Optimized dispatching method and device for wind power and cascade hydropower combined system
By adopting Monte Carlo tree search and quantile regression long and short-term memory neural network model in the combined system of wind power and cascade hydropower, the problems of system power generation and volatility optimization are solved, and efficient energy utilization and grid stability are achieved. .
Patent Information
- Application Number
- CN202411033762.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-12
AI Technical Summary
The existing intelligent optimization algorithms are difficult to effectively reduce the solution time in the combined system of wind power and cascade hydropower, and are easily trapped in the local optimal solution, which cannot fundamentally solve the problems of system power generation and volatility optimization.
The daily stochastic scheduling model of wind power and cascade hydropower combined system based on Monte Carlo tree search is adopted, combined with the quantile regression long and short-term memory neural network model, generate wind farm output, and solve the optimal scheduling scheme through the Monte Carlo tree search algorithm , maximize system power generation and minimize volatility.
It significantly improves the power generation and energy utilization efficiency of wind power and cascade hydropower combined systems, improves grid stability, and provides more reliable prediction and optimization scheduling support, reducing the volatility of system output.
Smart Images

Figure CN118944198A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric power systems, and in particular to a method and device for optimizing the dispatching of a wind power and cascade hydropower combined system. Background Art
[0002] In the wind power and cascade hydropower combined power generation system, the main role of cascade hydropower is the same as that of pumped storage power stations, which is to coordinate the total output of the wind power and cascade hydropower combined system, thereby reducing the impact of the output fluctuation of the wind power group on the local load safety. The adjustable capacity of cascade hydropower determines the total output of the combined system. The better the adjustment capacity of cascade hydropower, the greater the fluctuation of adjustable wind power. Under the condition of stable wind power output, the better the adjustment capacity of cascade hydropower, the greater the output difference of wind power, making the total output of the wind power and cascade hydropower combined system more stable, and at the same time, it can also reduce the abandoned water and wind of the combined system.
[0003] Although modern intelligent optimization algorithms have effectively solved the problem of large computational complexity and susceptibility to the curse of dimensionality in traditional optimization algorithms, they are unable to fundamentally solve the problem of premature convergence and susceptibility to falling into local optimal solutions. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method and device for optimizing the dispatching of a wind power and cascade hydropower combined system, so as to effectively reduce the solution time while ensuring a high-quality solution.
[0005] In order to solve the above technical problems, the present invention provides a method for optimizing the dispatching of a wind power and cascade hydropower combined system, comprising:
[0006] Step S1, establishing a daily stochastic optimization dispatch model of a wind power and cascade hydropower combined system based on Monte Carlo tree search, which is used to maximize the power generation of the wind power and cascade hydropower combined system within one day and minimize the total power generation fluctuation;
[0007] Step S2, determining the constraint conditions of the daily random optimization scheduling model;
[0008] Step S3, defining the cascade hydropower station with daily regulation capacity as an intelligent agent, the wind power and cascade hydropower combined system as an interactive environment, and discretizing the state space and action space;
[0009] Step S4, using a quantile regression long short-term memory neural network model to randomly generate wind farm output;
[0010] Step S5, applying the Monte Carlo tree search algorithm to solve the daily random optimization scheduling model to obtain the optimal scheduling solution.
[0011] Preferably, the step S1 specifically includes:
[0012] The calculation cycle is to maximize the power generation of the wind power and cascade hydropower combined system within one day;
[0013] Minimize the total fluctuation of wind power and cascade hydropower combined system power generation within one day of the calculation cycle;
[0014] Establish a daily stochastic optimization operation model for the combined wind power and cascade hydropower system.
[0015] Preferably, the method for maximizing the power generation of the wind power and cascade hydropower combined system within a calculation cycle of one day is:
[0016]
[0017] Where i is the number of the wind power station and cascade hydropower station; j = w,h, w represents the wind power station, h represents the cascade hydropower station; n j is the total number of wind power stations and cascade hydropower stations; t is the time period variable, which divides each day into 24 time periods, T is the number of time periods, indicating the total number of time periods in the cycle; E is the total power generation of the combined system of wind power stations and cascade hydropower stations; P j,i (t) is the power generation of the i-th wind power station or hydropower station of the j-th category in the t-th period; Q h,i (t) is the power generation flow of the ith hydropower station in the tth period; η h,i is the power generation efficiency of the i-th hydropower station, H h,i (t) is the water head of the ith hydropower station in the tth period;
[0018] The calculation method for minimizing the output fluctuation of the wind power and cascade hydropower combined system by measuring the standard deviation is:
[0019]
[0020] Where F is the total fluctuation value of wind power and cascade hydropower combined system power generation during the calculation period; It is the average power generation output of the combined wind power and cascade hydropower system during the period.
[0021] Preferably, the balance constraint of system power is:
[0022]
[0023] Where: P j,i (t) is the total power generation in the tth period. The distribution function curve of wind power output forecast in the tth period is discretized into N values, and the power corresponding to each value is P w,n (t)(n∈[1,N]), the corresponding probability is p w,i,n (t);
[0024] The weights for maximizing the combined system power generation and minimizing the total fluctuation of the combined system power generation are added together:
[0025]
[0026] Where: E and λ F are the weights of power generation proximity and total fluctuation proximity, E is power generation proximity, F is total fluctuation proximity, W is the comprehensive proximity index of wind power and cascade hydropower combined system; R(t) is the indicator function of time period t, V i (t) is the reservoir capacity of the i-th hydropower station in the cascade hydropower station during period t.
