A data-driven based method for reducing redundant constraints in power market clearing simulation

Abstract

The application discloses a data-driven power market clearing simulation redundant constraint reduction method. It aims to improve the efficiency and accuracy of power market clearing simulation optimization. It includes the following steps: collecting historical data of power market clearing and generating a large number of power market operation scenarios. Run the power market clearing model under different scenarios to generate corresponding clearing results, which are used as simulation data. Combine historical and simulation data as a data set, calculate the margin of each data constraint, and preliminarily judge the redundant constraint. The data set is used as input data, the clearing power of each unit is used as output data, and the historical data is divided into training set and test set. The training set is input into the random forest model for training, and the test set is used for testing accuracy. Under a certain market operation environment, the final power market clearing model after reduction is obtained by running, and the power market clearing result is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to a power market clearing method, more particularly to a data-driven power market clearing simulation redundant constraint reduction method. BACKGROUND

[0002] With the rapid development of renewable energy, the complexity of the power market has increased significantly, especially when dealing with multiple energy types and large-scale energy access, the operating conditions of the power system and the market clearing model face more and more constraints. The traditional power market clearing process often relies on fixed constraints to ensure the stability and safety of the power grid. However, some of these constraints may not be necessary, and too many redundant constraints will increase the complexity of the calculation, affecting the efficiency and accuracy of the market clearing. The commonly used redundant constraint identification method mainly relies on experience judgment, lacks systematicness and accuracy, and cannot effectively respond to the dynamic changes of the market environment, such as the patent number 202110034248.X, the name of the invention patent for inter-provincial and intra-provincial power market coupling clearing method, system, device and medium, which does not deal with redundant constraints, so there are the above problems. SUMMARY

[0003] In view of the deficiencies in the prior art, the purpose of the present application is to provide a data-driven power market clearing simulation redundant constraint reduction method, aiming to improve the efficiency and accuracy of the power market clearing process. This method first collects and processes the historical data of the power market, including the bidding information of the generator unit, the clearing quantity data and the load demand, etc., and generates a large number of simulation scenarios. In these scenarios, the corresponding clearing results are obtained by running the power market clearing model. The simulation data and historical data are combined as a research data set, and the redundancy of each constraint is calculated. Then, by using the random forest model to train the data set, the importance of each input feature is evaluated, and the redundant constraint set is identified. Finally, by gradually deleting the redundant constraints, repeatedly running the simulation model and verifying the optimization effect, the simplified power market clearing model is finally obtained.

[0004] To achieve the above purpose, the present application provides the following technical scheme: a data-driven power market clearing simulation redundant constraint reduction method, characterized by comprising the following steps:

[0005] Step 1, input the historical operation data of the power system and the historical clearing data of the power market, and clean the input system historical data, including deleting outliers and filling missing values;

[0006] Step two, based on the historical prediction data in the historical data after cleaning in step one, a large number of scenes are generated, the power market clearing model is introduced to generate the clearing results of the power market under each scene, the historical data and the simulation data are combined into a data set, and the preliminary judgment of the redundant constraint set is made according to the clearing result data;

[0007] Step three, the data set in step two is divided into a training set and a test set, the random forest model is trained using the training set, and the model is tested for accuracy using the test set; the importance of different constraints is sorted, and a threshold is set to determine that the constraint below the threshold is a preliminary redundant constraint;

[0008] Step four, input the current power market operation scene data into the power market clearing model, and gradually verify the reliability of each redundant constraint by deleting the redundant constraint one by one, and determine whether the deletion of a redundant constraint causes the model to fail to clear effectively, if yes, update the constraint set, if no, do not update, until the final redundant constraint is judged to be complete, and the final redundant constraint is obtained;

[0009] Step five, the final power market clearing model after reduction in step four is put into the market operation environment to run, and the power market clearing result is obtained, and the power market clearing simulation redundant constraint reduction is completed.

