step 201, calculating the day-ahead response capability of an EV unit.

[0037] The response characteristics of an independent EV unit when connected to the grid are as follows: Figure 2 As shown in the figure, the shaded area ABCFED is used to illustrate the maximum controllable charging and discharging range of the EV monomer when it is connected to the grid: the EV monomer is at t a The state of charge when connected to the grid is SOC a (Point A), and immediately start charging at the rated charging power. When the maximum allowable power SOC is reached, it switches to the idle state, corresponding to Figure 2 The middle boundary ABC; the boundary ADE means that the EV cell is discharged at the rated discharge power immediately after being connected to the grid. SOC Stop discharging and switch to idle state at t d The state of charge when leaving the grid can reach the minimum power SOC that meets the user's travel needs for the day. d (This value is within the upper and lower limits of the allowable power), and the EF segment means that the EV needs to be charged compulsorily to meet the minimum power requirement for travel.

[0038] The whole process of EV single-cell grid connection can be expressed as the operation point (t, SOC) moving continuously over time in the maximum controllable charge and discharge area, and its actual movement trajectory depends on the single-cell [t a ,t] The change of grid connection status during the period, such as Figure 1 As shown by the middle dotted line, after the EV is connected to the grid, the grid-connected state is continuously changed, and the moving trajectory of its operating point changes accordingly, but it is always within the maximum controllable area.

[0039] Taking electric vehicle j as an example, the change of battery charge state during the charging and discharging process is shown in formula (1).

[0040]

[0041] Where: is the battery charge state of EV cell j at time t; and are the charging efficiency and discharging efficiency of EV cell j, respectively; is the charge and discharge power of EV cell j at time t-1 (with the power absorbed from the grid as the positive direction), which is equal to the rated charge power in the charging state and equal to the negative value of the rated discharge power in the discharging state; is the battery capacity of EV cell j; δ is the length of a single evaluation period.

[0042] Without participating in any response (user daily charging), EV does not have the ability to discharge back to the grid, that is, there are only two grid-connected states: charging and idle. In order to better meet travel needs, users generally choose to charge EV immediately after connecting to the grid, and stop charging when the minimum off-grid power is reached to save charging costs, that is, Figure 2 Therefore, when not participating in the response, the grid-connected state of the jth EV is and the state of charge at time t The relationship between is shown in formula (2).

[0043]

[0044] Where: The value 0 and 1 indicate that the EV is in idle and charging states respectively.

[0045] The day-ahead response capability assessment of EV monomers refers to the estimation of the maximum response capability that EV monomers can provide at each evaluation time on the evaluation day when the specific control strategy is unknown. When responding, EV monomers can change the size of the grid-connected load by changing their own grid-connected state when they are not responding. Define the transition from idle state to discharge state as response situation I, the transition from charging state to idle state as response situation II, and the transition from idle state to charging state as situation III. The first two response situations can achieve load reduction, and the third situation can achieve load increase. It should be noted that when the actual response scheduling of EV monomers is carried out, considering the loss of EV batteries, the charging and discharging power should not change too much, that is, there is no response situation where the charging state changes directly to the discharging state.

[0046] Therefore, the day-ahead response capability of EV monomers can also be divided into the above three response conditions, as shown in equations (3) to (5).

[0047]

[0048]

[0049]

[0050] Where: are the day-ahead response capabilities of EV unit j at time t under three response conditions; are the rated charging power and rated discharging power of the j-th EV respectively.

[0051] Step 202, EV group division.

[0052] In order to reduce the complexity of the calculation of the final EVPP's day-ahead response capability, the embodiment of the present application first divides the EV monomers into groups and aggregates and expresses EVs with similar response characteristics.

[0053] From step 201 , it can be known that the day-ahead response capability of an EV unit is limited by the boundary of its maximum controllable grid-connected area, so the maximum controllable grid-connected range can be used to distinguish EV units with different response characteristics.

