Day-ahead optimal dispatch method for power system based on evolving virtual net load

Abstract

The application discloses a kind of new power system day-ahead optimal scheduling methods based on evolution virtual net load, belong to power system day-ahead optimal scheduling technical field.Utilize Informer model to the time series of wind and light output and load are carried out scene generation;Evolution virtual net load is defined on the basis of virtual net load, and the evolution virtual net load of power system is calculated;The operation characteristics of system are considered, to ensure the renewable energy power purchase is based on, to include the total cost of system including the operation cost of each unit and carbon emission cost Minimum as target, with system flexibility constraint condition reflects the demand of source and load uncertainty, establish new power system day-ahead optimal scheduling model based on evolution virtual net load;The model is solved, and the new power system day-ahead optimal scheduling plan based on evolution virtual net load is obtained.This method can effectively improve the ability of system to cope with source and load uncertainty fluctuation, reduce carbon emission cost and promote renewable energy power generation project on-grid power consumption.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of day-ahead optimal dispatching of power system in power system planning and operation, and particularly relates to a new type of day-ahead optimal dispatching method of power system based on evolved virtual net load. BACKGROUND

[0002] In the face of the problem of optimizing dispatching of new type of power system containing high proportion of renewable energy under the condition of considering uncertain factors, the early dispatching method research considering uncertain factors mainly focuses on chance constraint method, robust optimization method and traditional scene generation method, which has achieved certain effect, but increases the complexity of the model and is relatively cumbersome to calculate. At present, the renewable energy power consumption problem in power system has been deeply studied, but most of the researches are based on the Renewable Energy Power Purchase Supervision Measures of Power Grid Enterprises (hereinafter referred to as the "old measures") to develop dispatching method and power consumption plan. These methods and plans are no longer applicable after the Implementation of Renewable Energy Power Purchase Supervision Measures (hereinafter referred to as "measures").

[0003] "Net load" refers to the difference between the total load and wind and solar power generation in a certain period of the power system. Its concept provides convenience for the research of power system planning and operation, power grid stability analysis and demand side management. "Virtual load" is a multi-source heterogeneous controllable integrated load. Through the application of advanced communication technology and software architecture, diversified load resources distributed in different places are integrated, and the identification, complementation and coordinated management of these loads are realized. The virtual load management mode is proposed to promote the overall development of smart grid and ensure the efficient use of power resources and the stable and safe operation of power grid. And "virtual net load" refers to the load obtained by additionally removing the minimum technical output of thermal power and the forced output of hydropower from the net load. It unifies the research of power source operation constraints and source load uncertainty, making the power grid dispatching clear and simple. However, the virtual net load defined based on the "old measures" is no longer applicable after the "old measures" are abolished, and its dispatching method loses its theoretical basis. In order to distinguish from "net load", "virtual load" and "virtual net load" and implement the "measures", the "evolved virtual net load" is defined as follows: in a power system containing wind, solar, water and pumped storage, the total load in a certain period minus the power corresponding to the guaranteed purchase of renewable energy (wind, solar and water) and the minimum technical output of thermal power, and the obtained load is called the evolved virtual net load of the period.

[0004] Therefore, there is an urgent need in the prior art for a new type of day-ahead optimal dispatching method of power system considering wind and solar power generation and load power prediction and evolved virtual net load. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a new type of power system day-ahead optimal scheduling method based on evolving virtual net load, which is used to solve the problem of adverse effects of source and load uncertainty in the prior art on the formulation of scheduling plan, and the lack of full guarantee of renewable energy power consumption.

[0006] The technical solution adopted by the present application is to provide a new type of power system day-ahead optimal scheduling method based on evolving virtual net load, comprising the following steps:

[0007] 1) generating scenarios for the time series of wind and light output and load by using an Informer model, and establishing a wind and light generation and load power prediction model based on the Informer model;

[0008] 2) defining an evolving virtual net load on the basis of the virtual net load, and calculating the evolving virtual net load of the power system;

[0009] 3) considering the operating characteristics of the system, taking the guaranteed purchase of renewable energy power as the basis, taking the minimum total cost of the system including the operating cost and carbon emission cost of each unit as the target, and taking the system flexibility constraint condition to reflect the demand of source and load uncertainty, establishing a day-ahead optimal scheduling model based on the evolving virtual net load;

[0010] 4) solving the model to obtain a new type of power system day-ahead optimal scheduling plan based on the evolving virtual net load.

