Multi-energy virtual power plant optimization scheduling method and system

Abstract

The invention discloses a multi-energy virtual power plant optimization scheduling method and system, belongs to the technical field of power optimization scheduling, and solves the problems that an existing single prediction model cannot meet the requirements of multi-energy virtual power plant optimization scheduling, a data set applied to prediction calculation is large, and the calculation cost is high. According to the method, the influence factor indexes of the output of the wind and photovoltaic units are screened by adopting the random forest algorithm, the output of the wind and photovoltaic units is predicted by using the artificial neural network algorithm, and the defects of a single algorithm are further made up through the hybrid prediction algorithm, so that the prediction accuracy is improved; a Monte Carlo method is adopted for simulation analysis for load prediction of the electric vehicle, a function with the minimum operation cost of the virtual power plant as an objective is established, an optimal scheduling model of the multi-energy virtual power plant is established, and the problem that prediction and optimization can only be performed for single distributed energy in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power optimization and dispatching, and relates to a multi-energy virtual power plant optimization and dispatching method and system. Background Art

[0002] The core idea of ​​a virtual power plant is to bring together various types of decentralized adjustable power sources and loads, and form a virtual "power plant" through digital means for unified management and dispatching, while participating in the electricity market as a main body. With the development of renewable energy, more and more wind farms and photovoltaic power stations are incorporated into virtual power plants, and mobile trams are also connected as a kind of energy storage carrier. In a multi-energy virtual power plant, in order to achieve efficient utilization and flexible configuration of energy, various distributed energy sources, such as distributed power generation equipment and energy storage equipment, need to be coordinated and optimally dispatched. However, due to the strong uncertainty of wind power generation, photovoltaic power generation and other methods, how to accurately predict the power output and load of a virtual power plant has become one of the current difficulties in optimal dispatching.

[0003] Due to the intermittent and volatile nature of multiple energy sources, the quality and economic efficiency of power generation have become difficult issues. Therefore, in the process of controlling various distributed energy generation equipment, energy storage systems, and controllable loads in virtual power plants, coordinated control of them is the most critical part. For example, the invention patent with application publication number CN117196234A discloses a two-stage scheduling method for virtual power plants that integrate renewable energy and electric vehicles, which achieves the best bidding for virtual power plants in the day-ahead market and the optimal real-time scheduling of cold and hot power loads, facilitating the reasonable allocation of cold and hot power in virtual power plants.

[0004] At present, many mature algorithms have been formed in the computer field for numerical prediction, and they have a high prediction accuracy. Based on the meteorological, regional and other related data collected by the virtual power plant, the numerical prediction algorithm can be used to predict the load and power generation in the system. The more commonly used prediction algorithms include the ARIMA model based on statistical models, the support vector regression (SVR) model based on machine learning, and the artificial neural network (ANN) model.

[0005] The existing technology for optimizing the scheduling of virtual power plants through machine learning is mainly aimed at predicting a certain type of distributed energy. As more and more types of distributed energy are connected to virtual power plants, a single prediction model cannot meet the needs of optimizing the scheduling of multi-energy virtual power plants, and a more comprehensive analysis system is needed to predict power generation and load. At the same time, a single prediction model lacks feature selection, which makes the data set used in the prediction calculation large and the computing cost high. Therefore, there is an urgent need for a method that can combine multiple algorithms to optimize the scheduling of multiple distributed energy sources in virtual power plants to improve the efficiency and reliability of the energy system. Summary of the invention

[0006] The technical solution of the present invention is used to solve the problem that the existing single prediction model cannot meet the needs of optimized scheduling of multi-energy virtual power plants, and the data set used in the prediction calculation is large and the calculation cost is high.

[0007] The present invention solves the above technical problems through the following technical solutions:

[0008] A multi-energy virtual power plant optimization scheduling method comprises the following steps:

[0009] S1. Collect data on factors affecting wind turbine output, photovoltaic power generation unit output, and electric vehicle related data, clean the collected data, process outliers and missing values, and normalize them;

[0010] S2. Use the random forest algorithm to screen the factors affecting the output of wind and photovoltaic generators, and obtain the main factor indicators required for prediction;

[0011] S3, according to the main factor indicators selected in step S2, respectively establish wind and photovoltaic generator output prediction models based on artificial neural networks;

[0012] S4. Monte Carlo method is used to simulate the disorderly charging load of electric vehicles;

[0013] S5. Taking the maximization of the economic benefits of the virtual power plant as the objective function, construct the constraint models of each module and establish the optimization scheduling model of the multi-energy virtual power plant;

[0014] S6. Formulate the optimal energy dispatch plan for the virtual power plant based on the data calculated by the model.