[0054] From equations (1) to (5), it is not difficult to see that the parameters that affect the maximum controllable range of EV single-unit grid connection include: grid connection time Off-grid moments Battery state of charge upper limit Battery state of charge lower limit Grid charge status Minimum off-grid power and battery rated capacity

[0055] In the present application, the above parameters are referred to as characteristic parameters of EV monomers, and the characteristic parameters are used as clustering indicators. The K-means method is used to achieve EV clustering, and the elbow method is used to determine the optimal number of clusters. The specific steps of K-means clustering are as follows:

[0056] (1) Collect the characteristic parameter data of all EV monomers in the EVPP on the day of evaluation and normalize them as the basis for clustering.

[0057] (2) Each EV monomer is taken as a sample, the size of the characteristic parameter is used as the sample coordinate, and K initial center points are randomly selected.

[0058] (3) Use Euclidean distance to calculate the distance between K center points and each sample.

[0059] (4) Assign each sample to the cluster with the closest center point.

[0060] (5) Calculate the coordinate mean of all samples in each cluster in turn as the new center point of the cluster, and repeat steps (3)-(5) until the number of samples contained in each cluster no longer changes.

[0061] The number K of initial center points is the final number of clusters, and the elbow method can be used to determine the optimal number of clusters.

[0062] The sum of squared errors (SSE) is the core indicator of the elbow method. Its calculation method and theoretical basis are as follows:

[0063]

[0064] Where: k i is the number of sample points in the i-th cluster; pij represents the jth sample point in the i-th cluster; m i represents the center point of the i-th cluster; |p ij -m i | represents the sample point p ij With the center point m i The Euclidean distance between them; SSE is the clustering error of all samples, which represents the quality of clustering effect.

[0065] As the number of initial center points K increases, the SSE of the corresponding clustering results is calculated respectively, and we can get Figure 3 The relationship curve is shown in .

[0066] When K is less than the optimal number of clusters, the increase in K will greatly increase the degree of aggregation of each cluster, so the SSE decreases significantly. When K reaches the optimal number of clusters, the return of the degree of aggregation obtained by increasing K will rapidly decrease, and the SSE decrease will decrease sharply, and tend to be flat as the K value continues to increase. That is, the relationship between SSE and K is in the shape of an elbow, and the elbow (such as Figure 3 The K value corresponding to the solid point in the middle is the optimal number of clusters.

[0067] For the clustered EV group, its internal cells have similar battery characteristics and charging requirements, so the group's grid-in and grid-out time and characteristic parameters related to the state of charge can be expressed by the geometric center (i.e., mean) of the corresponding parameters of the internal cells, as shown in formula (7). At the same time, the battery capacity and rated charge and discharge power of the EV group can also be expressed by the corresponding parameters of the internal cells, as shown in formula (8).

[0068]

[0069]

[0070] Where: Represents the characteristic parameters of the i-th EV population (i.e., the i-th cluster) after aggregate expression.

[0071] Step 203: construct a day-ahead response capability assessment model for the EVPP.

[0072] After clustering, the grid-connected operation point movement trajectory of each EV group without participating in the response can be determined according to formula (9), and its grid-connected status at each evaluation time can be obtained. State of Charge

[0073]

[0074] in, is the grid connection status of the ith cluster at the evaluation time t on the day to be predicted, is the charge state of the ith cluster at the evaluation time t on the day to be predicted, is the aggregate expression of the network access time of the ith cluster, is the aggregated expression of the off-grid time of the ith cluster, is the aggregate expression of the minimum off-grid power of the ith cluster, is the aggregate expression of the upper limit of the battery state of charge of the ith cluster.

[0075] Then, according to equations (10), (11), and (12), the response capacity of the EV population under three response conditions is obtained:

[0076]

[0077]

[0078]

[0079] in, and They represent the response capabilities of response cases I, II and III of the i-th cluster at the evaluation time t of the predicted day, respectively. is the aggregate expression of the rated charging power of the ith cluster, is the aggregate expression of the rated discharge power of the ith cluster.

