A dynamic evaluation method of market power of generator sets based on grid search and decision tree

Abstract

This invention relates to a dynamic evaluation method for the market power of generating units based on grid search and decision trees. The method comprises: 1) collecting data on bidding generating units; 2) calculating unit indicator values ​​based on the collected data and the power market power risk monitoring indicator system, forming an indicator value vector describing the market behavior characteristics of the generating units, and using an expert system to label the unit categories in the set, thus constructing a market power characteristic dataset for bidding generating units; 3) finding the optimal parameters of the unit market power evaluation model through grid search and cross-validation algorithms, and creating a decision tree for unit market power evaluation based on the optimal parameters; 4) evaluating the exercise of unit market power using the constructed decision tree evaluation model in the next clearing period, forming a new database of market power characteristics for bidding generating units, which is used to correct the unit market power evaluation model. Compared with existing technologies, this invention has advantages such as simplifying the evaluation process and improving identification accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of power market, and in particular to a method for dynamically evaluating the market power of a generator set based on grid search and decision tree. Background Art

[0002] Market power refers to the ability of certain market players to change the market price level to make it deviate from the perfect competition price level and make a profit. The market power of the electricity market mainly refers to the ability of power generators to make the market price deviate from the perfect competition level and make a profit from it. The market power control of the electricity market is also mainly aimed at the market behavior of power generators abusing their market power. Therefore, measuring and preventing (mitigating) market power is an important part of the design of the electricity market (structure, model, rules).

[0003] Abuse of market power in the electricity market mainly refers to the risk brought about by market participants using their market power to make electricity prices continue to be high or fluctuate frequently and significantly. It can exist in various electricity commodity transactions such as electricity, spare capacity, and ancillary services, as well as in different market participants such as power generators, power sales companies, and power users. Traditional methods for evaluating the market power of generators mainly include the analytic hierarchy process, the G1 method, the ordinal relationship method, the entropy weight method, the grey correlation method, the fuzzy comprehensive evaluation method, etc. There is little historical data in the early stage of market construction. The above market power evaluation methods have good results in evaluating the market power of generators, but the evaluation process is relatively complex and is subject to a large degree of subjective influence. In addition, the electricity market is a relatively special market, and traditional market power evaluation methods cannot adapt themselves to changes in the way units exercise their market power.

[0004] Therefore, in order to ensure fair competition and healthy development of the market, there is an urgent need for a market power evaluation method that can be applied to the abuse of market power in the electricity market. This method can effectively process massive data and conform to the actual operation of the electricity market, while avoiding the problems of dimensionality disaster and excessive computation, so that market regulators can quickly and effectively identify the abuse of market power by power generators. Moreover, this method must start from the characteristics of market power abuse behavior and accurately judge the market power risk in a combination of qualitative and quantitative methods. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method for dynamically evaluating the market power of power generation units based on grid search and decision tree.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for dynamic evaluation of market power of power generating units based on grid search and decision tree, the method comprising:

[0008] Step 1: Collect bidding unit data including basic unit parameters, unit transaction behavior data, market settlement data and market management data through the power market technical support system.

[0009] The basic parameters of the unit include unit type, unit rated capacity, unit minimum technical output, unit status and unit maximum climbing power; the unit transaction behavior data include unit participation market quantity and price declaration data and unit clearing data; the market settlement data include market supply and demand data and market settlement electricity price; the market management data includes the deviation between the actual output of the unit and the cleared electricity quantity, electricity fee payment status and credit guarantee performance status.