[0011] The specific method for establishing the wind and light generation and load power prediction model based on the Informer model in step 1) is as follows:

[0012] ① The historical data of wind and light generation power and load power are normalized respectively, and then the normalized data set is divided into a training set, a test set and a validation set, and then the historical power data w p and the timestamp feature w time are reconstructed and combined to obtain W E ={w p , w time}, and W E is taken as the input data of the prediction model;

[0013] ② The long-term dependence relationship between data is mined through the multi-head probability sparse self-attention layer of the encoder in the prediction model to reduce the time complexity and realize the extraction of the wind and light generation power and load power sequence features;

[0014] ③ The output of each probability sparse self-attention layer is subjected to convolution and pooling operation through the self-attention distillation layer of the encoder in the prediction model, the output of each layer is halved on the basis of retaining the key information of data, and the calculation amount of the network is reduced;

[0015] IV. reconstructing and combining the partial historical power data of the encoder and the future power placeholder data w p and its corresponding timestamp feature w time to obtain W D ={w p ′, w time ′} as the input of the decoder in the prediction model, and performing attention calculation on the combined sequence of the input historical power and zero value through the masked multi-head probability sparse self-attention layer, in the process, the masked mechanism can be introduced to prevent the model from falling into autoregression, and the combined sequence feature can be extracted;

[0016] V. inputting the feature vector processed by the masked multi-head probability sparse self-attention layer in the decoder and the feature vector output by the encoder into the multi-head attention layer in the decoder to perform multi-head attention calculation, and outputting the calculation result using a generative method to realize one-time generation of the output sequence, thereby shortening the dynamic decoding time and reducing the memory occupation;

[0017] VI. setting a full connection layer to convert the output dimension of the decoder to the output dimension required for prediction, and finally performing inverse normalization processing on the output result to obtain the predicted value of the wind and solar power generation and and the predicted value of the load power P L .

[0018] The calculation method of the evolving virtual net load in the step 2) is formula (1),

[0019]

[0020] In the formula, P VL,t is the evolving virtual net load of the system at the t period; P L,t is the total load of the system at the t period; β 1,t is the wind and solar guarantee purchase coefficient at the t period; β 2,t is the hydropower guarantee purchase coefficient at the t period; and are the maximum power generation of the wind power, the photovoltaic power and the hydropower at the t period respectively; N h and N g are the total number of the hydropower units and the thermal power units respectively; is the minimum technical output of the thermal power unit i at the t period.

[0021] The objective function of the day-ahead optimal dispatching model based on the evolving virtual net load in the step 3) is formula (2),

[0022] min C=C g +C h +C w +C pv+C ps +C carbon (2)

[0023] wherein C is the total cost of the system; C g is the operation cost of thermal power units; C h is the operation cost of hydropower units; C w is the operation cost of wind farms; C pv is the operation cost of photovoltaic power stations; C ps is the start-stop cost of pumped storage units; and C carbon is the carbon emission cost.

[0024] The operation cost C g of thermal power units, the operation cost C h of hydropower units, the operation cost C w of wind farms, the operation cost C pv of photovoltaic power stations, the start-stop cost C ps of pumped storage units, and the carbon emission cost C carbon are calculated according to formulas (3)-(8),

[0025] a The operation cost C g of thermal power units is calculated according to formula (3),

[0026]

[0027] wherein T is the total number of time periods in the dispatching period; a i , b i , and c i are the cost coefficients of the thermal power unit i, respectively; P g,i,t is the power generation of the thermal power unit i in the t time period.

[0028] b The operation cost C h of hydropower units is calculated according to formula (4),

[0029]

[0030] wherein k h is the unit power generation cost coefficient of the hydropower unit; P h,i,t is the power generation of the hydropower unit i in the t time period.