[0080] Finally, considering the response willingness of EV users, the upper and lower boundaries of the day-ahead response capability of EVPP can be obtained as shown in formula (13). The actual response power of EVPP will be constrained by this boundary during demand response scheduling.

[0081]

[0082] Where: To evaluate the lower limit of the day-ahead response capability of the EV VPP at time t, To evaluate the upper limit of the day-ahead response capability of the EV VPP at time t, To evaluate the total charging power of EV units in the electric vehicle virtual power plant when no response is participated in at time t, γ is the collective response willingness of EV users, and N is the optimal number of clusters.

[0083] Based on EV grouping, the upper and lower boundaries of the day-ahead response capability of EVPP are finally obtained, and the power response range for EVPP to participate in demand response scheduling is determined. In order to ensure the rationality of EV group division, that is, to always classify EV monomers with similar response characteristics into one category, it is necessary to establish an identification mechanism for EV characteristic parameters, so as to make the day-ahead response capability assessment of EVPP more accurate.

[0084] In another exemplary embodiment of the present application, in the above steps 101 and 102, the characteristic parameters are divided into fixed characteristic parameters and non-fixed characteristic parameters, and the historical mean is taken as the predicted value of the fixed characteristic parameters based on the historical operation data of the virtual power plant. For non-fixed characteristic parameters, the LSTM-GA combination method is used to optimize the hyperparameters of the memory network, and then the final memory network is trained on the entire sample set, and the parameters are predicted, specifically including the following steps 301-304.

[0085] Step 301: The characteristic parameters of EV cells are mainly affected by battery characteristics and user behavior, among which the upper and lower limits of the battery state of charge are and battery capacity Depending on the capacity characteristics of the EV battery itself, its value will basically not change and can be considered as a fixed characteristic parameter (fixed parameter prediction). Therefore, the historical daily data one year before the date to be evaluated is averaged to obtain the predicted value of the parameter on the day to be evaluated.

[0086] The prediction of the fixed characteristic parameters of the j-th EV unit is shown in formula (14).

[0087]

[0088] Where: Represents the fixed characteristic parameter of the Dth day in the past year.

[0089] Step 302: Non-fixed characteristic parameters Affected by the user's own travel and charging habits, it has a certain volatility and is a non-fixed characteristic parameter: the characteristic parameters of adjacent days will fluctuate, but the amplitude is not large. If the historical data series of these characteristic parameters are input into the prediction model and repeatedly trained, the long-term and short-term time correlation of the time series can be effectively captured, and a more accurate prediction effect can be obtained.

[0090] Long Short-Term Memory (LSTM) is a variant of a recurrent neural network (RNN). Its hidden layer consists of multiple state memory units that are adjacent to each other in time sequence. The internal structure of a single state memory unit is as follows: Figure 4 As shown, where x t is the input at time step t, C t is the state information variable at time step t, H tis the output variable at time step t. LSTM introduces a special storage unit and gating mechanism that can capture and process long-term dependencies in sequence data, effectively solving the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequences, thereby achieving accurate prediction of long time series data. Therefore, this application uses the LSTM method to achieve parameter identification.

[0091] In LSTM, the hidden layer consists of multiple memory cells connected in series. The forget gate determines the forgotten part of the memory cell. t and the output H of the previous time step t-1 Weighted and biased, information is passed through the sigmoid activation function σ(x), which outputs a 0-1 F t , F t The closer it is to 1, the higher the information retention of the previous memory unit is, and the closer it is to 0, the higher the degree of forgetting is. The functional relationship is expressed as:

[0092] F t =σ(x t W xf +H t-1 W hf +b f ) (15)

[0093] Where: W xf and W hf is the weight value of the forget gate; b f is the bias value of the forget gate.