[0010] Step 2: Based on the various data collected in step 1 and the power market power risk monitoring index system, calculate the group index value, form the index value vector describing the market behavior characteristics of the unit, use the expert system to label the unit categories in the set, and construct the market power characteristic data set of the bidding unit. Among them:

[0011] The power market power risk monitoring index system includes bidding strategy indicators and clearing result indicators. The types of units in the set are marked by an expert system to form a market power characteristic data set of bidding units, where:

[0012] The bidding strategy indicators include the unit's declared electricity market share, the unit's average bidding relative level indicator, the unit's high bidding quantity and price index, the unit's high bidding quantity and price index relative level, the unit's high bidding ratio and the unit's high bidding ratio relative level indicator;

[0013] The clearing result indicators include the market share of the winning electricity volume, the relative level of the unit's winning rate, and the unit's out-of-sequence capacity index;

[0014] The expert system consists of an expert knowledge database and an inference engine, wherein:

[0015] The expert knowledge database includes reasoning rules based on transaction behavior, reasoning rules based on risk assessment, and reasoning rules based on market performance;

[0016] The inference engine analyzes the unit's behavior in participating in the market according to the inference rules in the expert knowledge base. When the unit reaches the risk threshold, it is determined that the generator unit is very likely to use market power to disrupt the market, and the unit is marked as a risky unit that uses market power to make a profit.

[0017] The expression of the relative level index of the average quotation of the unit is:

[0018]

[0019] In the formula, is the average relative price of unit i; y i,t is the profit margin of unit i participating in the market at time t; is the mean value of profit margin of unit i during the statistical period;

[0020] The unit high bid quantity price index is a relative level index of the unit high bid quantity price index, and its expression is:

[0021]

[0022] In the formula, and L i,t,h are the high bid quantity-price index and the relative level index of the high bid quantity-price index of unit i at time t, respectively, and N is the number of all bidding units in the market;

[0023] The expression of the relative level index of the unit high bid ratio is:

[0024]

[0025] In the formula, is the relative level of the high bid ratio of unit i at time t; R i,t,h is the high bid ratio of unit i at time t; N is the number of all bidding units in the market.

[0026] Step 3: Based on the market power characteristic data set of the bidding unit in step 2, find the optimal parameters of the unit market power evaluation model through grid search and cross-validation algorithm, and create a unit market power evaluation decision tree according to the optimal parameters; the specific steps include:

[0027] 31) Based on the power market power risk monitoring indicator system and the historical transaction data of the power market, the computer group indicator values ​​are calculated to form a complete sample, and the units that use market power in the set are marked through the expert system to form a market power characteristic data set for the bidding units;

[0028] 32) Construct a decision tree evaluation model, find the hyperparameters that make the decision tree classification effect best through grid search and cross-validation, and establish a unit market power evaluation model based on the optimal hyperparameters.

[0029] The specific steps of building a decision tree evaluation model include:

[0030] a) Determine the hyperparameters and value ranges that need to be adjusted for the model based on the size of the market and the number of indicators, and preset evaluation indicators as evaluation criteria for the decision tree; the decision tree model hyperparameters include the selection method of the optimal indicator, the maximum depth of the decision tree, and the strategy for dividing the decision tree nodes; the selection method of the optimal indicator includes two parameters: the Gini coefficient and the information gain;

[0031] b) constructing a hyperparameter combination based on the hyperparameter adjustment range;

[0032] c) Divide the training set data set into k disjoint subsets;

[0033] d) Select a set of decision tree hyperparameters, use any k-1 subsets in the training set to train the decision tree, and calculate the evaluation index of the decision tree based on the remaining subset. After k iterations, the evaluation index E of k decision trees under this hyperparameter combination can be obtained. i , and finally take its average value as the evaluation value of the decision tree under this hyperparameter combination;

[0034] e) Traversing all hyperparameter combinations, selecting a set of parameters with the highest decision tree evaluation value in step d) to construct a decision tree, and obtaining the unit market power evaluation model with the best classification effect.

[0035] Step 4: In the next clearing period, the constructed decision tree evaluation model is used to evaluate the market power of the units, and a new market power characteristic database of bidding units is formed to correct the market power evaluation model of the units. The specific contents are:

[0036] In the next clearing period, the exercise of the unit's market power is evaluated through the constructed decision tree evaluation model. Each unit traverses the decision tree evaluation model to complete the evaluation of the unit's market power and form a new bidding unit market power feature database to correct the unit's market power evaluation model and achieve closed-loop management of the unit's market power evaluation.