[0010] As a further improvement of the application, the historical data of the power system in step one includes: network transmission limit data, system load data and new energy unit predicted output, the power market historical data includes: the bidding amount, clearing amount and bidding price of each market participating unit, the bidding amount, clearing amount and bidding price of each market load, and the bidding amount, clearing amount and bidding price of each new energy unit, wherein the clearing amount is denoted as Y h , The jth data of the i data category is denoted as X h , The jth data of the i data category is denoted as X

[0011] As a further improvement of the application, the specific way of data cleaning in step one is:

[0012] Step one, use Z-score method to detect outliers:

[0013]

[0014] Wherein And Are the mean and variance of , when Determine the outliers; delete the identified outliers;

[0015] Step one two, check data missing value;

[0016] Step one three, fill in the missing and deleted values checked in step two, specifically, use known data to input the power market clearing model to calculate some special missing data, and the remaining missing values that cannot be calculated by the model are supplemented by average value calculation;

[0017] Step one four, normalize the input data X h .

[0018]

[0019] The normalized value is

[0020] As a further improvement of the application, the power market clearing model in step one three is as follows:

[0021] Where F is the objective function, which means the minimum system cost or the maximum social welfare; g, w, d are the numbers of ordinary units, new energy units and loads respectively; G, W, D are the quantities of ordinary units, new energy units and loads respectively; α g , α w , α d are the bidding prices of ordinary units, new energy units and loads respectively; are the clearing quantities of ordinary units, new energy units and loads respectively;

[0022] The related constraints are:

[0023]

[0024]

[0025] Where are the bidding quantities of ordinary units, new energy units and loads respectively; S L is the conversion matrix; are the power flow and its upper and lower limits of each line respectively.

[0026] As a further improvement of the application, the historical prediction data in step two includes the predicted output of new energy units of the power system and the predicted load According to the historical prediction data, the average and variance of the two can be calculated, denoted as and According to these data, S scenarios can be generated by the Monte Carlo sampling method, in the s scenario, the output of the new energy unit is and the load is After inputting data X h Normalization is performed:

[0027]

[0028] The normalized values, the imported power market clearing model is as follows:

[0029]

[0030] Wherein, F is the objective function, which means the minimum system cost or the maximum social welfare; g, w, d are the numbers of ordinary units, new energy units and loads respectively; G, W, D are the quantities of ordinary units, new energy units and loads respectively; α g , α w , α d are the bidding prices of ordinary units, new energy units and loads respectively; are the clearing quantities of ordinary units, new energy units and loads respectively;

[0031] The related constraints are:

[0032]

[0033]

[0034] Wherein, are the bidding quantities of ordinary units, new energy units and loads respectively; S L is the conversion matrix; are the power flows of each line and their upper and lower limits.

[0035] As a further improvement of the present application, the specific way of combining the historical data and the simulation data into a data set is: respectively running the clearing to obtain the corresponding clearing data, denoted as The total simulation data set is denoted as Then the historical data set and the simulation data set are combined to obtain the data set

[0036] For any constraint Calculate the margin Determine whether u ij is greater than the threshold value, which is set to 0.4, if it is greater than the threshold value, the constraint is preliminarily identified as a redundant constraint, and the set of identified redundant constraints is denoted as As a further improvement of the present application, the specific way of dividing the data set into training set and test set in step three, using the training set to train the random forest model, and using the test set to test the accuracy of the model is as follows:

[0037] Step three one, divide the data set [X ij , Y ij ] into training set and test set , the ratio is 7:3;

[0038] Step three two, input the training set into the random forest model for training;

[0039] Step three three, use the test set for prediction, and obtain the predicted value , and calculate the mean square error (MSE) according to the following formula:

[0040]

[0041] Wherein, n test is the number of tests.

[0042] As a further improvement of the present application, in step three three, the input variable X ij of different kinds i is judged for importance, in particular as follows:

[0043]

[0044] Wherein, N t is the number of inputs X ij , Importance(X i ) represents the importance of X i , and the importance of each X i is sorted, and if Importance(X i ) is less than θ, the value is set to 0.1, and if a certain X i is less than Importance(X i ), the constraint related to X i is determined as a redundant constraint, and is recorded as

[0045] As a further improvement of the present application, in step four, the specific way of judging whether the removal of a redundant constraint causes the model to be unable to effectively clear is as follows: input the value X ijs of a certain scene s into the power market clearing model M(·), judge whether the model can normally clear, and whether the clearing result is reasonable, if the model can normally clear and the result is reasonable, the constraint to be judged is an actual redundant constraint, then the constraint in U j (n) is updated to be deleted from U j(n+1); otherwise, the judged constraint is not a redundant constraint. Each preliminary identified redundant constraint is deleted one by one until the power market clearing model under 95% of the scenarios can be normally cleared, and the number of identified redundant constraints is unchanged, then the identification is completed. Until all the to-be-judged matrix tests are completed, the obtained U j (N) is the identified redundant constraint.