[0094] The input gate determines the remaining part of the memory cell and updates the cell information. t and H t-1 Passed to the sigmoid activation function and the tanh activation function to generate an I of [0,1] t and [-1,1] in is the initial state variable at time step t, I t For control The degree of superposition into the final state variable, the larger the number, the higher the degree of superposition retention. t and Multiply and according to F t The previous memory unit information is selectively recorded into the current state information variable, expressed as:

[0095] I t =σ(x t W xi +H t-1 W hi +b i ) (16)

[0096]

[0097]

[0098] Where: W xi and W hi is the weight value of the input gate; b i is the bias value of the input gate.

[0099] The output gate determines the output of the current state memory unit, which is expressed as:

[0100] O t =σ(x t W xo +H t-1 W ho +b o ) (19)

[0101] H t =O t ×tanh(C t ) (20)

[0102] Where: W xo and W ho is the weight value of the output gate; b o is the bias value of the output gate.

[0103] Prediction of non-fixed characteristic parameters, specific scheme: Take the example of day-ahead parameter identification. First, the historical daily data of the state of charge of the jth EV in the past year are collected as known data. The abnormal data are removed and the time series is completed by interpolation. The processed data is normalized and then a certain time window size t is used. w Rolling generation training samples: w The data of the day is used as the input sequence of a training sample. w +1 day The data can be predicted as:

[0104]

[0105]

[0106] Where: X (n) , Y (n) are the input sequence and the quantity to be predicted in the nth training sample respectively; is the state of charge of the jth EV entering the grid on the nth day.

[0107] Assuming a year of 365 days, the total is 365-t wThe total samples are divided into training set and test set in a ratio of 7:3. The network is trained with the training set, and the mean square error (MSE) is used on the test set to evaluate the performance of the prediction model. The calculation method of MSE is shown in formula (24):

[0108]

[0109] Where: N t is the number of test set samples; are the days to be predicted in the kth test set sample. The actual value and predicted value of ; the smaller the MSE value, the better the prediction ability of the LSTM model.

[0110] The GA method is used to optimize the hyperparameters of the LSTM network. Hyperparameters such as the size of the time window and the number of state memory units contained in the hidden layer will affect the prediction accuracy of the LSTM network. Manual parameter adjustment will waste a lot of time and the optimization effect is poor. Therefore, this application uses a genetic algorithm (GA) to optimize the hyperparameters of the LSTM network to obtain the hyperparameter combination with the highest prediction accuracy. The overall structure of the LSTM-GA combination method is as follows: Figure 5 As shown, the optimization process is as follows:

[0111] (1) Initializing the population: The time window size and the number of state memory units of the LSTM network are used as optimization variables (genetic information), and their upper and lower thresholds are set respectively. M groups of hyperparameter combinations are randomly generated as the initial population of the genetic algorithm.

[0112] (2) Genetic information encoding: Encoding the genetic information, this application uses binary encoding;

[0113] (3) Calculate fitness: Decode the binary code of each individual in the population into the hyperparameter combination represented, substitute them into the LSTM network and train them on the training set, and calculate 1 / MSE as the individual fitness on the test set;

[0114] (4) Roulette selection: Assign a selection probability to each individual according to its fitness, randomly select an individual as the mother according to the selection probability, then remove the individual and continue to select the mother using the roulette method until m / 2 mothers are selected;

[0115] (5) Crossover and mutation: Pair the mothers in pairs, cross-exchange the genetic information (binary code) of the two according to certain rules, generate m / 2 new individuals, and perform random mutation operations (0-1 substitution) on the genetic information (binary code) of the new individuals to increase the diversity of genetic information;

[0116] (6) Combining the mutated new individuals with the parent population to obtain a new generation of population;

[0117] (7) Repeat steps (2)-(6) until the fitness of the optimal individual no longer changes or the maximum number of iterations is reached, then stop the iteration, input the hyperparameters represented by the current optimal individual into the LSTM network, and train the optimal parameter model on the entire sample set.

[0118] After the training is completed, the predicted date will be w The daily related data sequence is input into the LSTM network with optimized parameters, and the predicted day can be output. The predicted value of .