[0031] c The operation cost C w of wind farms is calculated according to formula (5),

[0032] C w = k w P w,t (5)

[0033] wherein k w is the power generation cost coefficient of the wind farm; P w,tLet t be the power generation capacity of the wind farm during time period t;

[0034] d Photovoltaic power plant operating cost C pv See formula (6),

[0035] C pv =k pv P pv,t (6)

[0036] In the formula, k pv P represents the power generation cost coefficient for photovoltaic power generation. pv,t Let t be the power generation capacity of the photovoltaic power station during time period t;

[0037] C. Start-up and shutdown costs of e-pumped storage units ps See formula (7),

[0038]

[0039] In the formula, N ps This represents the total number of pumped storage units. These are the power generation and pumping state variables of pumped storage unit i during time period t, respectively. They are 1 when the unit is working and 0 when it is stopped. These are the power generation and pumping costs of pumped storage unit i, respectively.

[0040] f Carbon emission cost C carbon See formula (8),

[0041]

[0042] In the formula, C carbon η represents the carbon emission cost; E represents the carbon trading price; P E represents the total carbon emissions during the scheduling cycle. L The total carbon emission allowance within a scheduling cycle; δ i ε represents the carbon emissions of thermal power unit i; ε is the unit electricity quota coefficient for carbon emissions of the system.

[0043] The constraints of the day-ahead optimization scheduling model based on evolving virtual net load in step 3) include:

[0044] The operating constraints of thermal power unit a are given in formula (9).

[0045]

[0046] In the formula, These represent the upper and lower limits of the output of thermal power unit i, respectively; P g,i,t-1 The dispatch output of thermal power unit i during time period t-1; These are the upper and lower limits of the ramp rate for thermal power unit i, respectively.

[0047] b The water turbine operation constraint is shown in equation (10),

[0048]

[0049] where β2 is the water power guarantee purchase coefficient, which is 0.4, is the market transaction power corresponding to the market transaction electricity of the water turbine i in the t period; are the upper and lower limits of the output of the water turbine i, respectively; h,i,t-1 is the dispatch output of the water turbine i in the t-1 period; are the upper and lower limits of the ramping of the water turbine i, respectively;

[0050] c The wind farm constraint is shown in equation (11),

[0051]

[0052] where β1 is the non-water power guarantee purchase coefficient, which is 0.3, is the market transaction power corresponding to the market transaction electricity of the wind farm in the t period;

[0053] d The photovoltaic power station constraint is shown in equation (12),

[0054]

[0055] where β1 is the non-water power guarantee purchase coefficient, which is 0.3, is the market transaction power corresponding to the market transaction electricity of the photovoltaic power station in the t period;

[0056] e The pumped storage unit operation constraint is shown in equation (13),

[0057]

[0058] where P ps,i,t and P pm,i,t are the power generation and pumping power of the pumped storage unit i in the t period, respectively; are the upper limits of the power generation and pumping power of the pumped storage unit i, respectively; are the upper and lower reservoir capacities in the t period, respectively; are the upper limits of the upper and lower reservoir capacities, respectively; are the lower limits of the upper and lower reservoir capacities, respectively; is the upper reservoir capacity in the t-1 period; are the power generation water consumption and pumping amount of the pumped storage unit i in the t period, respectively; η gen and η pm are the water-electricity conversion coefficients of the power generation and pumping of the pumped storage unit i, respectively;

[0059] f The power balance constraint is shown in equation (14),

[0060]

[0061] wherein, is the peak shaving power of thermal power at time t;

[0062] g flexibility constraints include flexibility demand constraints as shown in equation (15), flexibility supply constraints as shown in equations (16)-(17) and flexibility margin constraints as shown in equation (18),

[0063]