[0027] Preferably, the step S2 of determining the model constraint conditions specifically includes:
[0028] Water balance constraints between cascade hydropower stations:
[0029]
[0030] Where: is the upstream power station of the i-th hydropower station, the i-1th hydropower station, at t-ΔT i-1 Downflow flow at time; ΔT i-1 is the number of time periods corresponding to the water flow delay from the i-1 reservoir to the i-th hydropower station reservoir; I i (t) is the average inflow from the i-1th hydropower station to the i-th hydropower station at time t;
[0031] Constraints on discharge flow from cascade reservoirs:
[0032]
[0033] Where: is the minimum and maximum discharge flow of the i-th hydropower station in the t-th period;
[0034] Constraints on water storage capacity of cascade reservoirs:
[0035]
[0036] Where: is the minimum and maximum water storage of the ith reservoir at time t. It is usually the maximum water level based on safety considerations of the reservoir, such as flood season, flood control and restrictions; Usually the lowest water level of dead storage capacity;
[0037] The daily reservoir capacity constraints of cascade hydropower stations are:
[0038] V0=V 24
[0039] In the formula: Cascade reservoirs are regulated daily, and the storage capacity of the first and last reservoirs is equal every day;
[0040] Output constraints of cascade hydropower stations:
[0041]
[0042] Where: are the minimum and maximum outputs of the i-th hydropower station of the j-th category in the t-th period, respectively;
[0043] Mutually exclusive constraints on the extraction of power from cascade hydropower stations:
[0044] n p (t)·n d (t) = 0, n p (t)≠n d (t)
[0045] Where: n p (t) indicates whether the hydropower station unit is in power generation state at time period t: if n p (t) is equal to 1, indicating that the unit is generating electricity in period t; if n p (t) is equal to 0, indicating that the unit did not generate electricity during period t; n d (t) indicates whether the hydropower station unit is in the pumping state at time period t: if n d (t) is equal to 1, indicating that the unit is pumping water during period t; if n d (t) is equal to 0, indicating that the unit did not pump water during period t.
[0046] Preferably, in step S3, the set of the storage capacity of the three cascade hydropower stations m, y and c, the random output of the wind farm, the maximum total power generation of the combined system and the minimum total volatility in the daily random optimization scheduling model of the wind power and cascade hydropower combined system is used as the state space:
[0047]
[0048] Among them, V m (t), V y (t), V c (t) are the storage capacities of the three cascade hydropower stations m, y and c in period t, P w (t) is the random output of the wind farm in period t;
[0049] The output / pumping power of the three cascade hydropower stations m, y, and c is used as the action space:
[0050] A(t)={P m (t),P y (t),P c (t)}.
[0051] Preferably, in step S3, discretization processing is performed on the state space and the action space, specifically including:
[0052] The reservoir capacity as a state variable is discretized into interval form from small to large, V i (t) is the reservoir capacity of the ith hydropower station in the cascade hydropower station during period t, ΔV m is the length of each interval after discretization of m hydropower stations:
[0053]
[0054] m hydropower station has (minV m ,ΔV),(ΔV,2ΔV)…(maxV m -ΔV,maxV m ) A total of A discrete intervals;
[0055] ΔV y is the length of each interval after the discretization of y hydropower station:
[0056]
[0057] y hydropower station has (minV y ,ΔV),(ΔV,2ΔV)…(maxV y -ΔV,maxV y ) A total of B discrete intervals;
[0058] ΔV c is the length of each interval after the discretization of the c hydropower station:
[0059]
[0060] c Hydropower station has (minV c ,ΔV),(ΔV,2ΔV)…(maxV c -ΔV,maxV c ) a total of C discrete intervals;
[0061] The action variables of the three hydropower stations m, y, and c of the cascade hydropower are discretized into interval forms, P h,i (t) is the power generation of the ith hydropower station in the tth period;
[0062] ΔP hm is the length of each interval after discretization of m hydropower stations:
[0063]
[0064] The m hydropower station has a total of (0,ΔP), (ΔP, 2ΔP)…(maxP p,m +maxP d,m-ΔP,maxP p,m +maxP d,m ) a total of D discrete intervals;
[0065] ΔP h,y is the length of each interval after the discretization of y hydropower station:
[0066]
[0067] The y hydropower station has a total of (0, ΔP), (ΔP, 2ΔP)…(maxP p,y +maxP d,y -ΔP,maxP p,y +maxP d,y ) a total of E discrete intervals;
[0068] ΔP h,c is the length of each interval after the discretization of the c hydropower station:
[0069]
[0070] The c hydropower station has a total of (0, ΔP), (ΔP, 2ΔP)…(maxP p,c +maxP d,c -ΔP,maxP p,c +maxP d,c ) a total of F discrete intervals;
[0071] maxP p is the maximum power generation capacity of the cascade hydropower station, maxP d They are the maximum pumping power of the pumped storage power station unit, maxP p +maxP d In order to obtain the power range under the maximum power generation and pumping power, the action variables of the three hydropower stations m, y and c of the cascade hydropower are discretized into D, E and F intervals.
[0072] Preferably, the step S4 specifically includes:
[0073] The quantile regression long short-term memory neural network model is used to predict the output of wind farms. The model includes an LSTM unit consisting of an input gate, a forget gate, and an output gate, and a quantile regression mechanism optimized by an asymmetric loss function based on quantile τ.
[0074] By training the model and using the data from the first five months of the wind farm's history, the wind power output in each period is predicted, and the conditional probability distribution of wind power output in 24 periods a day is derived, thereby obtaining prediction intervals at different confidence levels to achieve wind farm output prediction that takes into account output randomness and the reliability of random results.
[0075] Preferably, the step S5 specifically includes:
[0076] Step S51, set the simulation point number k, the cascade reservoir is daily regulation, and determine that the storage capacity of the initial and final reservoirs is equal V0 = V 24 ;
[0077] Step S52, start the nth search, and randomly select the pumping / power generation action A available in each power station;
[0078] Step S53, searching whether the current node is a child node, if not, selecting a node through UCB, if yes, selecting the current node;
[0079] Step S54, determine whether the value of the current node is 0, if so, simulate the whole day process, otherwise expand according to the current node and select the expanded node for simulation;
[0080] Step S55, recording the pumping / generating action A of each power station and the comprehensive proximity index W of the wind power and cascade hydropower combined system;
[0081] Step S56, updating the node;
[0082] Step S57, determine whether the specified number of operations has been reached, if not, return to step S52 to continue to complete the above steps.
[0083] The present invention also provides a wind power and cascade hydropower combined system optimization dispatching device, comprising:
[0084] A model building module is used to establish a daily stochastic optimization dispatch model of a wind power and cascade hydropower combined system based on Monte Carlo tree search to maximize the power generation of the wind power and cascade hydropower combined system within a day and minimize the total power generation fluctuation;
[0085] A constraint condition determination module, used to determine the constraint conditions of the daily random optimization scheduling model;
[0086] The discretization processing module is used to define the cascade hydropower station with daily regulation capacity as an intelligent agent, and to discretize the state space and action space by taking the wind power and cascade hydropower combined system as an interactive environment;
[0087] Wind power output generation module, used to randomly generate wind farm output using quantile regression long short-term memory neural network model;
[0088] The optimization solution module is used to apply the Monte Carlo tree search algorithm to solve the daily random optimization scheduling model to obtain the optimal scheduling solution.