[0015] Furthermore, the data on factors affecting the output of wind turbines include: wind direction, wind speed, humidity, temperature, atmospheric pressure, and air density data; the data on factors affecting the output of photovoltaic generators include: wind speed, humidity, temperature, direct solar radiation intensity, diffuse radiation intensity, total radiation intensity, weather type, and temperature; the electric vehicle related data include: time of leaving home, time of returning home to charge online, daily mileage, charging and discharging power, and duration.

[0016] Furthermore, the random forest algorithm described in step S2 is used to screen the factors affecting the output of wind and photovoltaic generators, and the main factor indicators required for prediction are obtained as follows:

[0017] S21. The historical data sets of factors affecting the output of wind and photovoltaic generators are used as training sets respectively. The Bootstraping method is used for random sampling with replacement. m samples are selected, and a total of n samplings are performed to generate n training sets. N decision tree models are trained for each of the n training sets.

[0018] S22, calculate the out-of-bag data error errOOB1 for the out-of-bag data OOB corresponding to each decision tree; randomly add noise interference to the feature X of all samples of the out-of-bag data OOB, calculate the out-of-bag data error errOOB2 again, and calculate the feature importance of feature X;

[0019] S23, according to the importance of features, remove the corresponding proportion of features to obtain a new feature set, and repeat the above process until p features remain;

[0020] S24. Calculate the out-of-bag error rate corresponding to each feature set, and select the feature set with the lowest out-of-bag error rate as the main factor indicator; the out-of-bag error rate is the proportion of misclassified samples in the out-of-bag samples.

[0021] Furthermore, the importance expression of the feature X in step S22 is:

[0022] Feature importance = ∑(errOOB2-errOOB1) / N

[0023] Where N is the number of decision trees in the random forest.

[0024] Furthermore, in step S3, according to the main factor indicators selected in step S2, the wind and photovoltaic generator output prediction models based on artificial neural networks are respectively established as follows:

[0025] The feature set screened in step S2 is selected as the model input variable of the artificial neural network, and the output variable is the power generation power of the corresponding unit. 80% of the data set is selected as the training set to train the neural network, and the artificial neural network parameters are continuously adjusted by the back propagation algorithm. The weight and bias of the neural network are adjusted by calculating the gradient of the cost function to the neural network parameters; 20% of the data set is selected as the validation set, and the trained model is evaluated using the validation set or cross-validation method, and the model is tuned according to the evaluation results.

[0026] Furthermore, the Monte Carlo method is used in step S4 to simulate the disorderly charging load of electric vehicles as follows:

[0027] S41, fitting the travel characteristics of electric vehicles, wherein the travel characteristics of electric vehicles include the distribution of the time when the electric vehicle arrives at home and leaves home, and the charging and discharging power and duration of the electric vehicle;

[0028] S42. Based on the fitted electric vehicle travel characteristics and data, the Monte Carlo method is used to randomly extract the initial parameters of the electric vehicle and set the scale and number of electric vehicles;

[0029] S43, simulating the time of returning home and connecting to the grid and the time of going out and leaving the grid according to the density function of the time of returning and leaving home of the electric vehicle, using the daily mileage density function to simulate the mileage to calculate the power consumption of the electric vehicle, and then calculating the charging time of a single electric vehicle;

[0030] S44, looping steps S42 to S43 until the number of simulated electric vehicles reaches the set scale value, and then accumulating to obtain the final total charging load.

[0031] Furthermore, in step S5, the objective function is to maximize the economic benefits of the virtual power plant, construct constraint models for each module, and establish an optimal scheduling model for a multi-energy virtual power plant, specifically:

[0032] S51. Taking the lowest operating cost of the virtual power plant as the goal, construct an objective function;

[0033] S52. Establish a constraint condition model, wherein the constraint condition model includes wind and solar unit output constraints, gas unit constraints, energy storage constraints, electric vehicle constraints, grid interaction constraints, and balance constraints.

[0034] Further, in step S51, the objective function expression is:

[0035] min Cost = C abondon +C gas +C w +C pv +C storage +C ev +C g

[0036] In the formula, min Cost is the objective function, C abondon The penalty cost for abandoning solar and wind power is C gas is the power generation cost of the gas unit, C w is the power generation cost of wind turbines, C pv is the power generation cost of the photovoltaic unit, C ev is the dispatching cost of electric vehicles, C storage is the energy storage charging and discharging cost, C g is the grid interaction cost.

[0037] Further, in step S52, the balance constraint expression is:

[0038]

[0039] In the formula, is the planned output of the virtual power plant at time t, P w,t is the wind turbine power output at time t, P pv,t is the wind turbine power output at time t, P gas,tis the power output of the gas generator set at time t, P dis,t is the energy storage discharge power at time t, is the power that the virtual power plant buys and sells to the grid at time t, is the discharge power of the electric vehicle at time t.

