Energy output optimization scheduling method and system considering random source load virtual power plant

Abstract

The invention relates to the field of short-term load prediction and power dispatching control of a power system, in particular to an energy output optimization dispatching method and system considering a random source load virtual power plant, and designs a virtual power plant double-layer optimization dispatching model. The lower-layer model constructs a user load prediction model of an improved Transform, and then a renewable energy output model is established by using a scene generation method; the upper layer model takes the power generation cost minimization of the wind-light storage and combustion virtual power plant as a target, considers constraint conditions such as gas turbine output power, ramp rate, energy storage charging and discharging power, electricity storage quantity and power balance, and proposes a single guide target function comparison algorithm considering random source load to carry out optimization solution on a virtual power plant optimization scheduling target function; and planning and deciding the output of each energy source in the next 24 hours. Experiments show that the new energy consumption level can be remarkably improved under the random source load condition, and better economic benefits are obtained while stable operation of a power distribution system is promoted.

Description

Technical Field

[0001] The present invention relates to the field of short-term load forecasting and power dispatching control of power systems, and in particular to an energy output optimization dispatching method and system considering random source and load virtual power plants. Background Art

[0002] With the proposal of the "dual carbon" goal and the changes in global energy forms, renewable energy such as photovoltaics and wind power has been vigorously developed, and the proportion of renewable energy in the distribution network has gradually increased. Its full absorption is of great significance to the efficient operation of the distribution system. However, the output of renewable energy and user load are highly random due to environmental and human factors, making it difficult to fully absorb renewable energy. In order to address the problem of insufficient absorption of renewable energy, virtual power plants can be used to aggregate energy storage, gas turbines, wind power, photovoltaics and other energy sources, and improve the absorption rate of renewable energy by dispatching various energy outputs.

[0003] Although some existing virtual power plant dispatching technologies have optimized the dispatching of virtual power plants, they have not fully considered the impact of source-load uncertainty, and it is difficult to ensure the stability and reliability of dispatching when facing actual working conditions of source-load changes. In addition, although some studies have considered the source-load uncertainty problem in virtual power plant dispatching planning, their new energy and user load data are derived from assumptions about the actual situation and related statistical results, which are difficult to reflect the actual situation of power grid operation and have certain limitations.

[0004] For example, the invention patent with publication number CN115375014A discloses a source-load combination probability prediction method, device and storage medium, which uses a CNN feature extraction module to replace the original multi-head attention module of the encoder in the Transformer model, which can extract data features and avoid the problem of being unable to manually select feature parameters due to the unclear internal coupling relationship between source load and meteorological data, thereby avoiding early complex data processing operations. However, it does not consider the impact of source-load uncertainty conditions on the scheduling optimization of virtual power plants.

[0005] Existing studies have focused on the impact of actual source-load uncertainty conditions on virtual power plant scheduling optimization. However, the optimization algorithms used have certain deficiencies in their optimization ability and convergence speed when dealing with high-complexity problems. They are prone to fall into local optimality, which affects the accuracy and efficiency of the optimization scheduling. In addition, existing methods usually have problems such as complex calculations, numerous parameters, and difficulty in adjusting parameters.

[0006] For example, the invention patent with publication number CN118971021A discloses an active distribution network energy storage control method containing high-penetration new energy. It solves the problem of accidental disconnection of new energy units from the existing new energy power grid after a fault. The main problem is insufficient frequency regulation capability of the system, but obviously, the calculation is very complicated. Summary of the invention

[0007] Purpose of the invention: In order to solve the problems arising from the above-mentioned prior art, the present invention provides a method for optimizing the energy output of a virtual power plant considering random source and load. The method solves the problems of poor stability of the distribution system and poor economic benefits caused by the low level of new energy consumption under random source and load conditions. The present invention also provides a system for optimizing the energy output of a virtual power plant considering random source and load.

[0008] Technical solution: According to a first aspect of the present invention, a method for optimizing energy output scheduling of a virtual power plant considering random source and load is provided, the method comprising:

[0009] S1 constructs a two-layer optimization scheduling model for a virtual power plant based on load and new energy output data prediction and minimization of power generation costs, which includes a lower model and an upper model; the lower model is a user load prediction model based on an improved Transformer, and the improved Transformer user load prediction model includes a Transformer model and an estimation model. The Transformer model can extract deep-level time series features, and then the user load is finally predicted through the estimation model to obtain user load prediction data;

[0010] S2 uses the scenario generation method to establish a renewable energy output model, thereby obtaining wind power and photovoltaic output forecast data for a certain period of time in the future;

[0011] S3 constructs an upper-level model based on user load forecast data and wind power and photovoltaic output forecast data. The upper-level model takes minimizing the power generation cost of the wind, solar, storage and fuel virtual power plant as the objective function, and considers the constraints of gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance;

[0012] S4 is based on the objective function of the upper model, and uses a monomer-guided objective function comparison algorithm to optimize and solve the objective function, thereby obtaining the optimal value of each energy output at a certain time in the future.