[0089] The implementation of the present invention has the following beneficial effects: the present invention establishes a multi-objective optimization scheduling model for the wind power and cascade hydropower combined system by comprehensively considering the uncertainty of wind power prediction and the water inflow data of small hydropower groups, achieving the dual goals of maximizing the power generation of the combined system and minimizing the total fluctuation, and significantly improving the energy utilization efficiency and grid stability. The present invention not only improves the prediction accuracy by adding a quantile regression mechanism on the basis of LSTM, but also provides fluctuation information of the predicted value, effectively analyzes the uncertainty of wind power output, and provides more reliable data support for the optimal scheduling of the combined system. By comparing algorithms such as Monte Carlo and Monte Carlo tree search, the present invention verifies the superiority and efficiency of Monte Carlo tree search in dealing with multi-state variable and multi-action variable problems, further improving the calculation accuracy and speed of the optimal scheduling of the combined system, which is of great significance for promoting the reliable grid connection and consumption of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS
[0090] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0091] Figure 1 It is a flow chart of a method for optimizing the dispatching of a wind power and cascade hydropower combined system according to an embodiment of the present invention.
[0092] Figure 2 It is a structural diagram of a quantile regression long short-term memory neural network QR-LSTM in an embodiment of the present invention.
[0093] Figure 3 It is a flow chart of the daily random optimization problem of the wind power and cascade hydropower combined system using the Monte Carlo tree search algorithm in an embodiment of the present invention.
[0094] Figure 4 It is a power generation structure diagram of a wind power and cascade hydropower combined system in an embodiment of the present invention.
[0095] Figure 5 It is a schematic diagram of a wind power and cascade hydropower complementary model in an embodiment of the present invention.
[0096] Figure 6 It is a flow chart of the optimization model of the wind power and cascade hydropower combined system in the embodiment of the present invention.
[0097] Figure 7 It is a diagram of the water level variation process of the cascade hydropower with daily regulation capability in an embodiment of the present invention. DETAILED DESCRIPTION
[0098] The following descriptions of the embodiments refer to the accompanying drawings to illustrate specific embodiments in which the present invention may be implemented.
[0099] Please refer to Figure 1 As shown, an embodiment of the present invention provides a method for optimizing the dispatching of a wind power and cascade hydropower combined system, comprising:
[0100] Step S1, establishing a daily stochastic optimization dispatch model of a wind power and cascade hydropower combined system based on Monte Carlo tree search, which is used to maximize the power generation of the wind power and cascade hydropower combined system within one day and minimize the total power generation fluctuation;
[0101] Step S2, determining the constraint conditions of the daily random optimization scheduling model;
[0102] Step S3, defining the cascade hydropower station with daily regulation capacity as an intelligent agent, the wind power and cascade hydropower combined system as an interactive environment, and discretizing the state space and action space;
[0103] Step S4, using a quantile regression long short-term memory neural network model to randomly generate wind farm output;
[0104] Step S5, applying the Monte Carlo tree search algorithm to solve the daily random optimization scheduling model to obtain the optimal scheduling solution.
[0105] Through the above steps, it can be seen that the embodiments of the present invention can significantly increase the power generation of the wind power and cascade hydropower combined system, while effectively reducing the volatility of the system output, making the total output of the combined system more stable, thereby optimizing energy utilization efficiency, improving grid stability, and promoting the reliable grid connection and consumption of renewable energy.
[0106] Specifically, step S1 of the embodiment of the present invention specifically includes:
[0107] (1) The power generation of the wind power and cascade hydropower combined system is maximized within one day of the calculation cycle.
[0108]
[0109] Where i is the number of the wind power station and cascade hydropower station; j = w,h, w represents the wind power station, h represents the cascade hydropower station; n j is the total number of wind power stations and cascade hydropower stations; t is the time period variable, which divides each day into 24 time periods, T is the number of time periods, indicating the total number of time periods in the cycle; E is the total power generation of the combined system of wind power stations and cascade hydropower stations; P j,i (t) is the power generation of the i-th wind power station or hydropower station of the j-th category in the t-th period; Q h,i (t) is the power generation flow of the ith hydropower station in the tth period; η h,iis the power generation efficiency of the i-th hydropower station, H h,i (t) is the water head of the ith hydropower station in the tth period.
[0110] (2) Minimize the total fluctuation of power generation of the wind power and cascade hydropower combined system within one day of the calculation cycle.
[0111] Commonly used volatility indicators are shown in Table 1 below:
[0112] Table 1 Commonly used volatility indicators
[0113]
[0114] The standard deviation in Table 1 is a commonly used volatility index for the combined system. The standard deviation in the present invention is calculated by taking the 24 time periods of a day as a whole. The calculation formula for minimizing the output fluctuation of the combined system of wind power and cascade hydropower by standard deviation is as follows:
[0115]
[0116] Where F is the total fluctuation value of wind power and cascade hydropower combined system power generation during the calculation period; It is the average power generation output of the combined wind power and cascade hydropower system during the period.
[0117] (3) Establish a daily random optimization operation model for the combined wind power and cascade hydropower system.
[0118] The balance constraint of system power is:
[0119]
[0120] Where: P j,i (t) is the total power generation in the tth period. Here, the distribution function curve of wind power output forecast in the tth period is discretized into N values, and the power corresponding to each value is P w,n (t)(n∈[1,N]), the corresponding probability is p w,i,n (t).
[0121] The weights for maximizing the combined system power generation and minimizing the total fluctuation of the combined system power generation are added together:
[0122]
[0123] Where: E and λ F are the weights of power generation proximity and total fluctuation proximity, E is power generation proximity, F is total fluctuation proximity, W is the comprehensive proximity index of wind power and cascade hydropower combined system; R(t) is the indicator function of time period t, V i (t) is the reservoir capacity of the i-th hydropower station in the cascade hydropower station during period t.
[0124] Step S2 determines the model constraints mainly including:
[0125] (1) Water balance constraints among cascade hydropower stations:
[0126]
[0127] Where: is the upstream power station of the i-th hydropower station, the i-1th hydropower station, at t-ΔT i-1 Downflow flow at time; ΔT i-1 is the number of time periods corresponding to the water flow delay from the i-1th reservoir to the i-th reservoir; I i (t) is the average inflow from the i-1th hydropower station to the i-th hydropower station at time t.
[0128] (2) Constraints on discharge flow from cascade reservoirs:
[0129]
[0130] Where: is the minimum and maximum discharge flow of the i-th hydropower station in the t-th period.