[0119] This combination method can also be used to achieve other non-fixed characteristic parameters Identification.

[0120] Combined with the above-mentioned day-ahead response capability assessment model, according to the identification results of the characteristic parameters in steps 102 and 103, EVs are grouped using step 103, so that the upper and lower limits of the day-ahead response capability of the EVPP are obtained by formula (13).

[0121] In an exemplary embodiment, Figure 6 As shown, a scheduling optimization method for an electric vehicle virtual power plant is provided, including the following steps 601-603.

[0122] Step 601, constructing an optimization model with the goal of optimizing the tracking effect of the actual response power on the dispatching demand signal and minimizing the amount of wind and solar power abandonment.

[0123] Step 602, based on the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant, determine the constraints of the optimization model; the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant are obtained using the day-ahead response capability evaluation method of the electric vehicle virtual power plant in the above-mentioned embodiment.

[0124] Step 603: Solve the optimization model based on the constraint conditions to obtain the optimal scheduling solution.

[0125] In another exemplary embodiment, EVPP participates in demand response by responding to the dispatch demand signal issued by the superior dispatch center (including the time period to be responded and the power size to be responded). The embodiment of the present application aims to optimize the improvement effect of EVPP on the system load curve (the tracking effect of the actual response power on the dispatch demand signal) and minimize the amount of wind and solar power abandoned by the system, and establish an optimization model:

[0126] minF=min(ω1f1+ω2f2) (24)

[0127]

[0128]

[0129]

[0130]

[0131] Where F is the extremely small objective function of the optimization model; f1 and f2 are the two sub-objective functions of the optimization model, and both f1 and f2 are extremely small, so formula (24) is the sum of the two; ω1 and ω2 are weight coefficients for dedimensionalization; T DR This is the time period when the superior dispatch center needs the electric vehicle virtual power plant to respond; is the actual operating power of the electric vehicle virtual power plant at the dispatching time t'; is the power response of the superior dispatch center to the demand of the electric vehicle virtual power plant at the dispatch time t'; They are the abandoned photovoltaic power generation and abandoned wind power generation; are the predicted output of photovoltaic power generation at the dispatching time t' and the actual output used to balance the load, and the difference is the abandoned photovoltaic power at that moment; They are respectively the predicted output of wind power at the dispatching time t' and the actual output used to balance the load, and the difference is the abandoned wind power at that moment.

[0132] The optimization variables of this optimization model are the operating power of EVPP at each scheduling time and the actual output of distributed renewable energy at each scheduling time. The remaining variables have been optimized and determined by the superior dispatching center.

[0133] The constraints of the model can be expressed as follows:

[0134]

[0135]

[0136]

[0137]

[0138] in, is the channel transmission power of the upper power grid at the scheduling time t', is the base load of the area where the EV virtual power plant is located at the dispatch time t', is the actual operating power of the electric vehicle virtual power plant at the dispatching time t', is the lower limit of the day-ahead response capability of the EV VPP at dispatch time t', is the upper limit of the day-ahead response capability of the EV VPP at dispatch time t', is the actual operating power of the electric vehicle virtual power plant at the dispatching time t', are the predicted output of photovoltaic power generation at the dispatching time t' and the actual output used to balance the load, are the predicted output of wind power generation at the dispatching time t' and the actual output for balancing the load. Furthermore, the above formula (29) is the system power balance constraint, The above equation (30) is the EVPP output constraint, which means that the actual output of EVPP should always be within the upper and lower boundaries of the day-ahead response capability. The above equations (31) and (32) are the output constraints of distributed photovoltaic power generation and distributed wind power generation, respectively.

[0139] In actual scheduling optimization, the sizes of ω1 and ω2 can be flexibly set according to the priority of system scheduling needs: when more emphasis is placed on improving the system load curve, that is, when EVPP needs to better track the scheduling demand signal, the size of ω1 can be appropriately increased; when more emphasis is placed on improving the utilization rate of new energy, that is, reducing the amount of wind and solar power abandonment, the size of ω2 can be appropriately increased.