[0046] As a further improvement of the present application, the specific way of putting the final power market clearing model after the reduction in step four into the market operation environment for running to obtain the power market clearing result in step five is to determine the power market operation environment under which the power market clearing model is run The input of the power market clearing model M(U j (N)) after deleting all redundant constraints can obtain the result of the power market clearing The present application has the following beneficial effects:

[0047] 1) By identifying and deleting redundant constraints, the present application significantly reduces the computational complexity of the power market clearing, shortens the clearing time, improves the response speed of the market, and ensures the reliability and economy of the power supply.

[0048] 2) The present application has higher adaptability by identifying redundant constraints in real time, can flexibly respond to market changes, and timely adjust the constraint conditions, thereby improving the flexibility and efficiency of the market operation. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The flowchart of the present application based on the data-driven power market clearing simulation redundant constraint reduction method. DETAILED DESCRIPTION

[0050] The present application will be further described in detail below with reference to the embodiments given in the accompanying drawings.

[0051] Referring to Figure 1 The present application based on the data-driven power market clearing simulation redundant constraint reduction method, including the following steps:

[0052] Step 1) Input the historical operation data of the power system and the historical clearing data of the power market, and perform data cleaning on the input system historical data, delete outliers, fill in missing values, etc.

[0053] Step 2) Generate a large number of scenarios according to the historical prediction data, import the power market clearing model to generate the clearing results of the power market under each scenario, combine the historical data and the simulation data into a data set, and preliminarily judge the redundant constraint set according to the clearing result data.

[0054] Step 3) Divide the data set into training set and test set, use the training set to train the random forest model, and use the test set to test the accuracy of the model. Get the importance ranking of different constraints, set the threshold, and determine the preliminary redundant constraints below the threshold.

[0055] Step 4) Input each power market operation scenario data into the power market clearing model, and gradually verify the reliability of each redundant constraint to determine whether the deletion of a redundant constraint leads to the model unable to effectively clear, and update the constraint set. If the final redundant constraint is determined, the final redundant constraint is obtained.

[0056] The step 1) is specifically:

[0057] The historical operation data of the power system and the historical clearing data of the power market are input, and the historical data of the input system are cleaned, including deleting abnormal values and filling missing values.

[0058] The historical data of the power system includes but is not limited to: network transmission limit data, system load data, new energy unit predicted output, etc.

[0059] The historical data of the power market includes but is not limited to: the bidding amount, clearing amount and bidding price of each market participating unit, the bidding amount, clearing amount and bidding price of each market load, the bidding amount, clearing amount and bidding price of each new energy unit, etc.

[0060] The clearing amount is denoted as Y h , wherein represents the jth data of the data category i.

[0061] Other system data is denoted as X h , wherein represents the jth data of the data category i.

[0062] The historical data of the input system are cleaned, including deleting abnormal values and filling missing values.

[0063] ① Check and process abnormal values

[0064] Use Z-score method to detect abnormal values:

[0065]

[0066] , wherein and are the mean and variance of , when determined as abnormal values.

[0067] Delete the identified abnormal values.

[0068] Check data missing value.

[0069] Fill in the missing and deleted values.

[0070] For some special missing data, the known other data can be used to input the power market clearing model for calculation. Other missing values that cannot be calculated by the model can be calculated by the average value.

[0071] The specific power market clearing model is:

[0072] Objective function:

[0073]

[0074] Where F is the objective function, which means the minimum system cost or the maximum social welfare; g, w, d are the numbers of conventional units, new energy units and loads, respectively; G, W, D are the quantities of conventional units, new energy units and loads, respectively; α g , α w , α d are the bidding prices of conventional units, new energy units and loads, respectively; are the clearing quantities of conventional units, new energy units and loads, respectively.