[0131] (3) Constraints on water storage capacity of cascade reservoirs:
[0132]
[0133] Where: is the minimum and maximum water storage capacity of the reservoir of the i-th hydropower station at the t-th time. It is usually the maximum water level based on safety considerations of the reservoir, such as flood season, flood control and restrictions; Usually the lowest water level of dead storage capacity.
[0134] (4) Daily reservoir capacity constraints for cascade hydropower stations:
[0135] V0=V 24
[0136] Wherein: The reservoirs of cascade hydropower stations are regulated daily, and the storage capacity of the reservoirs at the beginning and end of each day is the same.
[0137] (5) Output constraints of cascade hydropower stations:
[0138]
[0139] In the formula, are the minimum and maximum outputs of the i-th hydropower station of the j-th category in the t-th period, respectively.
[0140] (6) Mutually exclusive constraints on the extraction of power from cascade hydropower stations:
[0141] n p (t)·n d (t) = 0, n p (t)≠n d (t)
[0142] The above formula indicates that the unit cannot be in the state of generating electricity and pumping water at the same time. p (t) indicates whether the hydropower station unit is in power generation state at time period t: if n p (t) is equal to 1, indicating that the unit is generating electricity in period t; if n p (t) is equal to 0, indicating that the unit did not generate electricity during period t; n d (t) indicates whether the hydropower station unit is in the pumping state at time period t: if n d (t) is equal to 1, indicating that the unit is pumping water during period t; if n d (t) is equal to 0, indicating that the unit did not pump water during period t.
[0143] Step S3 is to define and discretize the decision environment.
[0144] First, a decision-making environment for the daily random optimization dispatch of the wind power and cascade hydropower combined system is established. It is necessary to define the intelligent agent and the interactive environment. The cascade hydropower can provide decision variables. The cascade hydropower station with the ability of daily regulation is defined as the intelligent agent. The interactive environment is the power system of the wind power and cascade hydropower combined system. The variables in the decision-making environment are defined as follows:
[0145] (1) State space
[0146] The set of the storage capacity of the three cascade hydropower stations m, y, and c, the random output of the wind farm, the maximum total power generation of the combined system, and the minimum total volatility in the daily random optimization scheduling model of the wind power and cascade hydropower combined system is used as the state space, that is:
[0147]
[0148] Among them, V m (t), V y (t), V c (t) are the storage capacities of the three cascade hydropower stations m, y and c in period t, P w (t) is the random output of the wind farm in period t.
[0149] (2) Action Space
[0150] The output / pumping power of the three cascade hydropower stations m, y, and c is used as the action space, that is:
[0151] A(t)={P m (t),P y (t),Pc (t)}
[0152] The output / pumping power of the three cascade hydropower stations m, y, and c in the wind power and cascade hydropower combined power generation system is the action variable, and the storage capacity of the three cascade hydropower stations m, y, and c is the state variable. They are all continuous variables. In the reinforcement learning algorithm, it is necessary to convert the action variables and state variables into discrete variables. The state variable and the action variable are closely related. The state variable can reflect the target well, and it can also reflect the size of the action variable, satisfying the "no aftereffect" and reflecting the entire recursive process. In the daily random optimization scheduling problem of the wind power and cascade hydropower combined system, the storage capacity of the three reservoirs is used as the state variable, and the reservoir capacity is discretized into interval form from small to large, V i (t) is the reservoir capacity of the i-th hydropower station in the cascade hydropower station during period t.
[0153] ΔV m is the length of each interval after discretization of m hydropower stations:
[0154]
[0155] m hydropower station has (minV m ,ΔV),(ΔV,2ΔV)…(maxV m -ΔV,maxV m ) A total of A discrete intervals.
[0156] ΔV y is the length of each interval after the discretization of y hydropower station:
[0157]
[0158] y hydropower station has (minV y ,ΔV),(ΔV,2ΔV)…(maxV y -ΔV,maxV y ) There are B discrete intervals in total.
[0159] ΔV c is the length of each interval after the discretization of the c hydropower station:
[0160]
[0161] c Hydropower station has (minV c ,ΔV),(ΔV,2ΔV)…(maxV c -ΔV,maxV c ) There are C discrete intervals in total.
[0162] The action variables include the pumping and power generation actions of the three hydropower stations m, y, and c of the cascade hydropower. The current state S can be determined by the predicted random output of wind power in period t and the water levels of the three hydropower stations m, y, and c. t ,By calculating the maximum total power generation and the minimum total volatility of the ,combined system, it is possible to determine whether the three power station units in the ,wind power and cascade hydropower combined system are in the pumping state or ,the power generation state.
[0163] The action variables of the three hydropower stations m, y, and c of the cascade hydropower are discretized into interval forms, P h,i (t) is the power generation capacity of the i-th hydropower station in the t-th period.
[0164] ΔP h,m is the length of each interval after discretization of m hydropower stations:
[0165]
[0166] The m hydropower station has a total of (0,ΔP), (ΔP, 2ΔP)…(maxP p,m +maxP d,m -ΔP,maxP p,m +maxP d,m ) There are D discrete intervals in total.
[0167] ΔP h,y is the length of each interval after the discretization of y hydropower station:
[0168]
[0169] The y hydropower station has a total of (0, ΔP), (ΔP, 2ΔP)…(maxP p,y +maxP d,y -ΔP,maxP p,y +maxP d,y ) There are E discrete intervals in total.
[0170] ΔP h,c is the length of each interval after the discretization of the c hydropower station:
[0171]
[0172] The c hydropower station has a total of (0, ΔP), (ΔP, 2ΔP)…(maxP p,c +maxP d,c -ΔP,maxP p,c +maxP d,c ) There are F discrete intervals in total.
[0173] maxP p is the maximum power generation capacity of the cascade hydropower station, maxP dThey are the maximum pumping power of the pumped storage power station unit, maxP p +maxP d is the power range under maximum power generation and pumping power, i.e. (0, ΔP), (ΔP, 2ΔP)…(maxP p +maxP d -ΔP,maxP p +maxP d ), the action variables of the three hydropower stations m, y, and c of the cascade hydropower should also be discretized into D, E, and F intervals.
[0174] Step S4 uses a quantile regression long short-term memory neural network model to randomly generate wind farm output.
