Time period aggregation method for SCUC model with high proportion of wind power based on neural network

Abstract

The present invention provides a time period aggregation method for a SCUC model containing a high proportion of wind power based on a neural network, which relates to the technical field of wind power SCUC models, and aims to achieve the requirements of accelerating the solution of unit combination problems and fully considering the uncertainty of wind power, including performing wind power multi-scenario clustering based on K-means; performing SCUC model time period aggregation based on wind power uncertainty. The present invention has the advantages of fast model solution and high reliability in the face of wind power uncertainty.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power SCUC models, and in particular to a time period aggregation method for a SCUC model containing a high proportion of wind power based on a neural network. Background Art

[0002] The security constrained unit commitment (SCUC) problem is the basis for clearing the day-ahead power market. The start and stop status of the unit is determined by solving the minimization of the target cost. In order to ensure the safety and economy of dispatch, the domestic market uses a 15-minute dispatch time interval for market clearing.

[0003] With the improvement of time resolution, the integer variables in the SCUC problem grow exponentially, resulting in a significant increase in calculation time. In the future, the proportion of new energy in the new power system will increase significantly, and the complexity of unit combination will be further improved; at the same time, the coordinated operation of the provincial power market in the future will also greatly increase the number of units and constraints, further increasing the difficulty of calculation. How to ensure that the complex calculation of unit combination is completed within the limited calculation time of the day-ahead market is of great significance. With the advancement of global energy transformation, the importance of renewable energy in the global power system is becoming increasingly prominent. Wind power and photovoltaic power, as the core, are growing rapidly, but the day-ahead forecast of wind power has great uncertainty. Therefore, for the SCUC problem with a high proportion of wind power, the uncertainty of wind power generation needs to be fully considered. At present, domestic and foreign research usually uses three types of methods to aggregate the time periods of the SCUC problem with a high proportion of new energy. The first type is based on clustering technology, which divides a day into 24 time periods of different durations without changing the number of time periods. Although this method customizes the time interval through clustering technology, it cannot reduce the time variable to accelerate the solution of the unit combination problem. The second type is to adjust the time resolution of the calculation. This method uses a coarser time resolution and simplified model for the last few hours, which easily ignores the volatility of load and power generation and may lead to large errors. A finer time resolution and detailed modeling are used in the initial hours of the scheduling time range, while a coarser time resolution is used in the last few hours. The third type is to adaptively aggregate the time step by proposing a prior mixed integer problem to reduce the original gradient that changes over time. This method considers wind power output as an accurate value and does not consider the day-ahead wind power output as a predicted value. It has a large uncertainty, which makes the solution of the unit combination have a large error.

[0004] Therefore, it is necessary to optimize the SCUC in the wind power field to accelerate the solution of the unit combination problem and fully consider the uncertainty of wind power. Summary of the invention

[0005] The object of the present invention is to provide a neural network-based time period aggregation method for a SCUC model containing a high proportion of wind power, which can achieve the requirements of accelerating the solution of unit combination problems and fully considering the uncertainty of wind power.

[0006] The present invention is achieved through the following technical solutions:

[0007] The time period aggregation method of the SCUC model with high proportion of wind power based on neural network includes the following steps:

[0008] Wind power multi-scenario clustering based on K-means;

[0009] SCUC model time period aggregation based on wind power uncertainty.

[0010] Preferably, the method for performing wind power multi-scenario clustering based on K-means comprises the following steps:

[0011] Step S101: Randomly generate n output scenarios based on the expected and variance of wind power output on the operating day, and set clustering parameters , in size Randomly select from the dataset The data points are used to obtain the cluster center, and the data points are selected based on the Euclidean distance to form the initial cluster. ;

[0012] Step S102: updating the cluster center and selecting data points based on the Euclidean distance to form a new cluster;

[0013] Step S103: Using the square error sum criterion function Determine whether the target clustering condition is met, if not, return to step S102 for iteration, if yes, stop iteration.

[0014] Preferably, the method of selecting data points based on Euclidean distance to form initial clusters is:

[0015] Calculate the data points To The cluster centers Euclidean distance , Representative Iterations;

[0016] like ,but .