[0040] The present invention also provides a multi-energy virtual power plant optimization scheduling system, including a data acquisition module, an influencing factor screening module, a prediction model construction module, a load simulation module, a scheduling model construction module and a scheduling plan formulation module;

[0041] The data acquisition module is used to collect data on factors affecting wind turbine output, photovoltaic power generation unit output, and electric vehicle related data, and to clean the collected data, process abnormal values ​​and missing values, and normalize them;

[0042] The influencing factor screening module is used to screen the influencing factors of wind and photovoltaic generator output respectively by using random forest algorithm, and obtain the main factor indicators required for prediction respectively;

[0043] The prediction model building module is used to establish wind and photovoltaic generator output prediction models based on artificial neural networks according to the selected main factor indicators;

[0044] The load simulation module is used to simulate the disorderly charging load of electric vehicles using the Monte Carlo method;

[0045] The scheduling model building module is used to build constraint models for each module and establish a multi-energy virtual power plant optimization scheduling model with maximizing the economic benefits of the virtual power plant as the objective function;

[0046] The scheduling plan formulation module is used to formulate the optimal plan for virtual power plant energy scheduling based on data calculated by the model.

[0047] The advantages of the present invention are:

[0048] (1) The present invention uses a random forest algorithm to screen the influencing factors of wind and photovoltaic power generation, removes the indicators with low classification effect, and uses an artificial neural network algorithm to predict the output of wind and photovoltaic power generation. According to the data of wind and photovoltaic power generation, the artificial neural network prediction model of wind and photovoltaic power generation output based on random forest feature selection is trained respectively. The hybrid prediction algorithm further makes up for the shortcomings of a single algorithm to improve the prediction accuracy; through feature selection, the data set used for calculation is reduced, the influencing factors of wind and photovoltaic power generation output are screened and simplified, the calculation cost is reduced, and the redundant analysis indicators of the prediction model in the existing technology are simplified, thereby improving the applicability of the model.

[0049] (2) The present invention adopts the Monte Carlo method to perform simulation analysis on the load forecast of electric vehicles, and establishes an optimization scheduling model for a multi-energy virtual power plant that takes the minimization of the operating cost of the virtual power plant as the objective function, including wind, photovoltaic units, electric vehicles, gas units, and energy storage equipment. This solves the problem that the existing technology can only predict and optimize a single distributed energy source, improves the applicability of the model, reduces the cost of data collection and processing of the virtual power plant, and ultimately realizes the optimization scheduling of the multi-energy virtual power plant based on the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a multi-energy virtual power plant optimization scheduling method according to an embodiment of the present invention;

[0051] Figure 2 is a schematic diagram of a process of using a random forest algorithm to screen a feature set in an embodiment of the present invention;

[0052] Figure 3 is a schematic diagram of a flow chart of simulating electric vehicle charging load using the Monte Carlo method according to an embodiment of the present invention;

[0053] Figure 4 It is a schematic diagram of a multi-energy virtual power plant optimization scheduling model constructed according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments:

[0056] Embodiment 1

[0057] like Figure 1 Specifically, a multi-energy virtual power plant optimization scheduling method is disclosed, comprising the following steps:

[0058] S1. Collect data on factors affecting wind turbine output, photovoltaic power generation unit output, and electric vehicle related data, clean the collected data, process outliers and missing values, and normalize them;

[0059] The data of factors affecting the output of wind turbine generator set include: wind direction, wind speed, humidity, temperature, atmospheric pressure, and air density data;

[0060] The data of factors affecting the output of the photovoltaic generator set include: wind speed, humidity, temperature, direct solar radiation intensity, diffuse radiation intensity, total radiation intensity, weather type, and temperature;

[0061] The electric vehicle related data include: time of leaving home, time of returning home to charge, daily mileage, charging and discharging power and duration.

[0062] S2. Use the random forest algorithm (RF) to screen the factors affecting the output of wind and photovoltaic generators, and obtain the main factor indicators required for prediction;

[0063] S21. The historical data sets of factors affecting the output of wind and photovoltaic generators are used as training sets respectively. The Bootstraping method is used for random sampling with replacement. m samples are selected, and a total of n samplings are performed to generate n training sets. N decision tree models are trained for each of the n training sets.

[0064] like Figure 2 As shown in the figure, in the above sampling method, one data point is selected from the training set each time and then put back into the original training set, so that after each sampling, the selection chance of all data points in the original training data set remains unchanged, which means that the same data point may be selected multiple times in a bootstrap sample, which can improve the stability of the model when the sample size is limited or the data fluctuates greatly.