[0037] The method for dynamic evaluation of market power of power generating units based on grid search and decision tree provided by the present invention has at least the following beneficial effects compared with the prior art:

[0038] 1) The method for dynamic evaluation of market power of power generating units based on grid search and decision tree of the present invention includes four steps: data collection of bidding units, construction of market power characteristic data set of bidding units, construction of market power evaluation model of units and correction of market power evaluation model. It can effectively process massive data and conform to the actual operation of power market, while avoiding the problems of dimensionality disaster and excessive computation, and can greatly simplify the evaluation process.

[0039] 2) The method of the present invention can learn labeled training samples and establish a learning process. By comparing the predicted results with the actual results of the "training data", the classifier model is continuously adjusted until the classification results of the model reach an expected accuracy rate, which greatly improves the market power evaluation efficiency and market power identification accuracy of the market operating organization, and inputs the evaluation results of the next time period as samples into the database to achieve closed-loop control of the unit market power evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of a method for dynamically evaluating the market power of a generator set based on grid search and decision tree in an embodiment. DETAILED DESCRIPTION

[0041] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should belong to the scope of protection of the present invention.

[0042] Example

[0043] The present invention relates to a method for dynamically evaluating the market power of a generator set based on grid search and decision tree. The method utilizes a decision tree algorithm to discover the strong relationship characteristics between the index value of the generator set and the potential results, and establishes a market power evaluation model for the generator set, so as to greatly improve the market power evaluation efficiency and market power identification accuracy of the market operation organization.

[0044] like Figure 1 As shown, the specific process of the method includes the following steps:

[0045] S1: Data collection steps for bidding units:

[0046] Through the power market technical support system, unit data is collected, including basic unit parameters, unit transaction behavior data, market settlement data and market management data.

[0047] S2: Steps for constructing the market power characteristics dataset of bidding units:

[0048] Based on the various data collected in S1 and the power market power risk monitoring index system, the computer group index values ​​are calculated to form an index value vector describing the market behavior characteristics of the unit. The expert system is used to label the unit categories in the set and construct a market power characteristic data set for the bidding unit.

[0049] S3: Steps for constructing the unit market power evaluation model:

[0050] Based on the market power characteristic data set of bidding units in S2, the optimal parameters of the market power evaluation model are found through grid search and cross-validation algorithms, and a unit market power evaluation decision tree is created according to the optimal parameters.

[0051] S4: Market power evaluation model modification steps:

[0052] In the next clearing period, the constructed decision tree evaluation model is used to evaluate the exercise of the unit's market power, forming a new bidding unit market power characteristic database, thereby correcting the unit's market power evaluation model.

[0053] In S1, the object of the dynamic evaluation model of the market power of power generating units is the generating units that declare quantity-price curves in the market and participate in transactions.

[0054] The basic parameters of the unit include unit type, unit rated capacity, unit minimum technical output, unit status (available status, debugging status, unavailable status) and unit maximum climbing power;

[0055] The unit transaction behavior data includes the unit's market participation quantity and price declaration data and unit clearance data;

[0056] Market settlement data includes market supply and demand data and market settlement electricity prices;

[0057] Market management data include the deviation between the actual output of the unit and the cleared electricity output, electricity bill payment, and credit guarantee performance.

[0058] In S2, the market power evaluation index system includes quotation strategy indicators and clearing result indicators. The unit categories in the system are marked through the expert system to form a market power characteristic data set of bidding units. Among them:

[0059] The quotation strategy indicators include the market share of electricity declared by the unit, the relative level of the average quotation of the unit, the volume and price index of the high quotation of the unit, the relative level of the volume and price index of the high quotation of the unit, the ratio of the high quotation of the unit, and the relative level of the ratio of the high quotation of the unit;

[0060] The indicators of clearing results include the market share of winning electricity, the relative level of unit winning rate and the unit out-of-sequence capacity index;

[0061] The expert system consists of an expert knowledge database and an inference engine, including:

[0062] The expert knowledge database includes inference rules based on transaction behavior, inference rules based on risk assessment and inference rules based on market performance;

[0063] The inference engine analyzes the unit's behavior in participating in the market according to the inference rules in the expert knowledge base. When the unit reaches the risk threshold, it is considered that the generator unit is very likely to use market power to disrupt the market, and the unit is marked as a risky unit that abuses market power.