[0064] wherein, ΔP VL,t is the virtual net load change between the next time period and time t; F t de,up , F t de,dn are the upward and downward flexibility demands of the system at time t, respectively; P w,t+1 , P pv,t+1 , P L，t+1 are the predicted outputs of wind power, photovoltaic power and load at time t+1; ω u and ω d , γ u and γ d , λ u and λ d are the upward and downward flexibility demand coefficients of wind power, photovoltaic power and load caused by prediction errors, respectively; λ e is the upward flexibility demand coefficient caused by unit fault repair;

[0065]

[0066] wherein, F t su,up is the upward flexibility supply of the system at time t; is the upward flexibility supply of thermal power, hydropower and pumped storage units at time t;

[0067]

[0068] wherein, F t su,dn is the downward flexibility supply of the system at time t; is the downward flexibility supply of thermal power, hydropower and pumped storage units at time t;

[0069]

[0070] wherein, F t up , F t dn represent the upward and downward flexibility margins of the system at time t, respectively.

[0071] Through the above design scheme, the present application can bring the following beneficial effects:

[0072] a) The present application can learn the long-term dependence and complex nonlinear relationship in time series in the wind and light power generation and load power prediction link, and has good effect on the prediction object driven by multiple factors and complex dynamic characteristics.

[0073] b) For renewable energy power generation enterprises, the present application can well implement the guarantee purchase mechanism, realize full consumption of the guarantee purchase power in the renewable energy on-grid power, make it obtain stable income; at the same time, it can encourage power generation enterprises to fully utilize marketized bidding opportunities, reduce power generation cost through technological progress to strive for more transaction opportunities, and make it obtain greater income.

[0074] c) For the power system dispatching department, the present application considers the power source flexibility constraint condition while tracking the evolving virtual net load, which can not only ensure the safe operation of the system, but also effectively reduce the power generation cost of each unit and improve the carbon income, so as to improve the overall economic benefit of the system. BRIEF DESCRIPTION OF DRAWINGS

[0075] Figure 1 It is a basic principle diagram of the new type of power system day-ahead optimal dispatching method based on the evolving virtual net load of the present application;

[0076] Figure 2 It is a wind and light power generation and load power prediction model diagram based on the Informer model of the new type of power system day-ahead optimal dispatching method based on the evolving virtual net load of the present application;

[0077] Figure 3 It is a full-province wind power prediction curve diagram of the new type of power system day-ahead optimal dispatching method based on the evolving virtual net load of the present application;

[0078] Figure 4 It is a full-province light power prediction curve diagram of the new type of power system day-ahead optimal dispatching method based on the evolving virtual net load of the present application;

[0079] Figure 5 It is a full-province load prediction curve diagram of the new type of power system day-ahead optimal dispatching method based on the evolving virtual net load of the present application;

[0080] Figure 6 It is a full-province virtual net load curve and evolving virtual net load curve diagram of the new type of power system day-ahead optimal dispatching method based on the evolving virtual net load of the present application;

[0081] Figure 7The method 1 scheduling after each unit output curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application;

[0082] Figure 8 The method 2 scheduling after each unit output curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application;

[0083] Figure 9 The method 3 scheduling after each unit output curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application;

[0084] Figure 10 The method 4 scheduling after each unit output curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application;

[0085] Figure 11 The method 4 scheduling after each unit output curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application;

[0086] Figure 12 The upward flexibility margin curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application;

[0087] Figure 13 The downward flexibility margin curve diagram for the new type of power system day-ahead optimal dispatch method based on the evolving virtual net load of the application. DETAILED DESCRIPTION

[0088] The application will be further described below in combination with the drawings and specific embodiments:

[0089] Specific embodiment: a new type of power system day-ahead optimal dispatch method based on the evolving virtual net load, the basic principle schematic diagram is Figure 1 , which comprises: constructing a prediction model of wind and light power generation and load power by using an Informer model, calculating the evolving virtual net load, establishing a new type of power system day-ahead optimal dispatch model based on the evolving virtual net load, and solving the new type of power system day-ahead optimal dispatch model based on the evolving virtual net load, and the specific content is:

[0090] 1) Parameter setting and scene prediction of wind and light power and load

[0091] The parameters of the hydropower unit and other parameters are shown in Table 1 and Table 2;

[0092] Table 1: Parameters of hydropower unit

[0093]

[0094] Table 2 other parameters

[0095]

[0096] Figure 2 For the wind and light power generation and load power prediction model graph based on the Informer model, the Informer model is used to construct the wind and light power generation and load power prediction model, which can overcome the shortcomings of high time complexity, high memory occupancy and the fixed limitations of the encoder-decoder architecture of the Transformer model, and is more excellent in processing and predicting data with complex time dependence.