[0065] S22, calculate the out-of-bag data error errOOB1 for the out-of-bag data OOB corresponding to each decision tree; randomly add noise interference to the feature X of all samples of the out-of-bag data OOB, calculate the out-of-bag data error errOOB2 again, and calculate the feature importance of feature X;

[0066] Out-of-bag data refers to the data obtained through repeated sampling each time a decision tree is established for training the decision tree. At this time, about 1 / 3 of the data is not used and does not participate in the establishment of the decision tree. This part of the data can be used to evaluate the performance of the decision tree and calculate the prediction error rate of the model, which is called the out-of-bag data error.

[0067] Specifically, in step S22, for each decision tree, select its corresponding out-of-bag data OOB to calculate the out-of-bag data error and record it as errOOB1. Randomly add noise interference to the feature X of all samples of the out-of-bag data OOB, that is, randomly change the value of all samples at feature X, and calculate the out-of-bag data error again and record it as errOOB2. Assuming that there are N decision trees in the random forest, the importance expression of feature X is:

[0068] Feature importance = ∑(errOOB2-errOOB1) / N

[0069] S23, according to the importance of features, remove the features of corresponding proportions to obtain a new feature set, and repeat the above process until p features are left, where p is a value set in advance;

[0070] Since the larger the value of feature importance is, the greater the contribution of the feature to the prediction ability of the model is, that is, the more important the feature is, so by setting a threshold for feature importance, features with a feature importance higher than this threshold will be retained, while features with a feature importance lower than this threshold will be removed.

[0071] S24, calculating the out-of-bag error rate corresponding to each feature set, and selecting the feature set with the lowest out-of-bag error rate as the main factor indicator; the out-of-bag error rate is the proportion of misclassified samples in the out-of-bag samples;

[0072] According to the feature sets of wind and photovoltaic power generation obtained in the above process, and the out-of-bag error rates corresponding to the feature sets, the feature set with the lowest out-of-bag error rate is selected;

[0073] For each sample, the classification probability of the sample can be estimated by the proportion of samples correctly classified by multiple decision trees. For classification problems, the out-of-bag error rate is the proportion of samples out-of-bag that are misclassified.

[0074] S3, according to the main factor indicators selected in step S2, respectively establish wind and photovoltaic generator output prediction models based on artificial neural network (ANN);

[0075] Specifically, the feature set screened in step S2 is selected as the model input variable of the artificial neural network (ANN), the output variable is the power generation power of the corresponding unit, 80% of the data set is selected as the training set to train the neural network, and the artificial neural network parameters are continuously adjusted through the backpropagation algorithm. The weights and biases of the neural network are adjusted by calculating the gradient of the cost function to the neural network parameters, thereby optimizing the performance of the network so that the model can better fit the data.

[0076] Select 20% of the data set as the validation set, use the validation set or cross-validation method to evaluate the trained model, and tune the model according to the evaluation results to improve the generalization ability of the model.

[0077] Through the above steps, artificial neural network prediction models based on random forest feature selection for wind and photovoltaic power generation are trained according to the data of wind and photovoltaic power generation groups. Feature selection reduces the data set used for calculation, screens and simplifies the influencing factors of wind and photovoltaic power generation output, reduces the computing cost, simplifies the redundant analysis indicators of the prediction model in the existing technology, and improves the applicability of the model.

[0078] S4. Monte Carlo method is used to simulate the disorderly charging load of electric vehicles to reflect the charging load of electric vehicles in the region, so as to optimize the scheduling; Figure 3 As shown, the following steps are included:

[0079] S41, fitting the travel characteristics of electric vehicles, wherein the travel characteristics of electric vehicles include the distribution of the time when the electric vehicle arrives at home and leaves home, and the charging and discharging power and duration of the electric vehicle;

[0080] Based on a private car travel behavior report, the electric vehicle (EV) travel characteristics are determined and optimization instructions are proposed based on the above.

[0081] The probability density function f of the tram user's last homecoming time, i.e. the tram's arrival time b t for:

[0082]

[0083] Where t is the time constant, σ b is the sample standard deviation of the time it takes for the tram to arrive home, μ b is the sample mean of the time it takes for the tram to arrive home.

[0084] Probability density function of tram user travel time and tram departure time for:

[0085]

[0086] In the formula, σ l is the sample standard deviation of the time when the tram leaves home, μ l is the sample mean of the time when the tram leaves home.

[0087] Probability density function of daily mileage of electric vehicles for:

[0088]

[0089] In the formula, S x is the daily mileage, σ d is the sample standard deviation of the daily mileage of electric vehicles, μ x is the sample mean of daily mileage of electric vehicles.