[0017] Preferably, the method for updating the cluster center is:

[0018] Calculate the average value of the data points in each of the initial clusters as the updated Iteration cluster centers:

[0019] ;

[0020] in, For the data points in the initial cluster of the cluster centers, is the total number of data points in all the initial clusters;

[0021] The method of selecting data points based on Euclidean distance to form a new cluster is the same as the method of selecting data points based on Euclidean distance to form an initial cluster.

[0022] Preferably, the error square sum criterion function The method to determine whether the target clustering condition is met is:

[0023] Calculate the error square sum criterion function The value of Represents the number of iterations;

[0024] ;

[0025] Preset judgment threshold ;

[0026] If satisfied And if the clustering ratio is between 15% and 30%, the target clustering condition is reached.

[0027] Preferably, the method for performing time period aggregation of the SCUC model based on wind power uncertainty is:

[0028] Optimize the weighted multi-scenario objective function and set constraints;

[0029] Perform neural network-based time period aggregation;

[0030] Select the number of aggregation time periods.

[0031] Preferably, the method of optimizing the weighted multi-scenario objective function and setting constraints is:

[0032] The multi-scenario objective function is:

[0033] ;

[0034] in, is the total number of scenes, is the total number of units, The total number of time periods divided into a day, For the scene Medium Unit In the period The output cost, For the scene Medium Unit In the period startup costs, For the scene The clustering ratio;

[0035] The constraints include:

[0036] Load balancing constraints:

[0037] ;

[0038] in, For the scene Medium Unit In the period of efforts, For the scene The total system load in For the scene Wind power output in

[0039] System positive and standby constraints:

[0040] ;

[0041] in, For the scene Medium Unit In the period The start-stop status, For the scene Medium Unit In the period The maximum output, For the scene Medium Unit of system positive spare capacity;

[0042] System negative standby constraints:

[0043] ;

[0044] in, For the scene Medium Unit In the period The minimum output, For the scene Medium Unit The system negative reserve capacity;

[0045] Output upper and lower limits:

[0046] ;

[0047] Ramp rate constraint:

[0048] ;

[0049] ;

[0050] in, For the crew The maximum climbing rate, For the crew Maximum downhill climbing rate;

[0051] Minimum continuous start and stop time constraints:

[0052] ;

[0053] ;

[0054] in, For the scene Medium Unit In the period Continuous boot time, For the scene Medium Unit In the period Continuous downtime. is the minimum startup time of the unit, The minimum downtime of the unit.

[0055] Preferably, the method for performing time period aggregation based on a neural network is:

[0056] The feedforward neural network is used to autonomously establish the relationship between the total system load change percentage, weighted wind power output change value, and the absolute value of weighted wind power output and the period aggregation score in each period;

[0057] If the unit commitment remains unchanged in adjacent time periods, the aggregation score of the time period is 1, otherwise it is 0.

[0058] Preferably, the method for selecting the number of aggregation time periods is:

[0059] Set the number of aggregation periods , sort the aggregate scores of each time period and select the one with the highest score Aggregate over time periods.

[0060] Preferably, the number of aggregation periods The value range is 5 to 7.

[0061] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0062] The present invention uses the K-means clustering algorithm to obtain the wind power output clustering scenario on the operating day, and uses the clustering ratio of each scenario as a weight to establish a multi-scenario SCUC model, thereby enhancing the safety and economy of the combination with a high proportion of wind turbines.

[0063] The present invention can comprehensively evaluate the thermal power output change, weighted wind power output change and weighted wind power absolute value of each time period through a neural network, and then predict the time period aggregation score, providing a scientific quantitative indicator for time period aggregation;

[0064] The present invention significantly reduces the solution time of the unit combination model by reasonably selecting the number of clusters while ensuring that the solution error of the unit combination model is extremely small;

[0065] The invention has reasonable design, is easy to realize and apply to the field of wind power generation, and is convenient for promotion and implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flow chart of performing wind power multi-scenario clustering based on K-means provided in Example 1 of the present invention;

[0067] Figure 2 This is a flow chart of the time period aggregation of the SCUC model taking into account wind power uncertainty provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0069] Example 1

[0070] This embodiment provides a time period aggregation method for a SCUC model containing a high proportion of wind power based on a neural network, comprising the following steps:

[0071] Wind power multi-scenario clustering based on K-means;

[0072] SCUC model time period aggregation based on wind power uncertainty.