[0075] The related constraints are:

[0076]

[0077] Where are the bidding quantities of conventional units, new energy units and loads, respectively; S L is the conversion matrix; are the power flow and its upper and lower limits of each line.

[0078] Step 4) Normalize the input data X h

[0079]

[0080] is the normalized value.

[0081] The step 2) is specifically:

[0082] According to the historical prediction data, a large number of scenarios are generated, the power market clearing model is imported to generate the clearing results of the power market under each scenario, and the redundant constraint set is preliminarily judged according to each clearing result.

[0083] The historical prediction data includes the predicted output of the new energy units of the power system and the predicted load ​The average and variance of the two can be calculated according to the historical prediction data, and are denoted as And According to these data, S scenarios can be generated by Monte Carlo sampling method. In the s-th scenario, the output of the new energy unit is The load quantity is

[0084] The input data X h is normalized.

[0085]

[0086] is the normalized value.

[0087] In different scenarios s, the corresponding power system scenario data is input into the power market clearing model.

[0088] The specific power market clearing model is:

[0089] Objective function:

[0090]

[0091] Where F is the objective function, which means the minimum system cost or the maximum social welfare; g, w, d are the numbers of ordinary units, new energy units and loads, respectively; G, W, D are the quantities of ordinary units, new energy units and loads, respectively; α g , α w , α d are the bidding prices of ordinary units, new energy units and loads, respectively; are the clearing quantities of ordinary units, new energy units and loads, respectively.

[0092] The related constraints are:

[0093]

[0094] Where are the bidding quantities of ordinary units, new energy units and loads, respectively; S L is the conversion matrix; are the power flows of each line and their upper and lower limits, respectively.

[0095] The clearing corresponding to each scenario can be obtained by running the clearing, denoted as The total simulation data set is denoted as

[0096] The historical data set and the simulation data set are combined to obtain the data set

[0097] For any constraint The margin can be calculated Determine u ij whether it is greater than a threshold value set to 0.4, if greater than the threshold value, the constraint is preliminarily identified as a redundant constraint, and the identified redundant constraint set is denoted as

[0098] The step 3) is specifically:

[0099] The historical data and simulation data set are divided into a training set and a test set, the random forest model is trained using the training set, and the model is tested for accuracy using the test set. The importance of different constraints is sorted, and a threshold value is set to determine that the constraint below the threshold value is a preliminary redundant constraint.

[0100] The data set [X ij ,Y ij ] is divided into a training set and a test set in a ratio of 7:3.

[0101] The training set is input into the random forest model for training. The specific random forest model is not described here.

[0102] The test set is used for prediction, and the predicted value can be calculated.

[0103]

[0104] Where n test is the number of tests.

[0105] The input variable (constraint) X ij of different types i can be judged for importance, and the specific method is according to the following formula:

[0106]

[0107] Where N t is the number of inputs X ij , and Importance(X i ) represents the importance of X i .

[0108] The importance of each X i is sorted, and if Importance(X i ) is less than θ, the value is set to 0.1, and if X i is less than Importance(X i ), X iThe relevant constraint judgment is a redundant constraint, denoted as

[0109] The step 4) is specifically:

[0110] The power market operation scenario data is input into the power market clearing model, the reliability of each redundant constraint is gradually verified, it is judged whether the deletion of a redundant constraint leads to the model being unable to effectively clear, and the constraint set is updated, if the final redundant constraints are all judged, the final redundant constraint is obtained.

[0111] The total set of redundant constraints obtained in steps 2 and 3 Each constraint in the total set of redundant constraints obtained in steps 2 and 3 is tested by deletion in the power market clearing model mentioned in step 2.

[0112] The operation value X of a certain scenario s ijs The power market clearing model M(·) is input, it is judged whether the model can normally clear, and whether the clearing result is reasonable. If the model can normally clear and the result is reasonable, the constraint to be judged is an actual redundant constraint, then the constraint in U j (n) is updated to U j (n+1) by deletion; otherwise, the constraint to be judged is not a redundant constraint. Each of the preliminary identified redundant constraints is deleted one by one until the power market clearing model under 95% scenarios can normally clear, and the number of identified redundant constraints is unchanged, then the identification is completed. Until all the to-be-judged matrices are tested, the obtained U j (N) is the identified redundant constraint.