[0175] Considering the volatility and randomness of wind power output, the wind power output prediction at the current moment is significantly affected by the output at different time steps. Therefore, the embodiment of the present invention adopts a quantile regression long short-term memory neural network (QuantileRegression Long Short-Term Memory, QR-LSTM) for prediction modeling, such as Figure 2 As shown in the figure, the LSTM unit consists of an input gate, a forget gate, and an output gate. The gated unit is used to improve the memory ability of long-term step-dependent information and forget unnecessary information, thereby achieving effective information memory and redundant information forgetting.
[0176] Input gate: i t =σ(W i [h t-1 ,x t ]+b i ) t =tanh(W g [h t-1 ,x t ]+b g )c t =c t-1 ⊙f t +i t ⊙g t
[0177] Forget gate: f t =σ(W f [h t-1 ,x t ]+b f )
[0178] Output gate: o t =σ(W o [h t-1 ,x t ]+b o ) t =ot ⊙tanh(c t )
[0179] Where: f t 、i t , g t , o t 、c t are the characteristic vectors of each gate at time t; W f , W i , W g , W o are the weight matrices of each gate; b f , b i , b g , b o are the bias terms of each gate respectively; x t is the input variable at time t; h t-1 and h t are the output variables at time t-1 and time t respectively; σ and tanh are activation functions; ⊙ is the Hadamard product.
[0180] Conventional wind power output point prediction methods pursue the accuracy of a single prediction point and often ignore the uncertainty and randomness of the wind power output prediction value. For this reason, the present invention adds a quantile regression mechanism based on LSTM, which can provide fluctuation information of the prediction value to analyze the uncertainty of wind power output. Quantile regression regresses the input feature variable X according to the conditional quantile of the explained variable Y, effectively deriving the conditional probability distribution of the explained variable Y, thereby obtaining the interval prediction result. QRTCN converts the loss function into the following formula to obtain the conditional quantile at the final τth quantile point:
[0181]
[0182] Where: N is the number of data samples; represents an asymmetric loss function based on quantile τ; y i , Represents the true value and predicted value of the i-th data.
[0183] The power output of a wind farm is determined based on the installed capacity of wind power in a certain area and the wind power output coefficient. The key to the random generation of wind farm output is to determine the prediction interval of wind farm output in 24 time periods of a day. In statistics, sample confidence intervals are generally used to estimate the interval of population parameters. The upper and lower confidence limits of the statistic are the intervals composed of the upper and lower bounds, which are confidence intervals. They indicate the probability that the statistical parameter falls within the range of the output result, which here indicates the credibility of the results of the measured parameters.
[0184] The present invention regards the power generation output stage of the wind farm, and the average output of each time period in the 24 time periods of a day as the total of a stage. The average power generation output of the wind farm in the 7 pm time period today is unrelated to the 7 pm time period of any day in the past or future, and the average output of other time periods in the 24 time periods of a day. According to the central limit theorem, a large number of random variables generally obey the conditions of normal distribution. Through the data of the first five months in the history of the wind farm, the wind power generation output of 24 time periods in a day can be obtained. Through the data of the first five months in the history of the wind farm, statistical analysis is performed to obtain the 95%, 90%, and 85% confidence intervals of the overall mean of each time period in the 24 time periods in a day, indicating that the wind power generation output in the tth time period in a day has a probability of 95%, 90%, and 85% falling into the corresponding confidence interval. Therefore, the predicted value randomly generated by the confidence interval of the power generation output stage of the wind farm of the present invention takes into account the randomness of the output and the reliability of the random results.
[0185] Step S5 applies the Monte Carlo tree search algorithm to the daily stochastic optimization problem of the wind power and cascade hydropower combined system, such as Figure 3 As shown, the specific steps are as follows:
[0186] Step S51: Set the simulation point number k, the cascade reservoir is daily regulated, and determine that the storage capacity of the initial and final reservoirs is equal V0 = V 24 .
[0187] Step S52: Start the nth search and randomly select the pumping / power generation action A among the available actions of each power station.
[0188] Step S53: Search whether the current node is a child node. If not, select a node through UCB. If yes, select the current node.
[0189] Step S54: Determine whether the value of the current node is 0. If so, simulate the whole day process. Otherwise, expand according to the current node and select the expanded node for simulation.
[0190] Step S55: Record the pumping / generating action A of each power station and the comprehensive proximity index W of the wind power and cascade hydropower combined system.
[0191] Step S56: Update the node.
[0192] Step S57: Determine whether the specified number of operations has been reached. If not, return to step S52 to continue the above steps.
[0193] The wind power and cascade hydropower combined power generation system studied in the present invention includes four cascade hydropower stations and a wind power group. The cascade hydropower takes the cascade power station group in the Xiaojinchuan River Basin in Sichuan Province, China as an example. The total installed capacity of the four cascade hydropower stations is 159MW. The wind farm in a certain area is composed of 80 wind turbines with an installed capacity of 1.5MW, and the total installed capacity is 120MW. The power generation of the wind power and cascade hydropower combined system directly supplies the local load. The scheduling cycle is in days, and the scheduling period is in hours. A case study of cascade hydropower and wind power is conducted based on the water inflow data of a typical daily small hydropower station group. As a non-dispatched power plant, the wind farm only superimposes the power generation output of the wind farm in the calculation process of the model, and the output is fully absorbed. The present invention adjusts the operation process of the cascade hydropower to enable the combined system to achieve a safer and more reliable operation state.
[0194] The model of the present invention takes the complementary system of wind power and cascade hydropower as the research object, and the overall schematic diagram is as follows: Figure 5 As shown in the figure, the energy flow represents the energy flow of the wind power and cascade hydropower combined system, and the information flow represents the information transmission within the combined system and the communication with the upper-level power grid. That is, the wind farm is fully fed into the power grid according to the predicted random power, and the cascade hydropower is used as a battery to provide corresponding power support through the wind power output transmitted in the control center.
[0195] like Figure 6 As shown, the present invention establishes a mathematical model of a wind farm and a cascade hydropower station, and comprehensively considers the uncertainty of wind power prediction and the water inflow data of small hydropower groups, and establishes a multi-objective optimization model for the day-ahead dispatch of wind power and cascade hydropower with the goal of maximizing the power generation of the combined system and minimizing the total fluctuation. On this basis, considering the regulation characteristics of cascade hydropower, such as water balance constraints, water storage constraints, downstream flow constraints and power station output constraints, the decision variables are the predicted output of wind power and the power generation flow of hydropower stations at all levels. Considering the volatility and randomness of wind power output, the wind power output prediction at the current moment is significantly affected by the output of different time steps, so QR-LSTM is used for prediction modeling. The LSTM unit is composed of an input gate, a forget gate, and an output gate. The gate control unit is used to improve the memory ability of long-term step-dependent information, forget unnecessary information, thereby realizing effective information memory and redundant information forgetting. Then, the overall parameter interval of wind power output is estimated through the sample confidence interval. The upper and lower confidence limits of the statistic are the upper and lower limits, and the interval formed is the confidence interval, which indicates the probability that the statistical parameter falls within the range of the output result, and here indicates the credibility of the result of the measured parameter. Then the target output value is discretized, and the solution is compared through random dynamic programming, Monte Carlo and Monte Carlo tree search.