[0064] The market share of reported electricity volume refers to the proportion of reported electricity volume of the unit to the total reported electricity volume in the market. The calculation formula is shown in formula (1). The larger the index value, the greater the influence of the unit in the market, and the more likely it is to use its market power to manipulate the market and obtain excess profits during the transaction process.

[0065]

[0066] Where: is the market share index of the declared electricity of unit i; N is the total number of units in the market; is the declared power of unit i.

[0067] As for the relative level of the average unit quotation, the power generation side of my country's power market is currently dominated by coal-fired units. The price fluctuation of the coal market will directly affect the power generation cost and electricity price of power generation enterprises. At this time, the change of the average unit quotation is not equivalent to the unit using market power. The present invention converts the coal price into the unit quotation and proposes a relative level indicator of the average unit quotation.

[0068]

[0069] In formula (2) is the average relative price of unit i; y i,t is the profit margin of unit i participating in the market at time t; is the mean profit margin of unit i during the statistical period. i,i The calculation formula is:

[0070]

[0071] In formula (3), is the average bid of unit i at time t, C i is the unit fuel cost of unit i (yuan / MWh); k i is the fuel cost proportion of coal-fired unit i, C i / k i is the power generation cost of unit i; g i is the coal consumption of power generation of unit i (g / kWh); P c is the price of coal in the market; w is the calorific value of primary energy; the constant v is the value-added tax rate for the purchase of primary energy by the power plant.

[0072] When the coal market price rises, the power generation cost of the unit will increase accordingly, and the added value is:

[0073]

[0074] As shown in formula (4), when the coal market price rises ΔP c When the power generation cost of the unit is i / k i Will increase ΔP i If the unit quotes normally, its average quote It will also increase ΔP as the cost of power generation changes i , at this time y i,t remain unchanged, Still close to 0; if the unit has the behavior of raising the market quotation, the unit's quotation Will increase ΔP′, ΔP′ includes not only ΔP i , including the ΔP that the unit raises the market electricity price h At this time, the average unit quotation is relatively level will be far away from 0. Therefore, the relative level index of the average quotation of the unit The judgment of whether the unit has used market power is not restricted by the price changes of the primary energy market, and when When it is far away from 0, it indicates that the unit has the behavior of mobilizing market power.

[0075] As for the unit high bid quantity and price index, since most of the power generation units in the domestic power market currently participate in the power market by adopting a multi-stage bid and quantity reporting method, the multi-stage bid and quantity reporting of multiple units leads to a large amount of field data, so the market operation agency cannot directly discover the unit's high bid behavior from the unit's bid data. Therefore, the present invention proposes a high bid quantity and price index based on the quantity and price index to reflect the degree of the unit's high price reporting behavior.

[0076] The content of the relative level of the unit high bid quantity-price index provided by the present invention is that when the price of primary energy rises, it is reasonable for power generation enterprises to appropriately increase their bids. Therefore, although the unit high bid quantity-price number can reflect whether the unit is suspected of raising market prices, it cannot directly confirm whether the unit is in violation of regulations. In this regard, the present invention innovatively proposes a relative level indicator of the unit high bid quantity-price index, which can more accurately reflect the high bidding behavior of the unit and help market operating institutions further determine whether the unit has mobilized market power. The calculation formula of this indicator is:

[0077]

[0078] In the formula, and L i,t,h are the high bid quantity-price index and the relative level index of the high bid quantity-price index of unit i at time t; N is the number of all bidding units in the market. The farther the relative level of the unit's high bid quantity-price index is from 0, the higher the probability that the unit will use market power to make a profit.