[0090] The charging and discharging power expression of electric vehicles is:

[0091]

[0092] In the formula, are the charging and discharging power of the i-th electric vehicle, respectively, and are the accumulated charging and discharging power of a single electric vehicle; are the charging and discharging status of the i-th electric vehicle, respectively. When 1 is taken, it means the i-th vehicle is charging or discharging, otherwise it is taken as 0; n is the number of electric vehicles participating in charging and discharging at this time.

[0093] Discharge duration of electric vehicles that can accept discharge scheduling The calculation formula is:

[0094]

[0095] In the formula, is the discharge time of the electric vehicle, It is the dispatchable discharge and charge state of electric vehicles. is the rated capacity of the electric vehicle battery, is the discharge efficiency of electric vehicles, Discharge power for electric vehicles.

[0096] Electric vehicles can be dispatched to discharge and charge states The calculation formula is:

[0097]

[0098]

[0099] In the formula, are the state of charge of the electric vehicle when it is connected to the grid and the state of charge at the initial moment, S is the daily mileage of the electric vehicle, α is the power consumption of the EV per 100 kilometers, The minimum state of charge for electric vehicles to receive discharge scheduling.

[0100] Furthermore, the charging time of electric vehicles is divided into two cases:

[0101] For electric vehicles that do not participate in discharge, the charging time The calculation is as follows:

[0102]

[0103] In the formula, Indicates the expected state of charge of the electric vehicle after charging is completed; Indicates the charging efficiency of electric vehicles.

[0104] For electric vehicles involved in discharge, the charging time The calculation is as follows:

[0105]

[0106] S42. Based on the fitted electric vehicle travel characteristics and data, the Monte Carlo method is used to randomly extract the initial parameters of the electric vehicle and set the scale and quantity of the electric vehicle.

[0107] S43, simulating the time of returning home and connecting to the grid and the time of going out and leaving the grid according to the density function of the time of returning and leaving home of the electric vehicle, using the daily mileage density function to simulate the mileage to calculate the power consumption of the electric vehicle, and then calculating the charging time of a single electric vehicle.

[0108] S44, looping steps S42 to S43 until the number of simulated electric vehicles reaches the set scale value, and then accumulating to obtain the final total charging load.

[0109] S5. Taking the maximization of the economic benefits of the virtual power plant as the objective function, construct the constraint models of each module and establish the optimization scheduling model of the multi-energy virtual power plant;

[0110] S51. Taking the lowest operating cost of the virtual power plant as the goal, construct an objective function; the objective function min Cost expression is as follows:

[0111] min Cost = C abondon +C gas +C w +C pv +C storage +C ev +C g

[0112] In the formula, C abondon The penalty cost for abandoning solar and wind power is C gas is the power generation cost of the gas unit, C w is the power generation cost of wind turbines, C pv is the power generation cost of the photovoltaic unit, C ev is the dispatching cost of electric vehicles, C storage is the energy storage charging and discharging cost, C g is the grid interaction cost;

[0113] Specifically, the penalty cost C for abandoning solar power and wind power is abondon The calculation formula is:

[0114]

[0115] Where P abw,t and P abpv,t are the wind power and solar power abandonment at time t, respectively, ab is the penalty unit price for curtailing wind and solar power, and Δt is the unit time.

[0116] The power generation cost of the gas unit is C gas The calculation formula is:

[0117]

[0118] Where P gas,t is the power output of the gas generator set at time t, pab is the marginal cost of the gas unit’s output at time t.

[0119] The calculation formula for the power generation cost of the wind and photovoltaic generator sets is:

[0120]

[0121] Where P w,t is the wind turbine power output at time t, P pv,t is the wind turbine power output at time t; p w,t 、p pv,t are the marginal costs of wind and photovoltaic generators at time t respectively.

[0122] The electric vehicle dispatching cost is composed of the charging income of electric vehicles participating in demand response and the discharge cost of electric vehicles. Therefore, the electric vehicle dispatching cost C ev The calculation formula is:

[0123]

[0124] In the formula, is the electric vehicle discharge compensation cost at time t, is the benefit of electric vehicles participating in demand response at time t, is the discharge power of the i-th electric vehicle at time t, is the compensation price for electric vehicle discharge by the virtual power plant at time t, N dis is the number of electric vehicles discharged at time t, are the charging loads before and after implementing demand response at time t, They are the charging prices before and after implementing demand response at time t respectively.