[0113] The step 5) is specifically:

[0114] Under a certain market operation environment, the final power market clearing model after deletion is obtained by operation, and the power market clearing result is obtained.

[0115] Under a certain market operation environment The power market clearing result j (N)) can be obtained by inputting the power market clearing model M(U

[0116] To sum up, the embodiment discloses a data-driven power market clearing simulation redundant constraint reduction method, aiming to improve the efficiency and accuracy of power market clearing simulation optimization. The method includes the following steps: collecting historical data of power market clearing, including generator bidding data and clearing quantity data, load demand, new energy unit clearing quantity and related constraint execution, and generating a large number of power market operation scenarios according to historical new energy prediction and load. Run the power market clearing model under different scenarios to generate corresponding clearing results as simulation data. Combine historical and simulation data as a data set, calculate the margin of each constraint corresponding to the data, and preliminarily judge the redundant constraint. Take the data set as input data and the clearing power of each unit as output data, divide the historical data into training set and test set, input the training set into the random forest model for training, and test the accuracy with the test set. The importance parameters of different inputs can be obtained to identify the preliminary redundant constraint. Input the power market running environment information of different scenarios into the power market clearing model for simulation and verification, and delete the preliminary redundant constraint step by step, update the constraint set, and if the final redundant constraint is judged, the final redundant constraint is obtained. Under a certain market operation environment, the final power market clearing model after reduction is run to obtain the power market clearing result.

[0117] The above only describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as the protection scope of the present application.

Claims

1. A data-driven based power market clearing simulation redundancy constraint reduction method, characterized in that: Comprise the following steps: Step one, input the historical operation data of power system and the historical clearing data of power market, clean the input system historical data, including deleting outliers and filling missing values; Step two, based on the historical prediction data in the historical data cleaned in step one, generate a large number of scenarios, import the power market clearing model to generate the clearing results of the power market under each scenario, combine the historical data and the simulation data into a data set, and preliminarily judge the redundant constraint set according to the clearing result data; Step three, divide the data set in step two into training set and test set, respectively use the training set to train the random forest model, and use the test set to test the accuracy of the model; Get the importance ranking of different constraints, and set a threshold to determine that the constraints below the threshold are the preliminary redundant constraints; Step four, input the current power market operation scenario data into the power market clearing model, and gradually test the reliability of each redundant constraint by deleting redundant constraints one by one, judge whether the deletion of a redundant constraint leads to the model unable to effectively clear, if yes, update the constraint set, if not, do not update, until the final redundant constraint is judged to be completed, then get the final redundant constraint; Step five, put the final power market clearing model after the reduction in step four into the market operation environment to run and get the power market clearing result, complete the power market clearing simulation redundant constraint reduction.

2. The data-driven based power market clearing simulation redundancy constraint reduction method of claim 1, wherein: The historical data of the power system in the step one includes network transmission limit data, system load data and new energy unit predicted output, the power market historical data includes the bidding quantity, clearing quantity and bidding price of each market participating unit, the bidding quantity, clearing quantity and bidding price of each market participating load, the bidding quantity, clearing quantity and bidding price of each new energy unit, wherein the clearing quantity is denoted as Y h , The data of the data category i is denoted as Xij h , The data of the data category i is denoted as Xij 3. The data-driven based power market clearing simulation redundancy constraint reduction method of claim 2, wherein: The specific way of data cleaning in step one is: Step one, use Z-score method to detect outliers: wherein with are respectively the mean and variance of, when an outlier; deleting the identified outliers; Step two, check the missing data; Step three, fill in the deleted and missing values checked in step two, specifically, use the known data to input the power market clearing model to calculate some special missing data, and use the average value to calculate the remaining missing values that cannot be calculated by the model; Step one four, normalizing the input data X h ; Normalized values.