[0013] Further, including:

[0014] In step S1, the process of constructing the user load prediction model based on the improved Transformer includes:

[0015] S11 data preprocessing: collect historical user load data and divide the historical user load data into training set data and test set data according to a certain ratio;

[0016] S12 model construction: using the improved Transformer model to process the training set data;

[0017] S13 model training and load forecasting: Use the preprocessed training set to train the improved Transformer model, set the maximum number of iterations, and when the maximum number of iterations is reached, the training is completed, and the test set is used to verify the performance of the model; finally, the load data of a certain power area on the same day is input into the trained improved Transformer model to obtain the load forecast data of the power area for several future times.

[0018] Further, including:

[0019] The estimation model is expressed as:

[0020]

[0021] Where n is the number of boundary vectors, α i is the weight multiplier, It is a function that maps the Transformer model output data x to a higher dimensional space, x' is the boundary vector, and b is the bias term;

[0022] in, The function expression is:

[0023]

[0024] In the formula, c ij is the autocorrelation coefficient, i,j=1,2,...u, and u is the dimension of the Transformer model output data.

[0025] Further, including:

[0026] In the step S2, a renewable energy output model is established according to the lower model using a scenario generation method, specifically including:

[0027] S21 uses non-parametric kernel density estimation method to fit the historical series data of wind power and photovoltaic output to obtain the kernel density estimation formula of wind power and photovoltaic;

[0028] S22 establishes a joint distribution model of wind power and photovoltaic output based on Copula function;

[0029] S23 uses the kernel density estimation formula to calculate the error between the estimated value obtained by the joint distribution model and the actual value of wind power and photovoltaic output, selects the optimal Copula function corresponding to the minimum error as the optimal wind power and photovoltaic joint probability distribution, and generates preliminary wind power and photovoltaic output scenario data;

[0030] S24 reduces the preliminary wind power and photovoltaic output scenario data according to the backward reduction method, and calculates the probability of each scenario after the reduction. The wind and photovoltaic output scenario with the highest probability of occurrence is the wind power and photovoltaic output forecast data for a certain period of time in the future.

[0031] Further, including:

[0032] In step S3, the upper model takes minimizing the power generation cost of the wind, solar, storage and fuel-fired virtual power plant as the objective function, and the objective function is expressed as:

[0033]

[0034] In the formula, s represents the photovoltaic output scenario, w represents the wind power output scenario; t is the trading period; The cost of purchasing electricity from the grid for the VPP; For gas turbine operating costs; The operating cost of energy storage;

[0035] The specific calculation formulas for each cost are as follows:

[0036]

[0037] In the formula, is the VPP power purchase price, in yuan / MW·h, is the amount of electricity purchased by the VPP, in MW·h;

[0038]

[0039] In the formula, c MT is the unit power generation cost of the gas turbine, in yuan / MW·h, is the power generation of the gas turbine, in MW·h;

[0040]

[0041] In the formula, is the real-time electricity price in the electricity market, η c is the charging efficiency of energy storage, is the charge capacity of energy storage, in MW·h.

[0042] Further, including:

[0043] In step S3, the gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance constraints are considered, including:

[0044] Gas turbine constraints:

[0045]

[0046] Where P min and P max are the minimum and maximum output power of the gas turbine, is the current output power of the gas turbine; r u and r d are the upward and downward ramp rates of the gas turbine, respectively;

[0047] Energy storage constraints:

[0048]

[0049] S min ≤S s,w,t ≤S max

[0050] Where P c,max and P d,max are the maximum charging and discharging power of energy storage, respectively. and They are respectively the current charging and discharging power of the energy storage; S s,w,t is the state of charge of the energy storage, S min and S max are the minimum and maximum storage capacity of energy storage, η c and η d are the charging and discharging efficiencies of energy storage, respectively;

[0051] Power balance constraints:

[0052]

[0053] In the formula, is the user load benchmark value, that is, the load forecast data for a certain period of time in the future obtained by the lower model. The wind power and photovoltaic output forecast data for several future periods of time obtained by the lower model.

[0054] Further, including:

[0055] In the step S4, based on the objective function of the upper model, the objective function is optimized and solved by using a monomer-guided objective function comparison algorithm, including:

[0056] S41 defines the search space according to the three constraints of gas turbine constraint, energy storage constraint and power balance constraint, sets the expected value of the objective function, and selects the random direction of the position vector search. The calculation formula is as follows:

[0057]

[0058] X l =X β -d β h;

[0059] X r =X β +d β h;

[0060] In the formula, h is a normalized random vector, rand(·) is a random function, k is the dimension of the solution, i.e., the 24-hour output value of each energy source, X r is the position vector on the right side of the search area, X l is the position vector on the left side of the search area, d is the distance between the search center and the exploration point, β is the number of iterations, and X β is the position vector of the iterative β-time solution, which corresponds to the planned output value of each energy source;

[0061] S42 updates the position, and the calculation formula is as follows:

[0062] X β+1 =X β +δ β *h*sign[f(X r )-f(X l )];

[0063] Let M = [f(X r )-f(X l )], M is the intermediate variable, then:

[0064]

[0065] Where f(·) represents the cost calculation function corresponding to each energy output plan; δ β is the search step size at the βth iteration; sign(·) represents the sign function;

[0066] S43 brings the output result of step S42, that is, the planned output value of each energy source, into the objective function to determine whether it meets the expected value. If it does, the solution is output; if it does not, the mutation probability W is introduced. r And update the search step size and the distance d between the search center and the exploration point, and then iteratively update the position according to step S42. The output solution that meets the expected value is the optimal solution to the complex object problem faced in the virtual power plant scheduling planning under distributed energy aggregation, that is, the optimal value of each energy output in the future.