[0196] Cascade reservoirs have their own regulation capacity, and the reservoir capacity coefficient β is generally used to represent the reservoir's regulation capacity. The reservoir capacity coefficient β is the ratio of the regulation storage capacity to the average runoff water volume of the reservoir over many years. The reservoir regulation capacity and the reservoir capacity coefficient β are positively correlated. The larger the reservoir capacity coefficient β, the stronger the reservoir regulation capacity. The corresponding relationship between the two is shown in Table 2.
[0197] Table 2 Corresponding relationship between storage capacity coefficient and regulation capacity
[0198]
[0199] The following is a comparative analysis of the optimization results.
[0200] Taking the wind power generation with an installed capacity of 120MW as an example, the combined power generation system after the complementarity of wind power and cascade hydropower is analyzed, and the uncertainty of wind power prediction and the water inflow data of small hydropower groups are comprehensively considered. On this basis, the regulation characteristics of cascade hydropower are taken into account, such as water balance constraints, water storage constraints, downstream flow constraints and power station output constraints. The decision variables are the predicted random output of wind power and the power generation flow of hydropower stations at all levels. It is concluded that the total power generation of the combined power generation system on this typical day is 5137.11MW·h, of which the total power generation of wind turbines is 1827.54MW·h, accounting for 35.58% of the total power generation of the combined system. Mengguqiao is a run-of-river hydropower station and does not have the ability to adjust on a daily basis. Mupo and Yangjiawan on Fubian River and Chunchangba Power Station on Wozhi River have been put into operation and have the ability to adjust on a daily basis. The power generation of Chunchangba Hydropower Station (i.e. the aforementioned C Hydropower Station) is 1037.27MW·h, accounting for 20.19% of the total power generation of the combined system; the power generation of Mupo Hydropower Station (i.e. the aforementioned M Hydropower Station) is 764.77MW·h, accounting for 14.89% of the total power generation of the combined system; the power generation of Yangjiawan Hydropower Station (i.e. the aforementioned Y Hydropower Station) is 909.29MW·h, accounting for 17.70% of the total power generation of the combined system; the power generation of Mengguqiao Hydropower Station is 598.24MW·h, accounting for 11.65% of the total power generation of the combined system. Among them, the wind power generation accounts for the highest proportion, and the run-of-river Mengguqiao Hydropower Station accounts for the lowest proportion.
[0201] Through the output structure of wind power and cascade hydropower stations in 24 hours a day after the wind power and cascade hydropower combined system complement each other, it can be seen that the proportion of wind power generation in the total power generation after the combined system complementation is relatively high in the 13-20 period, but the proportion of total power generation in the 14 and 19 periods becomes lower. In general, the proportion of cascade hydropower station power generation is higher in the 0-11 period and 22-24 period when the proportion of wind power generation is low, and the cascade hydropower and wind power generation are complementary; the power generation output of the three hydropower stations of Mupo, Yangjiawan and Mengguqiao in the hydropower station group is the largest in the 3-4 period, and the total proportion is about 85%. Since the Chunchangba Hydropower Station on the Wozhi River and the Mupo Hydropower Station on the Fubian River are not in the same basin, the output does not have similar characteristics. At the same time, the Mengguqiao Hydropower Station in the Fubian River Basin is a run-of-river hydropower station and does not have the ability to adjust daily. Therefore, its output ratio curve is very similar to that of the Yangjiawan Power Station in the same basin.
[0202] Then, taking wind power generation with an installed capacity of 120MW as an example, the water level process of cascade hydropower stations is analyzed. Figure 7 The water level change process after optimization is solved for the three hydropower stations of Chunchangba, Mupo and Yangjiawan with daily regulation capacity. From the rising and falling trends of the water levels of each hydropower station, Mupo and Yangjiawan hydropower stations of Fubian River are in the same basin, so they have hydraulic connection. Mupo and Yangjiawan hydropower stations have a high degree of similarity in the process of water storage and drawdown within a day. The water level change characteristics of Chunchangba, Mupo and Yangjiawan hydropower stations are as follows:
[0203] First, during the period of 14:00-20:00, although the Chunchangba, Mupo and Yangjiawan hydropower stations are located in different river basins, the water level change processes are extremely similar; then, during the period of 22:00-24:00, the Mupo and Yangjiawan hydropower stations began to store water and return to the initial water level, and the Chunchangba reservoir was located in a different river basin and began to release water to the initial water level. Because it has no hydraulic connection with the Mupo and Yangjiawan cascade hydropower stations, the water level processes of the hydropower stations in the two river basins at the end of the day are slightly different. In general, the water levels of the three hydropower stations of Chunchangba, Mupo and Yangjiawan are within the upper and lower limits of the water level threshold, and the water levels of the three hydropower stations have not reached the highest water level, indicating that no water abandonment is generated in the power generation process of the combined system of wind power and cascade hydropower, and the utilization rate of water energy reaches 100%; at the same time, the water levels of the three hydropower stations of Chunchangba, Mupo and Yangjiawan are guaranteed to be equal at the beginning and end of the day within the constraint conditions, and the three water levels are not approached in the entire operation optimization process. The upper and lower limits of the threshold, indicating that the overall regulation capacity of the three hydropower stations of Chunchangba, Mupo and Yangjiawan with daily regulation capacity is still sufficient for regulating 120MW of wind power generation. When wind power and cascade hydropower are in joint operation, on the basis of fully absorbing wind power, water resources are fully utilized, and the power generation of the combined system is maximized and the total fluctuation of the combined system power generation is minimized, which verifies the correctness and scientificity of the model and algorithm of the present invention.
[0204] The Monte Carlo and Monte Carlo tree search algorithms of the present invention are applied to the daily random optimization scheduling of the wind power and cascade hydropower combined system in a certain area. The comparison table of the optimal solution results of MC and MCTS is shown in Table 3 below.