4. The data-driven based power market clearing simulation redundancy constraint reduction method of claim 3, wherein: The power market clearing model in step three is as follows: Where F is the objective function, which means the minimum system cost or the maximum social welfare; g, w, d are the numbers of conventional units, new energy units, and loads, respectively; G, W, D are the quantities of conventional units, new energy units, and loads, respectively; a g , a w , a d are the bidding prices of conventional units, new energy units, and loads, respectively; are the clearing quantities of conventional units, new energy units, and loads, respectively. Its related constraints are: wherein are the normal unit, new energy unit, and load bidding amount, respectively; S L is the conversion matrix; P l L , P l L are the power flow and its upper and lower limits of each line, respectively.

5. The data-driven based power market clearing simulation redundancy constraint reduction method according to any one of claims 1 to 4, characterized in that: The historical prediction data in the second step includes predicted output of new energy units of the power system and predicted load According to the historical prediction data, the average and variance of the two can be calculated, respectively denoted as and According to these data, S scenarios can be generated by the Monte Carlo sampling method. In the s scenario, the output of the new energy unit is and the load is Then, the input data X h is normalized: For the normalized values, the imported power market dispatch model is as follows: Wherein, F is the target function, which means the minimum system cost or the maximum social welfare; g, w, d are the numbers of conventional units, new energy units and loads respectively; G, W, D are the quantities of conventional units, new energy units and loads respectively; α g , α w , α d are the bidding prices of conventional units, new energy units and loads respectively; are the clearing quantities of conventional units, new energy units and loads respectively. Its related constraints are: wherein, are the normal units, new energy units, and load bidding quantities, respectively; S L is a conversion matrix; P l L , P l L are the power flow and its upper and lower limits of each line, respectively.

6. The data-driven based power market clearing simulation redundancy constraint reduction method of claim 5, wherein: The specific way of combining the historical data and the simulation data into a data set is: running out the clearings respectively to obtain the corresponding clearing data The total simulation data set is denoted as Then the historical data set and the simulation data set are combined to obtain a data set For any constraint Compute the margin Judge u ij whether it is greater than a threshold value, set to 0.4, if greater than the threshold value, the constraint is preliminarily identified as a redundant constraint, and the set of identified redundant constraints is recorded as 7. The data-driven based power market clearing simulation redundancy constraint reduction method according to any one of claims 1 to 4, characterized in that: The specific way of dividing the training set and test set in step three is as follows: Step three, divide the data set [X ij , Y ij ] into training set and test set in the ratio of 7:3; Step three two, train the set input the random forest model for training; Step three, using the test set Making a prediction, resulting in a predicted value At the same time, the mean square error MSE is calculated according to the following formula: wherein n test is the number of tests.

8. The data-driven based power market clearing simulation redundancy constraint reduction method of claim 7, wherein: The step three of the method is to determine the importance of the input variables X ij The importance is determined in the following way: where N t is the number of inputs X ij , Importance(X i ) represents the importance of X i , and the importance of each X i is ranked. If Importance(X i ) is less than θ, the value is set to 0.

1. If a certain X i is less than Importance(X i ), the constraint related to X i is determined to be a redundant constraint, denoted as 9. The data-driven based power market clearing simulation redundancy constraint reduction method according to any one of claims 1 to 4, characterized in that: The specific way of judging whether the removal of a certain redundant constraint causes the model to be unable to effectively clear is: running the value X of a certain scenario s ijs The input power market clearing model M(·) is judged whether the model can normally clear and whether the clearing result is reasonable. If the model can normally clear and the result is reasonable, the constraint to be judged is an actual redundant constraint, and U j The constraint removal update in (n) is U j (n+1); otherwise, the constraint to be judged is not a redundant constraint. Each initially identified redundant constraint is added and removed one by one until the power market clearing model under 95% scenarios can normally clear, and the number of identified redundant constraints is unchanged, and then the identification is completed. Until all the to-be-judged matrixes are completed, the obtained U j (N) is the identified redundant constraint.

10. The data-driven based power market clearing simulation redundancy constraint reduction method of claim 9, wherein: The specific way of putting the final power market clearing model after the reduction in step four into the market operation environment to obtain the power market clearing result is to determine the power market clearing result under a certain market operation environment Input the power market clearing model M(U j (N)) after deleting all redundant constraints, and the power market clearing result can be obtained

Images