[0067] Further, including:

[0068] The calculation formula of the distance d between the search center and the exploration point is expressed as:

[0069]

[0070] On the other hand, the present invention also provides an energy output optimization scheduling system considering random source and load virtual power plants, the system comprising:

[0071] A two-layer optimization scheduling model construction module is used to construct a two-layer optimization scheduling model of a virtual power plant based on load and new energy output data prediction and minimization of power generation costs, which includes a lower model and an upper model; the lower model is a user load prediction model based on an improved Transformer, and the improved Transformer user load prediction model includes a Transformer model and an estimation model, and the Transformer model can extract deep-level time series features, and then the user load is finally predicted through the estimation model to obtain user load prediction data;

[0072] The wind and solar forecast module is used to establish a renewable energy output model using the scenario generation method, thereby obtaining wind power and photovoltaic output forecast data for a certain period of time in the future;

[0073] An upper model construction module is used to construct an upper model based on user load forecast data and wind power and photovoltaic output forecast data. The upper model takes minimization of the power generation cost of the wind, solar, storage and fuel virtual power plant as the objective function, and considers the constraints of gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance;

[0074] The optimal value calculation module is used to optimize and solve the objective function based on the objective function of the upper model by using a monomer-guided objective function comparison algorithm, so as to obtain the optimal value of each energy output at a certain time in the future.

[0075] Finally, the present invention also provides an energy output optimization scheduling device for virtual power plants considering random sources and loads. The scheduling device includes: a memory, a processor, and an energy output optimization scheduling program for virtual power plants considering random sources and loads stored in the memory and executable on the processor. When the energy output optimization scheduling program for virtual power plants considering random sources and loads is executed by the processor, the steps of the energy output optimization scheduling method for virtual power plants considering random sources and loads as described above are implemented.

[0076] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0077] This paper proposes a design method for a two-layer optimization scheduling model of a virtual power plant considering random source loads. Aiming at the influence of random fluctuations in user-side loads, an improved Transformer user load prediction model is proposed, and a scenario generation method is used to obtain typical renewable energy output scenarios. On this basis, a single-body guided objective function comparison algorithm is used to search for the optimal solution of the virtual power plant scheduling model objective function, and then plan and decide on the output of each energy source. Specifically:

[0078] The method of the present invention can effectively improve the generalization ability of the model and enhance the prediction accuracy of the model through the improved Transformer model, thereby obtaining accurate user-side load prediction information and providing a reliable data basis for further virtual power plant scheduling planning;

[0079] The single-unit guided objective function comparison algorithm proposed in the present invention can effectively handle the optimal value solution problem of complex objects faced in the scheduling and planning of virtual power plants under distributed energy aggregation. It has a fast convergence speed, is not easy to fall into the local optimum, and has simple calculation, few parameters, and is easy to adjust parameters.

[0080] Based on the prediction of user-side load and renewable energy output, the present invention coordinates and optimizes the output of various energy sources through the optimal scheduling of virtual power plants, thereby improving the renewable energy consumption rate and reducing the cost of power generation while promoting the stable operation of the distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is a framework diagram of the two-layer optimization scheduling model described in Example 1 of the present invention;

[0082] Figure 2 This is a flow chart of the improved Transformer algorithm described in Example 1 of the present invention;

[0083] Figure 3 This is a wind power output scenario diagram of the present invention;

[0084] Figure 4 This is a photovoltaic output scene diagram of the present invention;

[0085] Figure 5 This is a typical daily wind power output scene diagram of the present invention;

[0086] Figure 6 This is a typical daily photovoltaic output scene diagram of the present invention;

[0087] Figure 7 This is a schematic diagram of the virtual power plant scheduling result according to Embodiment 1 of the present invention;

[0088] Figure 8 A comparison chart of the new energy production capacity and dispatching output described in the present invention. DETAILED DESCRIPTION

[0089] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0090] Example 1

[0091] The present invention provides an energy output optimization scheduling method for a virtual power plant considering random source and load, the method comprising the following steps:

[0092] S1 constructs a two-layer optimization scheduling model for a virtual power plant based on load and new energy output data prediction and minimization of power generation costs, which includes a lower model and an upper model; the lower model is a user load prediction model based on an improved Transformer, and the load prediction model of the improved Transformer includes a Transformer model and an estimation model. The Transformer model can extract deep-level time series features, and then the user load is finally predicted through the estimation model to obtain user load prediction data;

[0093] S2 uses the scenario generation method to establish a renewable energy output model, thereby obtaining wind power and photovoltaic output forecast data for a certain period of time in the future;

[0094] S3 constructs an upper-level model based on user load forecast data and wind power and photovoltaic output forecast data. The upper-level model takes minimizing the power generation cost of the wind, solar, storage and fuel virtual power plant as the objective function, and considers the constraints of gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance;

[0095] S4 is based on the objective function of the upper model, and uses a monomer-guided objective function comparison algorithm to optimize and solve the objective function, thereby obtaining the optimal value of each energy output at a certain time in the future.