[0140] The upper dispatch center in this embodiment requires the electric vehicle virtual power plant to respond to the time period T DR , is a part of the time period within the day to be predicted in the embodiment of the method for evaluating the day-ahead response capability of the electric vehicle virtual power plant. In the specific implementation process, the upper limit and lower limit of the day-ahead response capability of the electric vehicle virtual power plant in the above step 602 can be obtained by the following steps:

[0141] First, determine the time period T during which the superior dispatch center needs the EV VPP to respond. DR Then, the method for evaluating the day-ahead response capability of the electric vehicle virtual power plant in the above embodiment is used to obtain the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant at each evaluation time within the day to be predicted, and then the time period T during which the superior dispatching center needs the electric vehicle virtual power plant to respond is selected from the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant at each evaluation time within the day to be predicted. DR The upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant at each dispatching time t' in the application embodiment. In the application embodiment, the dispatching time and the evaluation time in the same time period can correspond one to one, or the evaluation time of the corresponding part of the dispatching time can be used. For example, the time interval of the evaluation time is 1s, and the time interval of the dispatching time can be an integer multiple of 1s.

[0142] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for evaluating the day-ahead response capability of an electric vehicle virtual power plant or a method for optimizing the scheduling of an electric vehicle virtual power plant is implemented.

[0143] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0144] According to the specific embodiments provided in this application, this application has the following technical effects.

[0145] This application adopts the K-means clustering method, uses characteristic parameters as clustering indicators to group all EVs aggregated in the virtual power plant, and determines the optimal number of clusters through the elbow method, thereby realizing the aggregate expression of EVs with similar response characteristics. A day-ahead response capability evaluation model of EVPP is established with EV groups as the evaluation object, reducing the computational complexity of the evaluation model. Subsequently, EV groups can also be used as optimization objects to participate in the scheduling optimization of EVPP to improve the optimization speed.

[0146] 2. Based on the data-driven idea, this application adopts the LSTM-GA combination method to capture the time dependency between the historical operation data of the virtual power plant, realizes the accurate identification of the non-fixed characteristic parameters of EV monomers, and ensures the optimal hyperparameters of the prediction network, which more accurately reflects the differences between EV individuals in actual scenarios.

[0147] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0148] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0149] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0150] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0151] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for evaluating the day-ahead response capability of an electric vehicle virtual power plant, characterized in that: include: Based on the historical operation data of each EV unit in the electric vehicle virtual power plant, the predicted value of the fixed characteristic parameter of each EV unit in the electric vehicle virtual power plant on the predicted day is obtained by using the mean calculation method; the fixed characteristic parameters include: the upper limit of the battery state of charge, the lower limit of the battery state of charge and the battery capacity; Based on the historical operation data of each EV unit in the electric vehicle virtual power plant, the LSTM-GA combination method is used to obtain the predicted value of the non-fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day; the non-fixed characteristic parameters include: grid-connected charge state, minimum off-grid power, grid-connected time and off-grid time; According to the predicted values ​​of the characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day, each EV unit in the electric vehicle virtual power plant is clustered to obtain multiple clusters; each characteristic parameter includes each fixed characteristic parameter and each non-fixed characteristic parameter; Calculate the aggregate expression of each characteristic parameter in each cluster separately; According to the aggregated expression of each characteristic parameter in each cluster, the upper and lower limits of the day-ahead response capability of the EV virtual power plant are calculated.

2. The method for evaluating the day-ahead response capability of an electric vehicle virtual power plant according to claim 1, characterized in that: Based on the historical operation data of each EV unit in the electric vehicle virtual power plant, the average calculation method is used to obtain the predicted values ​​of the fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day, including: Based on the historical operation data of each EV unit in the electric vehicle virtual power plant, the following formula is used to obtain the predicted value of the fixed characteristic parameter of each EV unit in the electric vehicle virtual power plant on the day to be predicted; in, SOC j and are the upper limit of the battery state of charge of EV cell j on the day to be predicted, the lower limit of the battery state of charge and the predicted value of the battery capacity; and are the upper limit of the battery state of charge, the lower limit of the battery state of charge, and the actual value of the battery capacity of EV cell j on the Dth day in the past year.