[0097] Figure 3 、 Figure 4 and Figure 5 For the wind and light power generation and load power prediction model graph based on the Informer model, the Informer model is used to construct the wind and light power generation and load power prediction model, which can overcome the shortcomings of high time complexity, high memory occupancy and the fixed limitations of the encoder-decoder architecture of the Transformer model, and is more excellent in processing and predicting data with complex time dependence.

[0098] Figure 6 For the wind and light power generation and load power prediction model graph based on the Informer model, the Informer model is used to construct the wind and light power generation and load power prediction model, which can overcome the shortcomings of high time complexity, high memory occupancy and the fixed limitations of the encoder-decoder architecture of the Transformer model, and is more excellent in processing and predicting data with complex time dependence.

[0099] 2) Solve the new power system day-ahead optimal dispatching model based on the evolution virtual net load

[0100] Four different dispatching methods are designed for comparative analysis, and the design methods are as follows:

[0101] Method 1 is to use the prediction model of the application to predict the wind and light output and load power, on the basis of meeting the operating constraints of each unit and the system power balance constraints, according to the constant proportion of the total load of the system to set the rotating reserve capacity, and optimize the dispatching.

[0102] Method 2 is based on method 1, to ensure that when the wind and light output, load power and expected maximum deviation, the system can still meet the rotating reserve constraint (that is, considering the uncertainty of wind and light output and load power in the rotating reserve constraint), and optimize the dispatching.

[0103] Method 3 is to use the ARIMA model-based method to predict the wind and light output and load power scene, and use the scheduling method of method 1 to optimize scheduling.

[0104] Method 4 is to use the ARIMA model-based method to predict the wind and light scene and load power scene, and use the scheduling method of method 2 to optimize scheduling.

[0105] Figures 7-11 After using the new power system day-ahead optimization scheduling method based on the evolution of the virtual net load, method 1, method 2, method 3 and method 4 optimization scheduling method, the unit output curve of the power system is obtained.

[0106] Figure 12 For the upward flexibility margin of different scheduling methods, the upward flexibility margin of the five scheduling methods is positive, which can ensure the upward adjustment ability required for stable operation of the system, but too high flexibility margin may lead to cost increase and resource waste, thereby affecting the economy of system operation, therefore, under the condition of positive margin, lower upward flexibility margin is more conducive to improving economy. Through the analysis of the upward flexibility margin curve of each method, it can be seen that the proposed scheduling method can reduce resource waste on the basis of ensuring the necessary flexibility of the system, and has more advantages in economy.

[0107] Figure 13 For the downward flexibility margin of different scheduling methods, the smaller the downward flexibility margin, the better the economy of the system operation, but if the downward flexibility margin is lower than zero, it may mean that the system lacks in dealing with load fluctuation and renewable energy generation uncertainty, and it is difficult to meet the power balance demand, affecting the safety of the system. Through the analysis of the downward flexibility margin curve of each method, it can be seen that the proposed scheduling method can also improve economic benefits on the basis of ensuring the safe and stable operation of the system.

[0108] The daily operation cost of the system after using different scheduling methods for optimization scheduling is shown in Table 3. Compared with other scheduling methods, the operation cost of each unit of the scheduling method proposed in the application is relatively small, and the carbon benefit is higher, so as to reduce the overall power generation cost of the system, and the economy is better.

[0109] Table 3 System daily operation cost under different scheduling methods

[0110]

[0111] Through comparative analysis of the flexibility margin curve and daily power generation cost of the system, it can be seen that:

[0112] Method 1 focuses on maintaining sufficient rotational reserve to ensure the safe operation of the system, but does not fully consider the uncertainty of wind and light and load, resulting in excessive flexibility margin, and the operation cost of each unit of this method is high, which affects the overall economy and safety of the system.