[0125] According to economic demand theory, when the price of a commodity rises at a certain moment, the demand for the commodity in the market will decrease; conversely, when the price decreases, the demand for the commodity will increase; the same is true for charging load. Since the charging demand of electric vehicles is related to the electricity price at that moment, the elasticity coefficient is introduced to express the relationship between the charging demand of electric vehicles and the electricity price, as well as the relationship between the actual charging load demand and the real-time charging price:

[0126]

[0127] In the formula, ε st is the price elasticity of electricity demand, s and t are two dispatching periods respectively. When s=t, it is the self-elasticity coefficient, otherwise it is the cross-elasticity coefficient.

[0128] The energy storage charging and discharging cost C storage The calculation formula is:

[0129]

[0130] Where P ch,t , P dis,t is the energy storage charging and discharging power at time t, p ch,t 、p dis,t is the cost of charging and discharging energy storage at time t.

[0131] The grid purchase interaction cost is the cost of purchasing or selling electricity from the grid when the actual output of the virtual power plant deviates from the planned output. g The calculation formula is:

[0132]

[0133] In the formula, is the power that the virtual power plant buys and sells to the grid at time t, is the price at which the virtual power plant buys and sells electricity to the power grid at time t.

[0134] S52, such as Figure 4 As shown, a constraint condition model is established according to the resources and configuration of each module, and the constraint condition model includes wind and solar unit output constraints, gas unit constraints, energy storage constraints, electric vehicle constraints, grid interaction constraints, balance constraints, etc.;

[0135] The output constraints of wind and photovoltaic generators are:

[0136]

[0137] In the formula, is the predicted output of the hybrid model of wind and solar units at time t, μ α is the wind and solar power abandonment coefficient.

[0138] The gas unit constraints are:

[0139]

[0140] In the formula, is the upper and lower limits of the gas unit output power at time t, θ h ,θ l It is the ramp rate of the gas unit per unit time.

[0141] The energy storage constraint is:

[0142]

[0143] The state of charge (SOC) is calculated based on the charge and discharge status of the energy storage device during a certain period of time. It is related to the energy storage capacity, the device charge and discharge power, and efficiency. The energy storage capacity calculation formula is:

[0144]

[0145] In the formula, E t is the energy storage capacity at time t, P ch,t , P dis,t is the charging and discharging power of the energy storage at time t, η ch , η dis are the energy storage charging and discharging power respectively.

[0146] When the energy storage device is charging and discharging, its state of charge at time t is:

[0147]

[0148] After charging, the state of charge of an electric vehicle cannot exceed the maximum state of charge, and after discharging, it cannot be less than the minimum state of charge. When charging and discharging, the total discharge amount of the electric vehicle should be less than the total dispatchable discharge amount, and the total charge amount cannot exceed the total available charge amount. Finally, at the end of charging and discharging, its state of charge should be between the minimum and maximum states of charge. The electric vehicle constraints are specifically:

[0149]

[0150] In the formula, are the charging and discharging states of the electric vehicle at time t, When it is 1, it means charging, otherwise it is 0. When it is 1, it indicates discharge, otherwise it is 0; is the minimum and maximum state of charge allowed for electric vehicles, η ch,ev , η dis,ev are the charging and discharging efficiency of electric vehicles, They are the state of charge of the electric vehicle when it is connected to the grid and the state of charge at the initial moment, S is the daily mileage of the electric vehicle, and α is the power consumption of the EV per 100 kilometers.

[0151] The grid interaction constraints are:

[0152]

[0153] Where P line,t is the capacity of the interconnection line between the virtual power plant and the large power grid at time t, α t It is a 0-1 constraint for grid interaction, which is used to limit the virtual power plant to only buy or sell electricity at the same time.

[0154] The virtual power plant balance constraint is:

[0155]

[0156] In the formula, is the planned output of the virtual power plant at time t, Pw,t is the wind turbine power output at time t, P pv,t is the wind turbine power output at time t, P gas,t is the power output of the gas generator set at time t, P dis,t is the energy storage discharge power at time t, is the power that the virtual power plant buys and sells to the grid at time t, is the discharge power of the electric vehicle at time t.

[0157] S6. Formulate the optimal energy dispatching plan for the virtual power plant based on the data calculated by the model;

[0158] Specifically, according to the constructed multi-energy optimization scheduling model data based on the hybrid prediction algorithm, the CPLEX solver in MATLAB is used to simulate and solve the multi-energy optimization scheduling model to formulate the optimal solution for the energy scheduling of the virtual power plant.

[0159] Furthermore, other optimization algorithms may be selected to simulate and solve the multi-energy optimization scheduling model. Since there are many types of other existing optimization algorithms in this field, the present invention will not describe them in detail.