3. The method for evaluating the day-ahead response capability of an electric vehicle virtual power plant according to claim 1, characterized in that: Based on the historical operation data of each EV unit in the electric vehicle virtual power plant, the LSTM-GA combination method is used to obtain the predicted values ​​of the non-fixed characteristic parameters of each EV unit in the electric vehicle virtual power plant on the predicted day, including: Based on the actual value of the mth non-fixed characteristic parameter of each EV unit in the electric vehicle virtual power plant in the past year, a time window rolling method is used to generate training samples; m = 1, 2, ..., M, where M is the number of non-fixed characteristic parameters; Divide each training sample into a training set and a test set; Based on the training set and the test set, taking the mean square error as the loss function and adopting the GA algorithm, the LSTM network model is trained to obtain a trained LSTM network model; The EV monomer j before the prediction date t w The actual value of the mth non-fixed characteristic parameter on the day is input into the trained LSTM network model to obtain the predicted value of the mth non-fixed characteristic parameter of EV unit j on the predicted day, j = 1, 2, ..., J, J is the number of EV units in the electric vehicle virtual power plant, t w is the size of the time window.

4. The method for evaluating the day-ahead response capability of an electric vehicle virtual power plant according to claim 1, characterized in that: According to the predicted values ​​of each characteristic parameter of each EV monomer in the electric vehicle virtual power plant on the predicted day, each EV monomer in the electric vehicle virtual power plant is clustered to obtain multiple clusters, including: Initialize the number of cluster centers K; According to the predicted values ​​of the characteristic parameters of each EV unit in the electric vehicle virtual power plant on the prediction day, the K-means clustering algorithm is used to cluster each EV unit in the electric vehicle virtual power plant to obtain K clusters; Calculate the sum of squared errors of K clusters; Based on the elbow method, determine whether the absolute value of the difference between the sum of squared errors of the K clusters obtained in the current iteration and the sum of squared errors of the K-1 clusters obtained in the previous iteration is greater than a preset threshold, and obtain a determination result; If the judgment result is yes, the value of K is increased by 1, and the step of "clustering each EV monomer in the electric vehicle virtual power plant using the K-means clustering algorithm according to the predicted values ​​of each characteristic parameter of each EV monomer in the electric vehicle virtual power plant on the predicted day to obtain K clusters" is returned; If the judgment result is no, the value of K of the current iteration is used as the optimal clustering number, and the K clusters obtained in the current iteration are output.

5. The method for evaluating the day-ahead response capability of an electric vehicle virtual power plant according to claim 4 is characterized in that: The formula for calculating the sum of squared errors is: Among them, SSE is the sum of squared errors, p ij is the parameter vector of EV monomer j in the i-th cluster, m i is the parameter vector of the cluster center of the i-th cluster, k i is the number of EV monomers in the i-th cluster; the parameter vector of EV monomer j is composed of the predicted values ​​of each characteristic parameter of EV monomer j on the day to be predicted, and the parameter vector of the cluster center is composed of the average values ​​of the predicted values ​​of each characteristic parameter of each EV monomer in the i-th cluster on the day to be predicted.