[0096] Specifically, in this embodiment, the present invention designs a virtual power plant two-layer optimization scheduling model based on the load and new energy output data prediction and the minimization of power generation cost. Figure 1 In the lower model, a user load forecasting model based on the improved Transformer is constructed, and then the renewable energy output model is established using the scenario generation method; in the upper model, the goal is to minimize the power generation cost of the wind, solar, storage and fuel-fired virtual power plant, and consider the constraints such as gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, power balance, etc., and propose a virtual power plant optimization scheduling method based on the single-unit guided objective function comparison algorithm taking into account random source and load.

[0097] The Transformer model described in this embodiment is a Transformer neural network, and its architecture mainly consists of four parts: an input part (input-output embedding and position encoding), a multi-layer encoder, a multi-layer decoder, and an output part (output linear layer and Softmax).

[0098] Input part: Source text embedding layer: Convert the numerical representation of words in the source text into vector representation to capture the relationship between words.

[0099] Position Encoder: Generates a position vector for each position of the input sequence so that the model can understand the position information in the sequence. Target Text Embedding Layer (used in Decoder): Converts the vocabulary numerical representation in the target text into a vector representation.

[0100] Encoder part: It is composed of N encoder layers stacked together. Each encoder layer consists of two sub-layer connection structures: the first sub-layer is a multi-head self-attention sub-layer, and the second sub-layer is a feed-forward fully connected sub-layer. Each sub-layer is followed by a normalization layer and a residual connection.

[0101] Decoder part: It is composed of N decoder layers stacked together. Each decoder layer consists of three sub-layer connection structures: the first sub-layer is a masked multi-head self-attention sub-layer, the second sub-layer is a multi-head attention sub-layer (encoder to decoder), and the third sub-layer is a feed-forward fully connected sub-layer. Each sub-layer is followed by a normalization layer and a residual connection.

[0102] Output part: Linear layer: converts the vector output by the decoder into the final output dimension. Softmax layer: converts the output of the linear layer into a probability distribution for the final prediction.

[0103] In this embodiment, the user load forecasting model can predict future load information based on the user load historical data. In the power system, the input of the load forecasting model is usually a long time series. In terms of processing long time series data, Transformer can extract important features of the time series because it contains an attention mechanism. Compared with algorithms such as convolutional neural networks and LSTM, it has outstanding advantages. However, due to the simple structure of Transformer, its generalization ability is insufficient and the model accuracy is limited. Based on the above reasons, this application introduces an estimation model into Transformer to solve the problem of constructing a load forecasting model in the lower-level model, and proposes an improved Transformer hybrid prediction model. The flowchart of the improved Transformer algorithm is shown in the figure. Figure 2 shown.

[0104] The construction of load forecasting model includes the following steps:

[0105] A1. Data preprocessing:

[0106] Collect historical user load data and divide it into training set data and test set data in a 4:1 ratio.

[0107] A2. Model construction:

[0108] The Transformer model is used to process the training set data, and its self-attention mechanism is used to capture the long-term dependencies and periodic changes in the data. The output of the Transformer model is used as the input of the next estimation model, that is, the Transformer model and the estimation model constitute the improved Transformer model. The Transformer model can extract deep time series features, and then the SVM estimation model is used to enhance the generalization ability, and the user load is finally predicted based on the time series features.

[0109] The function of the estimated model is:

[0110]

[0111] Where n is the number of boundary vectors, α i is the weight multiplier, It is a function that maps the Transformer output data x to a higher dimensional space, x' is the boundary vector, and b is the bias term.

[0112] The function expression is:

[0113]

[0114] In the formula, c ij is the autocorrelation coefficient, i,j=1,2,...u, u is the dimension of Transformer output data.

[0115] A3. Model training and load forecasting:

[0116] The preprocessed training set is used to train the improved Transformer model, and the maximum number of iterations is set to 1000. When the maximum number of iterations is reached, the training is completed, and the test set is used to verify the performance of the model.

[0117] By inputting the load data of a certain power area on the current day into the trained improved Transformer model, the load forecast data of the power area for a period of time in the future can be obtained. In this embodiment, the period of time is set to 24 hours.

[0118] In order to evaluate the performance of the proposed algorithm, mean square absolute error, mean square error, root mean square error and mean absolute percentage error are introduced as evaluation indicators. Among them, mean square absolute error represents the average value of the absolute error between the predicted value and the actual value; mean square error is the average value of the square of the error; root mean square error is the standard deviation of the error; and mean absolute percentage error is the average value of the relative percentage of the absolute error between the predicted value and the actual value. The following are the calculation formulas for these four indicators:

[0119]

[0120] Where Y a is the ath actual value, y i is the ath predicted value, and N is the total number of samples.