[0073] The technical solution of this embodiment is established for the requirements of the new power system to speed up the solution of the unit combination problem and fully consider the uncertainty of wind power. This embodiment mainly includes wind power multi-scenario clustering based on K-means, and SCUC model period aggregation taking into account wind power uncertainty. The first part generates a large number of possible wind power output scenarios based on the wind power output forecast curve of the operating day, and uses the K-means algorithm to cluster to obtain 5 output clustering scenarios with clustering ratios between 15% and 30% to characterize the uncertainty of wind power output on the operating day. The second part uses the clustering ratio of each scenario as a weight, and solves the percentage change of the total load of the system in each period, the weighted wind power output change value and the absolute value of wind power output. Then, the percentage change of the total load of the system in each period, the weighted wind power output change value and the absolute value of wind power output can be used as input, and the neural network model can be used to evaluate the possibility of period aggregation in each period, and the appropriate number of periods can be selected for aggregation, thereby reducing the unit combination variables. Finally, the clustering ratio of each scenario is used as a weight, and the weighted multi-scenario system total cost is minimized as the optimization goal to solve the unit combination model.

[0074] In this embodiment, the method for performing wind power multi-scenario clustering based on K-means includes the following steps:

[0075] Step S101: Randomly generate n possible output scenarios based on the expected and variance of wind power output on the operation day, and set clustering parameters , in size Randomly select from the dataset The data points are used to obtain the cluster center, and the data points are selected based on the Euclidean distance to form the initial cluster. ;

[0076] Step S102: updating the cluster center and selecting data points based on the Euclidean distance to form a new cluster;

[0077] Step S103: Using the square error sum criterion function Determine whether the target clustering condition is met, if not, return to step S102 for iteration, if yes, stop iteration.

[0078] Furthermore, the method of selecting data points based on Euclidean distance to form initial clusters is:

[0079] Calculate the data points To The cluster centers Euclidean distance , Representative Iterations;

[0080] like ,but .

[0081] Next, the method for updating the cluster center may be:

[0082] Calculate the average value of the data points in each of the initial clusters as the updated Iteration cluster centers:

[0083] ;

[0084] in, For the data points in the initial cluster of the cluster centers, is the total number of data points in all the initial clusters;

[0085] The method of selecting data points based on Euclidean distance to form a new cluster is the same as the method of selecting data points based on Euclidean distance to form an initial cluster.

[0086] Furthermore, the error square sum criterion function The method to determine whether the target clustering condition is met is:

[0087] Calculate the error square sum criterion function The value of Represents the number of iterations;

[0088] ;

[0089] Preset judgment threshold ;

[0090] If satisfied And if the clustering ratio is between 15% and 30%, the target clustering condition is reached.

[0091] A flowchart of the actual implementation of K-means-based wind power multi-scenario clustering can be found in Figure 1 ,First, n possible output scenarios are generated according to the wind power output prediction curve, and then the clustering parameter k is set to initialize k cluster centers, and then the possible scenarios are assigned to the nearest class until the cluster center no longer changes and the cluster ratio is between 15% and 30%.

[0092] Based on the assumption that the wind power output in each period follows a normal distribution, this embodiment regards the predicted wind power output value of the operating day as the mean of the normal distribution, and multiplies the predicted value by a random value between 0 and 0.25 as the standard deviation. Based on this premise, n possible wind power output scenarios are generated and clustered into k output clustering scenarios. In order to avoid the problem of insufficient representativeness due to too few clustering scenarios, or overfitting due to too many clustering scenarios, the proportion of possible scenarios contained in each output clustering scenario is controlled between 15% and 30%.