[0205] Table 3 Comparison of optimization results of MC and MCTS
[0206]
[0207] The present invention establishes a decision-making environment for daily random optimization scheduling of a wind power and cascade hydropower combined system. Cascade hydropower can provide decision variables. Menggu Bridge is a run-of-river hydropower station and does not have the ability to adjust daily. Mupo, Yangjiawan on the Fubian River, and Chunchangba on the Wozhi River have been put into operation and have the ability to adjust daily. Therefore, the three power stations of Mupo, Yangjiawan and Chunchangba in the cascade hydropower are defined as intelligent entities, and the interactive environment is the power system of the wind power and cascade hydropower combined system. As can be seen from Table 3, the present invention has more state variables and action variables. The MCTS generates 57.54MW·h more power than the MC, the output fluctuation is reduced by 0.08MW, and the time is reduced by 38.73s. Therefore, MCTS is superior to the random simulation of MC in terms of accuracy and speed.
[0208] Corresponding to the above-mentioned method for optimizing the dispatching of a wind power and cascade hydropower combined system in the first embodiment of the present invention, the second embodiment of the present invention further provides a device for optimizing the dispatching of a wind power and cascade hydropower combined system, including:
[0209] A model building module is used to establish a daily stochastic optimization dispatch model of a wind power and cascade hydropower combined system based on Monte Carlo tree search to maximize the power generation of the wind power and cascade hydropower combined system within a day and minimize the total power generation fluctuation;
[0210] A constraint condition determination module, used to determine the constraint conditions of the daily random optimization scheduling model;
[0211] The discretization processing module is used to define the cascade hydropower station with daily regulation capacity as an intelligent agent, and to discretize the state space and action space by taking the wind power and cascade hydropower combined system as an interactive environment;
[0212] Wind power output generation module, used to randomly generate wind farm output using quantile regression long short-term memory neural network model;
[0213] The optimization solution module is used to apply the Monte Carlo tree search algorithm to solve the daily random optimization scheduling model to obtain the optimal scheduling solution.
[0214] For the working principle and process of this embodiment, please refer to the description of the aforementioned embodiment 1 of the present invention, which will not be repeated here.
[0215] Compared with the prior art, the beneficial effect brought by the embodiment of the present invention is that the present invention establishes a multi-objective optimization scheduling model for the wind power and cascade hydropower combined system by comprehensively considering the uncertainty of wind power prediction and the water inflow data of small hydropower groups, achieving the dual goals of maximizing the power generation of the combined system and minimizing the total fluctuation, and significantly improving the energy utilization efficiency and grid stability. By adding a quantile regression mechanism on the basis of LSTM, the present invention not only improves the prediction accuracy, but also provides the fluctuation information of the predicted value, effectively analyzes the uncertainty of wind power output, and provides more reliable data support for the optimal scheduling of the combined system. By comparing algorithms such as Monte Carlo and Monte Carlo tree search, the present invention verifies the superiority and efficiency of Monte Carlo tree search in dealing with multi-state variable and multi-action variable problems, further improves the calculation accuracy and speed of the optimal scheduling of the combined system, and is of great significance for promoting the reliable grid connection and consumption of renewable energy.
[0216] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.
Claims
1. A method for optimizing the dispatching of a wind power and cascade hydropower combined system, characterized in that: include: Step S1, establishing a daily stochastic optimization dispatch model of a wind power and cascade hydropower combined system based on Monte Carlo tree search, which is used to maximize the power generation of the wind power and cascade hydropower combined system within one day and minimize the total power generation fluctuation; Step S2, determining the constraint conditions of the daily random optimization scheduling model; Step S3, defining the cascade hydropower station with daily regulation capacity as an intelligent agent, the wind power and cascade hydropower combined system as an interactive environment, and discretizing the state space and action space; Step S4, using a quantile regression long short-term memory neural network model to randomly generate wind farm output; Step S5, applying the Monte Carlo tree search algorithm to solve the daily random optimization scheduling model to obtain the optimal scheduling solution.
2. The method according to claim 1, characterized in that The step S1 specifically includes: The calculation cycle is to maximize the power generation of the wind power and cascade hydropower combined system within one day; Minimize the total fluctuation of wind power and cascade hydropower combined system power generation within one day of the calculation cycle; Establish a daily stochastic optimization operation model for the combined wind power and cascade hydropower system.
3. The method according to claim 2, characterized in that The way to maximize the power generation of the wind power and cascade hydropower combined system within a calculation cycle of one day is: Where i is the number of the wind power station and cascade hydropower station; j = w,h, w represents the wind power station, h represents the cascade hydropower station; n j is the total number of wind power stations and cascade hydropower stations; t is the time period variable, which divides each day into 24 time periods, T is the number of time periods, indicating the total number of time periods in the cycle; E is the total power generation of the combined system of wind power stations and cascade hydropower stations; P j,i (t) is the power generation of the i-th wind power station or hydropower station of the j-th category in the t-th period; Q h,i (t) is the power generation flow of the ith hydropower station in the tth period; η h,i is the power generation efficiency of the i-th hydropower station, H h,i (t) is the water head of the ith hydropower station in the tth period; The calculation method for minimizing the output fluctuation of the wind power and cascade hydropower combined system by measuring the standard deviation is: Where F is the total fluctuation value of wind power and cascade hydropower combined system power generation during the calculation period; It is the average power generation output of the combined wind power and cascade hydropower system during the period.
4. The method according to claim 3, characterized in that: The balance constraint of system power is: Where: P j,i (t) is the total power generation in the tth period. The distribution function curve of wind power output forecast in the tth period is discretized into N values, and the power corresponding to each value is P w,n (t)(n∈[1,N]), the corresponding probability is p w,i,n (t); The weights for maximizing the combined system power generation and minimizing the total fluctuation of the combined system power generation are added together: Where: E and λ F are the weights of power generation proximity and total fluctuation proximity, E is power generation proximity, F is total fluctuation proximity, W is the comprehensive proximity index of wind power and cascade hydropower combined system; R(t) is the indicator function of time period t, V i (t) is the reservoir capacity of the i-th hydropower station in the cascade hydropower station during period t.