[0113] Method 2 considers the uncertainty of wind and light power generation in scheduling to ensure the safety of the system in extreme cases. Compared with method 1, although it can reduce the excess flexibility margin, this method is still conservative in reducing the operation cost of the unit and promoting renewable energy consumption. The scheduling result is too focused on safety and damages the economy.

[0114] Methods 3 and 4 use ARIMA models to predict wind and light power generation and load power scenarios. The model only relies on historical power data for prediction without additional variables, but the model requires the input historical data to be close to the Gaussian distribution, and when the non-stationary sequence is differentiated to achieve stationarity, excessive differentiation may lose key information of the data, thereby affecting the accuracy of the scenario prediction. Compared with the prediction method proposed in the present application, methods 3 and 4 perform poorly in terms of the economy of scheduling.

[0115] The scheduling method proposed in the present application uses a wind and light power generation and load power prediction model based on the Informer model for scenario prediction. Through its advanced attention mechanism, it can better extract time-dependent features from data, mine and learn the complex nonlinear relationship in power data, thereby nearly accurately predicting wind and light power generation and load power. The optimization scheduling method proposed in the present application not only ensures the stable operation of the power system, but also effectively avoids the redundant consumption of resources by reducing unnecessary flexibility margin. At the same time, the method can complete the consumption of renewable energy on-grid power by optimizing the load curve operation. Compared with the other four methods, the scheduling method proposed in the present application has more significant advantages in improving economic benefits and ensuring the safe and stable operation of the system.

[0116] The embodiments of the present application are not limited by the above examples, and any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principles of the present application shall be equivalent replacement methods and shall be included in the protection scope of the present application.

Claims

1. A power system day-ahead optimal dispatching method based on evolving virtual net load, characterized in that: Includes the following steps: 1) Use the Informer model to generate scenarios from time series of wind and solar power output and load, and establish a wind and solar power generation and load prediction model based on the Informer model; 2) Define the evolving virtual net load based on the virtual net load, and calculate the evolving virtual net load of the power system; The formula for calculating the evolved virtual net load is: In the formula, P VL,t P represents the virtual net load of the system during time period t; L,t β represents the total load of the system during time period t; 1,t The guaranteed purchase coefficient for wind and solar power during time period t; β 2,t The hydropower guaranteed purchase coefficient for time period t; and These represent the maximum power generation capacity achievable by wind power, solar power, and hydropower during time period t; N h N g These represent the total number of hydropower units and thermal power units, respectively. The minimum technical output of thermal power unit i during time period t; 3) Considering the operating characteristics of the system, based on the guaranteed purchase of renewable energy, with the goal of minimizing the total system cost including the operating cost and carbon emission cost of each unit, and with the system flexibility constraints reflecting the uncertainty of source and load, a day-ahead optimization scheduling model based on the evolving virtual net load is established. Flexibility constraints include flexibility demand constraints, flexibility supply constraints, and flexibility margin constraints. The formula for the flexibility demand constraint is: In the formula, ΔP VL,t F represents the virtual net load change between the next time period and time period t. t de,up F t de,dn These represent the upward and downward flexibility requirements of the system during time period t; P w,t+1 P pv,t+1 P L，t+1 Forecast power output for wind power, solar power, and load during period t+1; ω u and ω d γ u and γ d , λ u and λ d These are the upside and downside flexibility demand coefficients for wind power, solar power, and load, respectively, resulting from forecast errors; λ e P represents the upward flexibility requirement factor during unit failure maintenance; VL,t+1 The virtual net load representing the evolution of the system during time period t+1; The formula for flexible supply constraints is: In the formula, F t su,up Provides upward flexibility for the system during time period t; To provide upward flexibility for thermal power, hydropower, and pumped storage units during time period t; In the formula, F t su,dn Provides downward flexibility for the system during time period t; For the downward flexibility supply of thermal power, hydropower and pumped storage units during time period t; The formula for flexibility margin constraint is: In the formula, F t up F t dn These represent the upward and downward flexibility margins of the system during time period t, respectively. 4) Solve the model to obtain the day-ahead optimal dispatch plan of the power system based on the evolving virtual net load.