[0093] In the optimization of unit combination in multiple scenarios, it is necessary to ensure that the objective function is minimized in all scenarios. To this end, the clustering ratio of each scenario is used as the weight, and the ultimate goal of optimization is to minimize the total cost of the weighted multi-scenario system. As a preferred solution of this embodiment, the method for performing SCUC model period aggregation based on wind power uncertainty is:

[0094] Optimize the weighted multi-scenario objective function and set constraints;

[0095] Perform neural network-based time period aggregation;

[0096] Select the number of aggregation time periods.

[0097] As a preferred solution, the method of optimizing the weighted multi-scenario objective function and setting constraints is:

[0098] The multi-scenario objective function is:

[0099] ;

[0100] in, is the total number of scenes, is the total number of units, The total number of time periods divided into a day, For the scene Medium Unit In the period The output cost, For the scene Medium Unit In the period startup costs, For the scene The clustering ratio;

[0101] The constraints include:

[0102] Load balancing constraints:

[0103] ;

[0104] in, For the scene Medium Unit In the period of efforts, For the scene The total system load in For the scene Wind power output in

[0105] System positive and standby constraints:

[0106] ;

[0107] in, For the scene Medium Unit In the period The start-stop status, For the scene Medium Unit In the period The maximum output, For the scene Medium Unit of system positive spare capacity;

[0108] System negative standby constraints:

[0109] ;

[0110] in, For the scene Medium Unit In the period The minimum output, For the scene Medium Unit The system negative reserve capacity;

[0111] Output upper and lower limits:

[0112] ;

[0113] Ramp rate constraint:

[0114] ;

[0115] ;

[0116] in, For the crew The maximum climbing rate, For the crew Maximum downhill climbing rate;

[0117] Minimum continuous start and stop time constraints:

[0118] ;

[0119] ;

[0120] in, For the scene Medium Unit In the period Continuous boot time, For the scene Medium Unit In the period Continuous downtime. is the minimum startup time of the unit, The minimum downtime of the unit.

[0121] Specifically, the method for performing time period aggregation based on a neural network is preferably:

[0122] The relationship between the percentage change of total system load, the change value of weighted wind power output, the absolute value of weighted wind power output and the aggregation score of each period is established autonomously through a feedforward neural network, so as to judge the feasibility of period aggregation.

[0123] The setting of the time period aggregation score is based on the changes in the unit combination. If the unit combination remains unchanged in adjacent time periods, the aggregation score of the time period is 1, otherwise it is 0.

[0124] It is particularly noted that several parameters of this embodiment are specifically defined as:

[0125] Time Percentage change in total system load :

[0126] ;

[0127] in, and The time periods and Total system load;

[0128] Time Weighted wind power output change value :

[0129] ;

[0130] ;

[0131] in, For the scene Wind power in the period The output change, and Respectively for scenes and Wind power in the period The absolute value of output;

[0132] Time The absolute value of weighted wind power output :

[0133] ;

[0134] in, For the scene The clustering ratio.

[0135] Finally, the method for selecting the number of aggregation time periods is preferably:

[0136] Set the number of aggregation periods , sort the aggregate scores of each time period and select the one with the highest score Aggregate over time periods. .

[0137] The above operations can reduce the unit combination variables and simplify the solution of the SCUC model. However, the choice of the number of aggregation periods has a significant impact on the computational efficiency of the model. As the number of aggregation periods increases, the time period variables in the model decrease, and the calculation speed is significantly improved. However, the calculation speed does not continue to increase with the increase in the number of aggregation periods. When the number of aggregation periods exceeds a certain critical value, the number of scenarios that need to be considered for the calculation of this period increases significantly, and the number of constraints increases accordingly. The difficulty of the model in finding the optimal solution also increases, resulting in an increase in computational complexity and time. Therefore, as a further optimization solution, the number of aggregation periods is The value range of is preferably 5 to 7.