5. The method according to claim 4, characterized in that The step S2 of determining the model constraint conditions specifically includes: Water balance constraints between cascade hydropower stations: Where: is the upstream power station of the i-th hydropower station, the i-1th hydropower station, at t-ΔT i-1 Downflow flow at time; ΔT i-1 is the number of time periods corresponding to the water flow delay from the i-1 reservoir to the i-th hydropower station reservoir; I i (t) is the average inflow from the i-1th hydropower station to the i-th hydropower station at time t; Constraints on discharge flow from cascade reservoirs: Where: is the minimum and maximum discharge flow of the i-th hydropower station in the t-th period; Constraints on water storage capacity of cascade reservoirs: Where: is the minimum and maximum water storage of the ith reservoir at time t. It is usually the maximum water level based on safety considerations of the reservoir, such as flood season, flood control and restrictions; Usually the lowest water level of dead storage capacity; The daily reservoir capacity constraints of cascade hydropower stations are: V0=V 24 In the formula: Cascade reservoirs are regulated daily, and the storage capacity of the first and last reservoirs is equal every day; Output constraints of cascade hydropower stations: Where: are the minimum and maximum outputs of the i-th hydropower station of the j-th category in the t-th period, respectively; Mutually exclusive constraints on the extraction of power from cascade hydropower stations: n p (t)·n d (t)=0,n p (t)≠n d (t) Where: n p (t) indicates whether the hydropower station unit is in power generation state at time period t: if n p (t) is equal to 1, indicating that the unit is generating electricity in period t; if n p (t) is equal to 0, indicating that the unit did not generate electricity during period t; n d (t) indicates whether the hydropower station unit is in the pumping state at time period t: if n d (t) is equal to 1, indicating that the unit is pumping water during period t; if n d (t) is equal to 0, indicating that the unit did not pump water during period t.
6. The method according to claim 5, characterized in that In step S3, the set of the storage capacity of the three cascade hydropower stations m, y, and c, the random output of the wind farm, the maximum total power generation of the combined system, and the minimum total volatility in the daily random optimization scheduling model of the wind power and cascade hydropower combined system is used as the state space: Among them, V m (t), V y (t), V c (t) are the storage capacities of the three cascade hydropower stations m, y and c in period t, P w (t) is the random output of the wind farm in period t; The output / pumping power of the three cascade hydropower stations m, y, and c is used as the action space: A(t)={P m (t),P y (t),P c (t)}。 7. The method according to claim 6, characterized in that In step S3, the state space and the action space are discretized, specifically including: The reservoir capacity as a state variable is discretized into interval form from small to large, V i (t) is the reservoir capacity of the ith hydropower station in the cascade hydropower station during period t, ΔV m is the length of each interval after discretization of m hydropower stations: m hydropower station has (minV m ,ΔV),(ΔV,2ΔV)…(maxV m -ΔV,maxV m ) A total of A discrete intervals; ΔV y is the length of each interval after the discretization of y hydropower station: y hydropower station has (minV y ,ΔV),(ΔV,2ΔV)…(maxV y -ΔV,maxV y ) A total of B discrete intervals; ΔV c is the length of each interval after the discretization of the c hydropower station: c Hydropower station has (minV c ,ΔV),(ΔV,2ΔV)…(maxV c -ΔV,maxV c ) a total of C discrete intervals; The action variables of the three hydropower stations m, y, and c of the cascade hydropower are discretized into interval forms, P h,i (t) is the power generation of the ith hydropower station in the tth period; ΔP hm is the length of each interval after discretization of m hydropower stations: The m hydropower station has a total of (0,ΔP), (ΔP, 2ΔP)…(maxP p,m +maxP d,m -ΔP,maxP p,m +maxP d,m ) a total of D discrete intervals; ΔP h,y is the length of each interval after the discretization of y hydropower station: The y hydropower station has a total of (0, ΔP), (ΔP, 2ΔP)…(maxP p,y +maxP d,y -ΔP,maxP p,y +maxP d,y ) a total of E discrete intervals; ΔP hc is the length of each interval after the discretization of the c hydropower station: The c hydropower station has a total of (0, ΔP), (ΔP, 2ΔP)…(maxP p,c +maxP d,c -ΔP,maxP p,c +maxP d,c ) a total of F discrete intervals; maxP p is the maximum power generation capacity of the cascade hydropower station, maxP d They are the maximum pumping power of the pumped storage power station unit, maxP p +maxP d In order to obtain the power range under the maximum power generation and pumping power, the action variables of the three hydropower stations m, y and c of the cascade hydropower are discretized into D, E and F intervals.
8. The method according to claim 1, characterized in that The step S4 specifically includes: The quantile regression long short-term memory neural network model is used to predict the output of wind farms. The model includes an LSTM unit consisting of an input gate, a forget gate, and an output gate, and a quantile regression mechanism optimized by an asymmetric loss function based on quantile τ. By training the model and using the data from the first five months of the wind farm's history, the wind power output in each period is predicted, and the conditional probability distribution of wind power output in 24 periods a day is derived, thereby obtaining prediction intervals at different confidence levels to achieve wind farm output prediction that takes into account output randomness and the reliability of random results.
9. The method according to claim 1, characterized in that The step S5 specifically includes: Step S51, set the simulation point number k, the cascade reservoir is daily regulation, and determine that the storage capacity of the initial and final reservoirs is equal V0 = V 24 ; Step S52, start the nth search, and randomly select the pumping / power generation action A available in each power station; Step S53, searching whether the current node is a child node, if not, selecting a node through UCB, if yes, selecting the current node; Step S54, determine whether the value of the current node is 0, if so, simulate the whole day process, otherwise expand according to the current node and select the expanded node for simulation; Step S55, recording the pumping / generating action A of each power station and the comprehensive proximity index W of the wind power and cascade hydropower combined system; Step S56, updating the node; Step S57, determine whether the specified number of operations has been reached, if not, return to step S52 to continue to complete the above steps.
10. A wind power and cascade hydropower combined system optimization dispatching device, characterized in that: include: A model building module is used to establish a daily stochastic optimization dispatch model of a wind power and cascade hydropower combined system based on Monte Carlo tree search to maximize the power generation of the wind power and cascade hydropower combined system within a day and minimize the total power generation fluctuation; A constraint condition determination module, used to determine the constraint conditions of the daily random optimization scheduling model; The discretization processing module is used to define the cascade hydropower station with daily regulation capacity as an intelligent agent, and to discretize the state space and action space by taking the wind power and cascade hydropower combined system as an interactive environment; Wind power output generation module, used to randomly generate wind farm output using quantile regression long short-term memory neural network model; The optimization solution module is used to apply the Monte Carlo tree search algorithm to solve the daily random optimization scheduling model to obtain the optimal scheduling solution.