6. The method for evaluating the day-ahead response capability of an electric vehicle virtual power plant according to claim 4, characterized in that: According to the aggregate expression of each characteristic parameter in each cluster, the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant are calculated, including: According to the aggregate expression of each characteristic parameter in each cluster, the following formula is used to calculate the grid connection status of each cluster in the electric vehicle virtual power plant at each evaluation time on the forecast day; in, is the grid connection status of the ith cluster at the evaluation time t on the day to be predicted, is the charge state of the ith cluster at the evaluation time t on the day to be predicted, is the aggregate expression of the network access time of the ith cluster, is the aggregated expression of the off-grid time of the ith cluster, is the aggregate expression of the minimum off-grid power of the ith cluster, is the aggregate expression of the upper limit of the battery state of charge of the ith cluster; According to the aggregate expression of each characteristic parameter in each cluster and the grid connection status of each cluster at each evaluation time on the day to be predicted, the following formula is used to calculate the response capacity of each cluster under different response conditions at each evaluation time on the day to be predicted; Among them, P t agg,i,Ⅰ , P t agg,i,Ⅱ and P t agg,i,Ⅲ They represent the response capabilities of response cases I, II and III of the i-th cluster at the evaluation time t of the predicted day, respectively. is the aggregate expression of the rated charging power of the ith cluster, is the aggregate expression of the rated discharge power of the i-th cluster; response case I is the case of transitioning from the idle state to the discharge state, response case II is the case of transitioning from the charging state to the idle state, and response case III is the case of transitioning from the idle state to the charging state; According to the aggregate expression of each characteristic parameter in each cluster and the response capability of each cluster at each evaluation time of the forecast day, the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant are calculated using the following formula; in, To evaluate the lower limit of the day-ahead response capability of the EV VPP at time t, To evaluate the upper limit of the day-ahead response capability of the EV VPP at time t, P t EVPP0 To evaluate the total charging power of EV units in the electric vehicle virtual power plant when no response is participated in at time t, γ is the collective response willingness of EV users, and N is the optimal number of clusters.

7. A scheduling optimization method for an electric vehicle virtual power plant, characterized in that: include: An optimization model is constructed with the goal of achieving the best tracking effect of the actual response power on the dispatch demand signal and minimizing the amount of wind and solar power abandoned. Based on the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant, determining the constraint conditions of the optimization model; the upper and lower limits of the day-ahead response capability of the electric vehicle virtual power plant are obtained by using the day-ahead response capability evaluation method of the electric vehicle virtual power plant described in any one of claims 1 to 6; Based on the constraints, the optimization model is solved to obtain the optimal scheduling solution.

8. The scheduling optimization method of the electric vehicle virtual power plant according to claim 7 is characterized in that: The optimization model is: minF=min(ω1f1+ω2f2); Among them, F is the extremely small objective function of the optimization model; f1 and f2 are the two sub-objective functions of the optimization model; ω1 and ω2 are the weight coefficients used for dedimensionalization; T DR This is the time period when the superior dispatch center needs the electric vehicle virtual power plant to respond; is the actual operating power of the EV virtual power plant at the dispatching time t'; is the power response of the superior dispatch center to the demand of the electric vehicle virtual power plant at the dispatch time t'; They are the abandoned photovoltaic power generation and abandoned wind power generation; They are respectively the predicted output of photovoltaic power generation at the dispatching time t' and the actual output used to balance the load; They are respectively the predicted output of wind power generation at the dispatching time t' and the actual output used to balance the load.

9. The scheduling optimization method of the electric vehicle virtual power plant according to claim 7 is characterized in that: The constraints are: in, is the channel transmission power of the upper power grid at the scheduling time t', is the base load of the area where the EV virtual power plant is located at the dispatch time t', is the actual operating power of the electric vehicle virtual power plant at the dispatching time t', is the lower limit of the day-ahead response capability of the EV VPP at dispatch time t', is the upper limit of the day-ahead response capability of the EV VPP at dispatch time t', is the actual operating power of the electric vehicle virtual power plant at the dispatching time t', are the predicted output of photovoltaic power generation at the dispatching time t' and the actual output used to balance the load, They are respectively the predicted output of wind power generation at the dispatching time t' and the actual output used to balance the load.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the day-ahead response capability assessment method for an electric vehicle virtual power plant as described in any one of claims 1-6, or the scheduling optimization method for an electric vehicle virtual power plant as described in any one of claims 7-9.

Images