[0121] In order to evaluate the performance of the proposed load forecasting model, the test data sets are input into CNN, LSTM, bidirectional LSTM, Transformer and improved Transformer models for regression prediction and performance index calculation. The performance index results of the five models are shown in Table 1.

[0122] Table 1 Performance index results of five models

[0123]

[0124] As can be seen from Table 1, although Transformer can encode the input data and extract important features of the time series, its generalization ability is limited. Introducing the estimation model into Transformer can effectively improve the generalization ability of the model and thus improve the model accuracy. The mean square absolute error, mean square error, root mean square error and mean absolute percentage error of the improved Transformer model are all smaller than those of other algorithms. In general, the performance of the proposed model is better than that of the other four models.

[0125] On the basis of the construction of the user load forecasting model, considering the randomness and volatility of renewable energy output, the scenario generation method can accurately describe the changing rules of wind and light output and generate typical output scenarios. The process of using the scenario generation method to construct a renewable energy output model is as follows:

[0126] B1: Use non-parametric kernel density estimation method to fit the historical series data of wind power and photovoltaic output to obtain the kernel density estimation formula of wind power and photovoltaic;

[0127] B2: Based on the Copula function, a joint distribution model of wind power and photovoltaic output is established;

[0128] The expression of Copula is:

[0129] F(x1,x2,...,x m )=C(F X1 (x1),F X2 (x2),...F Xm (x m )) (3)

[0130] In the formula, m represents the number of variables; F Xg (x g) is the marginal distribution function of a single variable (g=1,2,3…,m); C(·) is the Copula connection function; F(x1,x2,x3,…x m ) is the joint distribution model of m variables.

[0131] B3: Use the kernel density estimation formula to calculate the error between the estimated value obtained by the joint distribution model and the actual value of wind power and photovoltaic output, select the optimal Copula function corresponding to the minimum error as the optimal wind power and photovoltaic joint probability distribution, and generate preliminary wind power and photovoltaic output scenario data; use the optimal wind power and photovoltaic joint probability distribution to obtain 1000 sets of wind power and photovoltaic output scenario data as follows Figure 3 and Figure 4 As shown. Figure 3 and Figure 4 It can be seen that wind power produces power throughout the 24 hours, while photovoltaic power only produces power during periods of high sunshine intensity.

[0132] B4: 1000 sets of wind power and photovoltaic output scene data are reduced according to the backward reduction method, and the probability of each scene after reduction is calculated. The wind and photovoltaic output scene with the highest probability is the wind power and photovoltaic output forecast data for the next 24 hours, such as Figure 5 and Figure 6 As shown. Figure 5 and Figure 6 It can be seen that the output curves of the five typical day wind and solar scenes are generally similar.

[0133] In this embodiment, the occurrence probability of each typical day power generation scenario can be determined by calculating the Spearman correlation coefficient between the actual power generation scenario. The calculation formula is as follows:

[0134]

[0135] Where z is the number of output scenario data points, l r is the level difference between the two sets of scene data after sorting from small to large. The probability calculation formula for the output scene is as follows:

[0136]

[0137] In the formula, Γ Si represents the probability of the scene appearing; u is the number of scenes; ρ r is the Spearman correlation coefficient between the rth output scenario and the actual output. The calculated probabilities of the five typical daily output scenarios are shown in Table 2. The scenario with the highest probability is taken as the final wind and solar output scenario for example analysis.

[0138] Table 2 Probability of occurrence of wind and solar output scenarios on a typical day

[0139]

[0140] In the upper model of this embodiment, the goal is to minimize the power generation cost of the wind, solar, storage and fuel-fired virtual power plant, and consider the constraints such as gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, power balance, etc. The single-unit guided objective function comparison algorithm is used to optimize and solve the virtual power plant optimization scheduling objective function, and plan and decide on the output of each energy source in the next 24 hours. The specific process is as follows:

[0141] The energy sources are coordinated and optimized with the goal of minimizing the operating cost. The objective function of the scheduling model can be expressed as:

[0142]

[0143] In the formula, s represents the photovoltaic output scenario, w represents the wind power output scenario; t is the trading period; The cost of purchasing electricity from the grid for the VPP; For gas turbine operating costs; The operating cost of energy storage.

[0144] The specific calculation formulas for each cost are as follows:

[0145]

[0146] In the formula, is the VPP power purchase price, in yuan / MW·h, is the power purchase amount of VPP, in MW·h.

[0147]

[0148] In the formula, c MT is the unit power generation cost of the gas turbine, in yuan / MW·h, is the power generation of the gas turbine, in MW·h.

[0149]

[0150] In the formula, is the real-time electricity price in the electricity market, η c is the charging efficiency of energy storage, is the charging capacity of energy storage, in MW·h.

[0151] Set the model constraints as follows:

[0152] Gas turbine constraints:

[0153]

[0154] Where P min and Pmax are the minimum and maximum output power of the gas turbine, is the current output power of the gas turbine; r u and r d are the upward and downward ramp rates of the gas turbine, respectively.