[0138] In summary, a SCUC model period aggregation process taking into account wind power uncertainty can be found in Figure 2 ,After calculating the absolute value and change rate of the weighted wind power output on the ,operating day, the clustering scores of each time period on the ,operating day can be solved by the neural network, and ,then the maximum aggregation number N is set to perform time period aggregation and ,finally solve the SCUC.

[0139] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A neural network-based time period aggregation method for a SCUC model with a high proportion of wind power, characterized in that: The following steps are involved: Wind power multi-scenario clustering based on K-means; SCUC model time period aggregation based on wind power uncertainty; The method for performing wind power multi-scenario clustering based on K-means comprises the following steps: Step S101: Randomly generate n output scenarios based on the expected and variance of wind power output on the operating day, and set clustering parameters , in size Randomly select from the dataset The data points are used to obtain the cluster center, and the data points are selected based on the Euclidean distance to form the initial cluster. ; Step S102: updating the cluster center and selecting data points based on the Euclidean distance to form a new cluster; Step S103: Using the square error sum criterion function Determine whether the target clustering condition is met, if not, return to step S102 for iteration, if yes, stop iteration; The method for period aggregation of SCUC model based on wind power uncertainty is as follows: Optimize the weighted multi-scenario objective function and set constraints; Perform neural network-based time period aggregation; Select the number of aggregation time periods; The method for optimizing the weighted multi-scenario objective function and setting constraints is: The multi-scenario objective function is: ; in, is the total number of scenes, is the total number of units, The total number of time periods divided into a day, For the scene Medium Unit In the period The output cost, For the scene Medium Unit In the period startup costs, For the scene The clustering ratio; The constraints include: Load balancing constraints: ； in, For the scene Medium Unit In the period of effort, For the scene The total system load in For the scene Wind power output in System positive and standby constraints: ; in, For the scene Medium Unit In the period The start and stop status, For the scene Medium Unit In the period The maximum output, For the scene Medium Unit of system positive spare capacity; System negative standby constraints: ; in, For the scene Medium Unit In the period The minimum output, For the scene Medium Unit The system negative reserve capacity; Output upper and lower limits: ； Ramp rate constraint: ； ； in, For the crew The maximum climbing rate, For the crew Maximum downhill climbing rate; Minimum continuous start and stop time constraints: ; ; in, For the scene Medium Unit In the period Continuous startup time, For the scene Medium Unit In the period Continuous downtime. is the minimum startup time of the unit, The minimum downtime of the unit; The method for performing time period aggregation based on a neural network is: The feedforward neural network is used to autonomously establish the relationship between the total system load change percentage, weighted wind power output change value, and the absolute value of weighted wind power output and the period aggregation score in each period; If the unit commitment remains unchanged in adjacent time periods, the aggregation score of the time period is 1, otherwise it is 0; The method for selecting the number of aggregation time periods is: Set the number of aggregation periods , sort the aggregate scores of each time period and select the one with the highest score Aggregate over time periods.

2. The neural network-based SCUC model with high wind power proportion period aggregation method according to claim 1 is characterized in that: The method for selecting data points based on Euclidean distance to form initial clusters is: Calculate the data points To The cluster centers Euclidean distance , Representative Iterations; like ,but .

3. The neural network-based SCUC model with high wind power proportion period aggregation method according to claim 2 is characterized in that: The method for updating the cluster center is: Calculate the average value of the data points in each of the initial clusters as the updated Iteration cluster centers: ； in, For the data points in the initial cluster of the cluster centers, is the total number of data points in all the initial clusters; The method of selecting data points based on Euclidean distance to form a new cluster is the same as the method of selecting data points based on Euclidean distance to form an initial cluster.

4. The neural network-based SCUC model with high wind power proportion period aggregation method according to claim 3 is characterized in that: The error square sum criterion function The method to determine whether the target clustering condition is met is: Calculate the error square sum criterion function The value of Represents the number of iterations; ； Preset judgment threshold ; If satisfied And if the clustering ratio is between 15% and 30%, the target clustering condition is reached.

5. The neural network-based SCUC model with high wind power proportion period aggregation method according to claim 1 is characterized in that: The number of aggregation periods The value range is 5 to 7.

Images