2. The day-ahead optimal dispatching method for power systems based on evolving virtual net load according to claim 1, characterized in that: The specific method for establishing the wind and solar power generation and load power prediction model based on the Informer model in step 1) is as follows: ① Normalize the historical data of wind and solar power generation and load power respectively. Then, divide the normalized dataset into training set, test set and validation set. Then, analyze the historical power data w p and timestamp features w time Reconstruct and combine to obtain W E ={w p w time }, and with W E As input data for the predictive model; ② By mining long-term dependencies between data through the multi-head probabilistic sparse self-attention layer of the encoder in the prediction model, the time complexity can be reduced and the features of the input wind and solar power generation and load power sequences can be extracted; ③ By performing convolution and pooling operations on the output of each probability sparse self-attention layer through the self-attention distillation layer of the encoder in the prediction model, the output of each layer is halved while retaining the key information of the data, thereby reducing the computational load of the network. ④ The w is formed by combining some historical power data of the encoder with future power placeholder data. p ′ and its corresponding timestamp feature w time ′ is reconstructed and combined to obtain W D ={w p ′, w time As the input of the decoder in the prediction model, the attention calculation is performed on the combination sequence of the input historical power and zero value through the masked multi-head probabilistic sparse self-attention layer. In this process, the masking mechanism can be introduced to prevent the model from falling into autoregression, and the features of the combination sequence can be extracted. ⑤ Input the feature vector processed by the mask multi-head probability sparse self-attention layer in the decoder and the feature vector output by the encoder into the multi-head attention layer in the decoder for multi-head attention calculation, and output the calculation result using a generative method to realize the one-time generation of the output sequence, thereby shortening the dynamic decoding time and reducing memory usage; ⑥ Set up a fully connected layer to transform the output dimension of the decoder into the output dimension required for prediction, and finally perform inverse normalization on the output result to obtain the predicted value of wind and solar power generation. and With the predicted load power P L .

3. The day-ahead optimal dispatching method for power systems based on evolving virtual net load according to claim 1, characterized in that: The objective function formula for the day-ahead optimization scheduling model based on the evolved virtual net load in step 3) is: minC=C g +C h +C w +C pv +C ps +C carbon In the formula, C represents the total system cost; C g C is the operating cost of thermal power units; h C is the operating cost of the hydropower unit; w C is the operating cost of the wind farm. pv C represents the operating cost of a photovoltaic power plant. ps C is the start-up and shutdown cost of pumped storage units; carbon Cost of carbon emissions.

4. The day-ahead optimal dispatching method for power systems based on evolving virtual net load according to claim 3, characterized in that: The operating cost C of the thermal power unit g Operating cost of hydropower units C h Wind farm operating costs C w and the operating cost of photovoltaic power plants C pv And the start-up and shutdown costs of pumped storage units, C ps And carbon emission costs C carbon The calculation formula is as follows: Operating cost of thermal power unit C g : In the formula, T represents the total number of time periods within the scheduling cycle; a i b i and c i These are the cost coefficients for thermal power unit i; P g,i,t Let be the power generation capacity of thermal power unit i during time period t; b. Operating cost of hydropower unit C h : In the formula, k h P is the unit power generation cost coefficient for hydropower units. h,i,t Let be the power generation capacity of hydropower unit i during time period t; c Wind farm operating costs C w : C w =k w P w,t In the formula, k w P represents the power generation cost coefficient for wind farms. w,t Let t be the power generation capacity of the wind farm during time period t; d Photovoltaic power plant operating cost C pv : C pv =k pv P pv,t In the formula, k pv P represents the power generation cost coefficient for photovoltaic power generation. pv,t Let t be the power generation capacity of the photovoltaic power station during time period t; C. Start-up and shutdown costs of e-pumped storage units ps : In the formula, N ps This represents the total number of pumped storage units. These are the power generation and pumping state variables of pumped storage unit i during time period t, respectively. They are 1 when the unit is working and 0 when it is stopped. These are the power generation and pumping costs of pumped storage unit i, respectively. f Carbon emission cost C carbon : In the formula, C carbon η represents the carbon emission cost; E represents the carbon trading price; P E represents the total carbon emissions during the scheduling cycle. L The total carbon emission allowance within a scheduling cycle; δ i ε represents the carbon emissions of thermal power unit i; ε is the unit electricity quota coefficient for carbon emissions of the system.