[0155] Energy storage constraints:

[0156]

[0157] S min ≤S s,w,t ≤S max ; (14)

[0158] Where P c,max and P d,max are the maximum charging and discharging power of energy storage, respectively. and They are respectively the current charging and discharging power of the energy storage; S s,w,t is the state of charge of the energy storage, S min and S max are the minimum and maximum storage capacity of energy storage, η c and η d are the charging and discharging efficiencies of energy storage, respectively.

[0159] Power balance constraints:

[0160]

[0161] In the formula, is the user load benchmark value, that is, the load forecast data for the next 24 hours obtained by the lower model. The wind and solar output forecast data for the next 24 hours obtained by the lower model.

[0162] The single-unit guided objective function comparison algorithm is proposed to search for the optimal solution of the objective function of the scheduling model, and then plan and decide the output of each energy source in the next 24 hours. The operation steps of the single-unit guided objective function comparison algorithm are as follows:

[0163] Step 1: Define the search space according to the three constraints of gas turbine constraint, energy storage constraint and power balance constraint, set the expected value of the objective function, and select the random direction of the position vector search. The calculation formula is as follows:

[0164]

[0165] X l =X β -d β h; (17)

[0166] X r =Xβ +d β h; (18)

[0167] In the formula, h is a normalized random vector, rand(·) is a random function, k is the dimension of the solution, and X r is the position vector on the right side of the search area, X l is the position vector on the left, d is the distance between the search center and the exploration point, β is the number of iterations, X β is the position vector of the iterative β-time solution, and the position vector corresponds to the planned output value of each energy source.

[0168] Step 2: Update the position. The calculation formula is as follows:

[0169] X β+1 =X β +δ β *h*sign[f(X r )-f(X l )]; (19)

[0170] Let M = [f(X r )-f(X l )], M is the intermediate variable, then:

[0171]

[0172] Where f(·) represents the cost calculation function corresponding to each energy output plan; δ β is the search step size at the βth iteration; sign(·) represents the sign function.

[0173] Step 3: According to the output result of step 2, that is, the planned output value of each energy source, it is brought into the objective function to determine whether it meets the expected value. If it does, the solution is output; if it does not, the mutation probability W is introduced. r And update the search step size and the distance d between the search center and the exploration point, and then iterate and update the position according to step 2. The distance d between the search center and the exploration point is calculated as:

[0174]

[0175] The output solution that meets the expected value is the optimal solution to the complex object problem faced in the virtual power plant scheduling planning under distributed energy aggregation, that is, the optimal value of each energy output in the next 24 hours.

[0176] In order to verify the effectiveness of the proposed method, the historical data of summer power load in a residential area is selected for example analysis. The wind-solar-gas-storage virtual power plant used is composed of four parts: a wind power station, a photovoltaic power station, two gas turbines and energy storage. The parameters of the gas turbine and energy storage equipment are shown in Table 3.

[0177] Table 3 Gas turbine and energy storage parameters

[0178]

[0179] Since the main power generation costs of distributed energy sources in the virtual power plant come from gas turbines and energy storage, in order to minimize the cost of purchasing electricity from energy storage, it is possible to choose to charge at the lowest time of the real-time electricity price of the power system. The real-time electricity price is shown in Table 4. When the energy storage is charging, the virtual power plant purchases electricity from the grid or supplies electricity through gas turbines combined with renewable energy.

[0180] Table 4 Real-time electricity prices

[0181]

[0182] As can be seen from Table 4, there are significant differences in electricity prices at different times, and the lowest electricity price is during the off-peak period, so the cost of charging energy storage during the off-peak period is lower. Considering the load changes and renewable energy output differences at different times, after obtaining the user load information and renewable energy output information for the next 24 hours, the virtual power plant dispatches the output of each energy source as shown below: Figure 7 shown.

[0183] from Figure 8 It can be seen that the output curve of renewable energy dispatched by the virtual power plant basically coincides with the new energy production capacity, indicating that the proposed method can make full use of new energy and greatly improve the absorption rate of renewable energy.

[0184] By introducing the estimation model into Transformer, the present invention can effectively improve the generalization ability of the model, improve the model prediction accuracy, and then obtain accurate user-side load prediction information, providing a reliable data basis for further virtual power plant scheduling and planning; the proposed single-body guided objective function comparison algorithm can effectively solve the optimal value solution problem of complex objects faced in the virtual power plant scheduling and planning under the aggregation of multiple distributed energy sources, and it has few parameters, simple algorithm, and is not easy to fall into local optimality; based on the user-side load and new energy output prediction, the optimized scheduling of wind, solar, storage and fuel virtual power plants can coordinate and optimize the output of various energy sources, while promoting the stable operation of the distribution system, improving the new energy consumption rate and reducing the power generation cost.