[0160] The present invention uses a random forest algorithm to screen the influencing factors of wind and photovoltaic generator output, removes the indicators with low classification effect, and uses an artificial neural network algorithm to predict the output of wind and photovoltaic generators. According to the data of wind and photovoltaic generators, the artificial neural network prediction model of wind and photovoltaic generator output based on random forest feature selection is obtained. The hybrid prediction algorithm further makes up for the shortcomings of a single algorithm to improve the prediction accuracy; through feature selection, the data set used for calculation is reduced, the influencing factors of wind and photovoltaic generator output are screened and simplified, the calculation cost is reduced, and the redundant analysis indicators of the prediction model in the existing technology are simplified, thereby improving the applicability of the model.

[0161] The present invention adopts the Monte Carlo method to perform simulation analysis on the load forecast of electric vehicles, and establishes an optimization scheduling model for a multi-energy virtual power plant with the minimum operating cost of the virtual power plant as the objective function, which includes wind, photovoltaic units, electric vehicles, gas units, and energy storage equipment. It solves the problem that only a single distributed energy can be predicted and optimized in the prior art, improves the applicability of the model, reduces the cost of virtual power plant data collection and processing, and finally realizes the optimization scheduling of the multi-energy virtual power plant based on the model.

[0162] The present invention also provides a multi-energy virtual power plant optimization scheduling system, including a data acquisition module, an influencing factor screening module, a prediction model construction module, a load simulation module, a scheduling model construction module and a scheduling plan formulation module;

[0163] The data acquisition module is used to collect data on factors affecting wind turbine output, photovoltaic power generation unit output, and electric vehicle related data, and to clean the collected data, process abnormal values ​​and missing values, and normalize them;

[0164] The influencing factor screening module is used to screen the influencing factors of wind and photovoltaic generator output respectively by using random forest algorithm, and obtain the main factor indicators required for prediction respectively;

[0165] The prediction model building module is used to establish wind and photovoltaic generator output prediction models based on artificial neural networks according to the selected main factor indicators;

[0166] The load simulation module is used to simulate the disorderly charging load of electric vehicles using the Monte Carlo method;

[0167] The scheduling model building module is used to build constraint models for each module and establish a multi-energy virtual power plant optimization scheduling model with maximizing the economic benefits of the virtual power plant as the objective function;

[0168] The scheduling plan formulation module is used to formulate the optimal plan for virtual power plant energy scheduling based on data calculated by the model.

[0169] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-energy virtual power plant optimization scheduling method, characterized in that: The following steps are involved: S1. Collect data on factors affecting wind turbine output, photovoltaic power generation unit output, and electric vehicle related data, clean the collected data, process outliers and missing values, and normalize them; S2. Use the random forest algorithm to screen the factors affecting the output of wind and photovoltaic generators, and obtain the main factor indicators required for prediction; S3, according to the main factor indicators selected in step S2, respectively establish wind and photovoltaic generator output prediction models based on artificial neural networks; S4. Monte Carlo method is used to simulate the disorderly charging load of electric vehicles; S5. Taking the maximization of the economic benefits of the virtual power plant as the objective function, construct the constraint models of each module and establish the optimization scheduling model of the multi-energy virtual power plant; S6. Formulate the optimal energy dispatch plan for the virtual power plant based on the data calculated by the model.

2. The multi-energy virtual power plant optimization scheduling method according to claim 1 is characterized in that: The data of factors affecting the output of wind turbine generator set include: wind direction, wind speed, humidity, temperature, atmospheric pressure, and air density data; The data of factors affecting the output of the photovoltaic generator set include: wind speed, humidity, temperature, direct solar radiation intensity, diffuse radiation intensity, total radiation intensity, weather type, and temperature; The electric vehicle related data include: time of leaving home, time of returning home to charge, daily mileage, charging and discharging power and duration.

3. The multi-energy virtual power plant optimization scheduling method according to claim 1 is characterized in that: In step S2, the random forest algorithm is used to screen the factors affecting the output of wind and photovoltaic generators, and the main factor indicators required for prediction are obtained as follows: S21. The historical data sets of factors affecting the output of wind and photovoltaic generators are used as training sets respectively. The Bootstraping method is used for random sampling with replacement. m samples are selected, and a total of n samplings are performed to generate n training sets. N decision tree models are trained for each of the n training sets. S22, calculate the out-of-bag data error errOOB1 for the out-of-bag data OOB corresponding to each decision tree; randomly add noise interference to the feature X of all samples of the out-of-bag data OOB, calculate the out-of-bag data error errOOB2 again, and calculate the feature importance of feature X; S23, according to the feature importance, remove the corresponding proportion of features to obtain a new feature set, and repeat the above process until p features remain; S24. Calculate the out-of-bag error rate corresponding to each feature set, and select the feature set with the lowest out-of-bag error rate as the main factor indicator; the out-of-bag error rate is the proportion of misclassified samples in the out-of-bag samples.