5. The day-ahead optimal dispatching method for power systems based on evolving virtual net load according to claim 4, characterized in that: The constraints of the day-ahead optimization scheduling model based on evolving virtual net load in step 3) include: Operating constraints of thermal power units: In the formula, These represent the upper and lower limits of the output of thermal power unit i, respectively; P g,i,t-1 The dispatch output of thermal power unit i during time period t-1; These represent the upper and lower limits of the ramp rate for thermal power unit i, respectively; U i,t Let be the operating and shutdown status variable of thermal power unit i during time period t, which is 1 when operating and 0 when shutting down; b. Operating constraints of hydropower units: In the formula, the hydropower guaranteed purchase coefficient β2 is 0.

4. The power corresponding to the market transaction volume of hydropower unit i during time period t; These represent the upper and lower limits of the output of hydropower unit i, respectively; P h,i,t-1 The scheduled output of hydropower unit i during time period t-1; These are the upper and lower limits of the ramp rate for hydropower unit i, respectively. c. Wind farm constraints: In the formula, the non-hydropower guaranteed purchase coefficient β1 is 0.

3. The power corresponding to the market transaction volume of the wind farm during time period t; d. Constraints of photovoltaic power plants: In the formula, the non-hydropower guaranteed purchase coefficient β1 is 0.

3. The power corresponding to the market transaction volume of the photovoltaic power station during time period t; Operating constraints of pumped storage units: In the formula, P ps,i,t P pm,i,t These represent the power generation and pumping power of pumped storage unit i during time period t, respectively. These are the upper limits of power generation and pumping power for pumped storage unit i, respectively. These represent the upper and lower reservoir capacities during time period t; These are the upper and lower reservoir capacity limits, respectively. These are the lower limits of the upper and lower reservoir capacities, respectively. The reservoir capacity during time period t-1; Let η represent the water consumption for power generation and the water pumping rate of pumped storage unit i during time period t; gen η pm These are the hydroelectric conversion coefficients for power generation and pumping of the pumped storage unit i, respectively. These are the power generation and pumping state variables of pumped storage unit i during time period t-1, respectively. They are 1 when the unit is working and 0 when it is stopped. f. Power balance constraint: In the formula, The peak-shaving power of thermal power during time period t; The flexibility constraints include flexibility demand constraints, flexibility supply constraints, and flexibility margin constraints. The formula for the flexibility demand constraint is: In the formula, ΔP VL,t F represents the virtual net load change between the next time period and time period t. t de,up F t de,dn These represent the upward and downward flexibility requirements of the system during time period t; P w,t+1 P pv,t+1 P L，t+1 Forecast power output for wind power, solar power, and load during period t+1; ω u and ω d γ u and γ d , λ u and λ d These are the upside and downside flexibility demand coefficients for wind power, solar power, and load, respectively, resulting from forecast errors; λ e P represents the upward flexibility requirement factor during unit failure maintenance; VL,t+1 The virtual net load representing the evolution of the system during time period t+1; The formula for flexible supply constraints is: In the formula, F t su,up Provides upward flexibility for the system during time period t; To provide upward flexibility for thermal power, hydropower, and pumped storage units during time period t; In the formula, F t su,dn Provides downward flexibility for the system during time period t; For the downward flexibility supply of thermal power, hydropower and pumped storage units during time period t; The formula for flexibility margin constraint is: In the formula, F t up F t dn These represent the upward and downward flexibility margins of the system during time period t, respectively.

Images