[0185] Example 2

[0186] The present invention also provides an energy output optimization scheduling system considering random source and load virtual power plants, the system comprising:

[0187] A two-layer optimization scheduling model construction module is used to construct a two-layer optimization scheduling model of a virtual power plant based on load and new energy output data prediction and minimization of power generation costs, which includes a lower model and an upper model; the lower model is a user load prediction model based on an improved Transformer, and the improved Transformer user load prediction model includes a Transformer model and an estimation model, and the Transformer model can extract deep-level time series features, and then the user load is finally predicted through the estimation model to obtain user load prediction data;

[0188] The wind and solar forecast module is used to establish a renewable energy output model using the scenario generation method, thereby obtaining wind power and photovoltaic output forecast data for a certain period of time in the future;

[0189] An upper model construction module is used to construct an upper model based on user load forecast data and wind power and photovoltaic output forecast data. The upper model takes minimization of the power generation cost of the wind, solar, storage and fuel virtual power plant as the objective function, and considers the constraints of gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance;

[0190] The optimal value calculation module is used to optimize and solve the objective function based on the objective function of the upper model by using a monomer-guided objective function comparison algorithm, so as to obtain the optimal value of each energy output at a certain time in the future.

[0191] Other technical features of the energy output optimization scheduling system for virtual power plants considering random source and load described in the present invention are similar to the corresponding energy output optimization scheduling method for virtual power plants considering random source and load, and will not be repeated here.

[0192] Finally, the present invention also provides an energy output optimization scheduling device for virtual power plants considering random sources and loads. The scheduling device includes: a memory, a processor, and an energy output optimization scheduling program for virtual power plants considering random sources and loads stored in the memory and executable on the processor. When the energy output optimization scheduling program for virtual power plants considering random sources and loads is executed by the processor, the steps of the energy output optimization scheduling method for virtual power plants considering random sources and loads as described above are implemented.

[0193] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0194] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0195] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for optimizing energy output scheduling of a virtual power plant considering random source and load, characterized in that: The method includes: S1 constructs a two-layer optimization scheduling model for a virtual power plant based on load and new energy output data prediction and minimization of power generation costs, which includes a lower model and an upper model; the lower model is a user load prediction model based on an improved Transformer, and the improved Transformer user load prediction model includes a Transformer model and an estimation model. The Transformer model can extract deep-level time series features, and then the user load is finally predicted through the estimation model to obtain user load prediction data; S2 uses the scenario generation method to establish a renewable energy output model, thereby obtaining wind power and photovoltaic output forecast data for a certain period of time in the future; S3 constructs an upper-level model based on user load forecast data and wind power and photovoltaic output forecast data. The upper-level model takes minimizing the power generation cost of the wind, solar, storage and fuel virtual power plant as the objective function, and considers the constraints of gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance; S4 is based on the objective function of the upper model, and uses a monomer-guided objective function comparison algorithm to optimize and solve the objective function, thereby obtaining the optimal value of each energy output at a certain time in the future.

2. The energy output optimization scheduling method of a virtual power plant considering random source and load according to claim 1 is characterized in that: In step S1, the process of constructing the user load prediction model based on the improved Transformer includes: S11 data preprocessing: collect historical user load data and divide the historical user load data into training set data and test set data according to a certain ratio; S12 model construction: using the improved Transformer model to process the training set data; S13 model training and load forecasting: Use the preprocessed training set to train the improved Transformer model, set the maximum number of iterations, and when the maximum number of iterations is reached, the training is completed, and the test set is used to verify the performance of the model; finally, the load data of a certain power area on the same day is input into the trained improved Transformer model to obtain the load forecast data of the power area for several future times.

3. The energy output optimization scheduling method of a virtual power plant considering random source and load according to claim 1 or 2 is characterized in that: The estimation model is expressed as: Where n is the number of boundary vectors, α i is the weight multiplier, It is a function that maps the Transformer output data x to a higher dimensional space, x' is the boundary vector, and b is the bias term; in, The function expression is: In the formula, c ij is the autocorrelation coefficient, i,j=1,2,...u, u is the dimension of Transformer output data.

4. The method for optimizing energy output of a virtual power plant considering random source and load according to claim 3 is characterized in that: In the step S2, a renewable energy output model is established according to the lower model using a scenario generation method, specifically including: S21 uses non-parametric kernel density estimation method to fit the historical series data of wind power and photovoltaic output to obtain the kernel density estimation formula of wind power and photovoltaic; S22 establishes a joint distribution model of wind power and photovoltaic output based on Copula function; S23 uses the kernel density estimation formula to calculate the error between the estimated value obtained by the joint distribution model and the actual value of wind power and photovoltaic output, selects the optimal Copula function corresponding to the minimum error as the optimal wind power and photovoltaic joint probability distribution, and generates preliminary wind power and photovoltaic output scenario data; S24 reduces the preliminary wind power and photovoltaic output scenario data according to the backward reduction method, and calculates the probability of each scenario after the reduction. The wind and photovoltaic output scenario with the highest probability of occurrence is the wind power and photovoltaic output forecast data for a certain period of time in the future.