4. The multi-energy virtual power plant optimization scheduling method according to claim 3 is characterized in that: The importance expression of the feature X in step S22 is: Feature importance = ∑(errOOB2-errOOB1) / N Where N is the number of decision trees in the random forest.

5. The multi-energy virtual power plant optimization scheduling method according to claim 4 is characterized in that: In step S3, according to the main factor indicators selected in step S2, the wind and photovoltaic generator output prediction models based on artificial neural networks are respectively established as follows: The feature set selected in step S2 is selected as the model input variable of the artificial neural network, the output variable is the power generation power of the corresponding unit, 80% of the data set is selected as the training set to train the neural network, the artificial neural network parameters are continuously adjusted through the back propagation algorithm, and the weight and bias of the neural network are adjusted by calculating the gradient of the cost function to the neural network parameters; Select 20% of the data set as the validation set, use the validation set or cross-validation method to evaluate the trained model, and tune the model according to the evaluation results.

6. The multi-energy virtual power plant optimization scheduling method according to claim 1 is characterized in that: The Monte Carlo method is used in step S4 to simulate the disorderly charging load of electric vehicles as follows: S41, fitting the travel characteristics of electric vehicles, wherein the travel characteristics of electric vehicles include the distribution of the time when the electric vehicle arrives at home and leaves home, and the charging and discharging power and duration of the electric vehicle; S42. Based on the fitted electric vehicle travel characteristics and data, the Monte Carlo method is used to randomly extract the initial parameters of the electric vehicle and set the scale and number of electric vehicles; S43, simulating the time of returning home and connecting to the grid and the time of going out and leaving the grid according to the density function of the time of returning and leaving home of the electric vehicle, using the daily mileage density function to simulate the mileage to calculate the power consumption of the electric vehicle, and then calculating the charging time of a single electric vehicle; S44, looping steps S42 to S43 until the number of simulated electric vehicles reaches the set scale value, and then accumulating to obtain the final total charging load.

7. The multi-energy virtual power plant optimization scheduling method according to claim 1 is characterized in that: In step S5, the objective function is to maximize the economic benefits of the virtual power plant, construct constraint models for each module, and establish an optimal scheduling model for a multi-energy virtual power plant. Specifically, S51. Taking the lowest operating cost of the virtual power plant as the goal, construct an objective function; S52. Establish a constraint condition model, wherein the constraint condition model includes wind and solar unit output constraints, gas unit constraints, energy storage constraints, electric vehicle constraints, grid interaction constraints, and balance constraints.

8. The multi-energy virtual power plant optimization scheduling method according to claim 7 is characterized in that: In step S51, the objective function expression is: min Cost=C abondon +C gas +C w +C pv +C storage +C ev +C g In the formula, min Cost is the objective function, C abondon The penalty cost for abandoning solar and wind power is C gas is the power generation cost of the gas unit, C w is the power generation cost of wind turbines, C pv is the power generation cost of the photovoltaic unit, C ev is the dispatching cost of electric vehicles, C storage is the energy storage charging and discharging cost, C g is the grid interaction cost.

9. The multi-energy virtual power plant optimization scheduling method according to claim 7 is characterized in that: In step S52, the balance constraint expression is: In the formula, is the planned output of the virtual power plant at time t, P w,t is the wind turbine power output at time t, P pv,t is the wind turbine power output at time t, P gas,t is the power output of the gas generator set at time t, P dis,t is the energy storage discharge power at time t, is the power that the virtual power plant buys and sells to the grid at time t, is the discharge power of the electric vehicle at time t.

10. A multi-energy virtual power plant optimization scheduling system, characterized in that: It includes data acquisition module, influencing factor screening module, prediction model building module, load simulation module, scheduling model building module and scheduling plan formulation module; The data acquisition module is used to collect data on factors affecting wind turbine output, photovoltaic power generation unit output, and electric vehicle related data, and to clean the collected data, process abnormal values ​​and missing values, and normalize them; The influencing factor screening module is used to screen the influencing factors of wind and photovoltaic generator output respectively by using random forest algorithm, and obtain the main factor indicators required for prediction respectively; The prediction model building module is used to establish wind and photovoltaic generator output prediction models based on artificial neural networks according to the selected main factor indicators; The load simulation module is used to simulate the disorderly charging load of electric vehicles using the Monte Carlo method; The scheduling model building module is used to build constraint models for each module and establish a multi-energy virtual power plant optimization scheduling model with maximizing the economic benefits of the virtual power plant as the objective function; The scheduling plan formulation module is used to formulate the optimal plan for virtual power plant energy scheduling based on data calculated by the model.

Images