5. The method for optimizing energy output of a virtual power plant considering random source and load according to claim 1 is characterized in that: In step S3, the upper model takes minimizing the power generation cost of the wind, solar, storage and fuel-fired virtual power plant as the objective function, and the objective function is expressed as: In the formula, s represents the photovoltaic output scenario, and w represents the wind power output scenario; t is the trading period; The cost of purchasing electricity from the grid for the VPP; For gas turbine operating costs; The operating cost of energy storage; The specific calculation formulas for each cost are as follows: In the formula, is the VPP power purchase price, in yuan / MW·h, is the amount of electricity purchased by the VPP, in MW·h; In the formula, c MT is the unit power generation cost of the gas turbine, in yuan / MW·h, is the power generation of the gas turbine, in MW·h; In the formula, is the real-time electricity price in the electricity market, η c is the charging efficiency of energy storage, is the charging capacity of energy storage, in MW·h.

6. The method for optimizing energy output of a virtual power plant considering random source and load according to claim 1 is characterized in that: In step S3, the gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance constraints are considered, including: Gas turbine constraints: Where P min and P max are the minimum and maximum output power of the gas turbine, is the current output power of the gas turbine; r u and r d are the upward and downward ramp rates of the gas turbine, respectively; Energy storage constraints: Where P c,max and P d,max are the maximum charging and discharging power of energy storage, respectively. and They are respectively the current charging and discharging power of the energy storage; S s,w,t is the state of charge of the energy storage, S min and S max are the minimum and maximum storage capacity of energy storage, η c and η d are the charging and discharging efficiencies of energy storage, respectively; Power balance constraints: In the formula, is the user load benchmark value, that is, the load forecast data for a certain period of time in the future obtained by the lower model. The wind power and photovoltaic output forecast data for several future periods of time obtained by the lower model.

7. The method for optimizing energy output of a virtual power plant considering random source and load according to claim 6 is characterized in that: In the step S4, based on the objective function of the upper model, the objective function is optimized and solved by using a monomer-guided objective function comparison algorithm, including: S41 defines the search space according to the three constraints of gas turbine constraint, energy storage constraint and power balance constraint, sets the expected value of the objective function, and selects the random direction of the position vector search. The calculation formula is as follows: X l =X β -d β h; X r =X β +d β h; In the formula, h is a normalized random vector, rand(·) is a random function, k is the dimension of the solution, i.e., the 24-hour output value of each energy source, X r is the position vector on the right side of the search area, X l is the position vector on the left side of the search area, d is the distance between the search center and the exploration point, β is the number of iterations, and X β is the position vector of the iterative β-time solution, which corresponds to the planned output value of each energy source; S42 updates the position, and the calculation formula is as follows: X β+1 =X β +δ β *h*sign[f(X r )-f(X l )]; Let M = [f(X r )-f(X l )], M is the intermediate variable, then: Where f(·) represents the cost calculation function corresponding to each energy output plan; δ β is the search step size at the βth iteration; sign(·) represents the sign function; S43 brings the output result of step S42, that is, the planned output value of each energy source, into the objective function to determine whether it meets the expected value. If it does, the solution is output; if it does not, the mutation probability W is introduced. r And update the search step size and the distance d between the search center and the exploration point, and then iteratively update the position according to step S42. The output solution that meets the expected value is the optimal solution to the complex object problem faced in the virtual power plant scheduling planning under distributed energy aggregation, that is, the optimal value of each energy output in the future.

8. The method for optimizing energy output of a virtual power plant considering random source and load according to claim 7 is characterized in that: The calculation formula of the distance d between the search center and the exploration point is expressed as:

9. An energy output optimization scheduling system considering random source and load virtual power plants, characterized in that: The system includes: A two-layer optimization scheduling model construction module is used to construct a two-layer optimization scheduling model of a virtual power plant based on load and new energy output data prediction and minimization of power generation costs, which includes a lower model and an upper model; the lower model is a user load prediction model based on an improved Transformer, and the improved Transformer user load prediction model includes a Transformer model and an estimation model, and the Transformer model can extract deep-level time series features, and then the user load is finally predicted through the estimation model to obtain user load prediction data; The wind and solar forecast module is used to establish a renewable energy output model using the scenario generation method, thereby obtaining wind power and photovoltaic output forecast data for a certain period of time in the future; An upper model construction module is used to construct an upper model based on user load forecast data and wind power and photovoltaic output forecast data. The upper model takes minimization of the power generation cost of the wind, solar, storage and fuel virtual power plant as the objective function, and considers the constraints of gas turbine output power, ramp rate, energy storage charging and discharging power, storage capacity, and power balance; The optimal value calculation module is used to optimize and solve the objective function based on the objective function of the upper model by using a monomer-guided objective function comparison algorithm, so as to obtain the optimal value of each energy output at a certain time in the future.

10. An energy output optimization scheduling device considering random source and load virtual power plants, characterized in that: The scheduling device includes: a memory, a processor, and an energy output optimization scheduling program for virtual power plants considering random sources and loads, which is stored in the memory and can be run on the processor. When the energy output optimization scheduling program for virtual power plants considering random sources and loads is executed by the processor, the steps of the energy output optimization scheduling method for virtual power plants considering random sources and loads are implemented as described in any one of claims 1 to 8.

Images