[0079] The unit high bid ratio is different from the high bid volume price index. The high bid ratio analyzes whether the generator unit has used market power from the perspective of the unit's declared capacity. The high bid ratio is defined as the ratio of the unit's declared capacity in the high bid section to the unit's registered capacity when it enters the market. The calculation formula is shown in formula (6). The larger the unit's high bid ratio, the stronger its speculative behavior is, and the more harmful it is to the market operation.

[0080]

[0081] In the formula, R i,t,h is the high bid ratio of unit i at time t; is the declared capacity of the high bid segment in the declared curve of unit i at time t; is the registered capacity of the unit. When the market quotes full power, q i,t Declare capacity for the last section of the unit.

[0082] Regarding the relative level of the unit's high bid ratio: when the market price of primary energy increases, the unit's high bid ratio will also increase while the unit reasonably increases its bid, which will cause the market operation agency to misjudge the compliant unit. For this reason, the present invention innovatively proposes a high bid ratio relative level indicator, and the calculation formula is shown in formula (7). When the relative level of the unit's high bid ratio is far away from 0, it indicates that the unit's high bid ratio is far away from the average high bid ratio of all units in the market, and the unit has a large degree of abnormal bidding.

[0083]

[0084] In the formula, is the relative level of the high bid ratio of unit i at time t; R i,t,h is the high bid ratio of unit i at time t; N is the number of all bidding units in the market.

[0085] For the winning electricity market share: the calculation formula for the winning electricity market share of the unit is shown in formula (8). When the clearing price increases, the unit with a larger winning electricity market share index value has a greater influence on the market, and the greater the possibility of causing the market clearing price to increase through its own behavior during the transaction process.

[0086]

[0087] In the formula, is the market share index of the winning electricity volume of unit i; N is the total number of units in the market; is the winning bid power of unit i.

[0088] Regarding the relative level of the unit's bid winning rate: In an ideal power market, the bid curves of most power generating units are close to their cost electricity prices. When the load demand and the unit's bid bidding strategy remain unchanged, the unit's bid winning rate will decrease as the number of market players increases. At this time, the change in the unit's bid winning rate is not equivalent to the unit's use of market power. For this reason, the method of the present invention proposes a relative level indicator of the unit's bid winning rate. When the number of market players N increases, the average bid winning rate of all units in the market will also decrease. At this time, the number of market players increases but the relative level of the unit's bid winning rate W i,t Therefore, the relative level indicator of the unit's bid winning rate can help market operators more accurately analyze whether the unit has used market power through the unit's bid winning situation, which is of great value in determining whether the unit has violated regulations.

[0089]

[0090] In formula (9), and W i,t are the relative level and index of the bid-winning rate of unit i at time t; N is the number of all bidding units in the market; in formula (10), and are the winning bid value and declared value of the electricity of unit i at time t. When the winning bid rate of all units is close to 0, the market competition situation is good. When the winning bid rate of the unit is far from 0, it indicates that the abnormality of the unit is higher.

[0091] For the unit out-of-sequence capacity index: From the perspective of unit capacity, the unit out-of-sequence capacity index describes the level of the unit's non-generated capacity in the market. The larger the index value, the more likely the unit is to have physical retention behavior, and the greater the suspicion that the unit uses market power to make profits.

[0092] In S3, the dynamic evaluation model of market power of power generators based on grid search and decision tree algorithm specifically includes the following processes:

[0093] S31: Based on the market power evaluation index system and the historical transaction data of the power market, the computer group index values ​​are calculated to form a complete sample, and the units that use market power in the set are marked through the expert system to form a market power characteristic data set of the bidding units;

[0094] S32: Find the hyperparameters that make the decision tree classification effect the best through grid search and cross validation, and establish the unit market power evaluation model based on the optimal hyperparameters;

[0095] The specific steps to build a decision tree evaluation model are as follows:

[0096] a) Determine the hyperparameters and value ranges that need to be adjusted for the model based on the size of the market and the number of indicators, and preset evaluation indicators as the evaluation criteria for the decision tree; the decision tree model hyperparameters include criterion, max_depth, and splitter. The criterion parameter is the method for selecting the optimal indicator, including the Gini coefficient (gini) and information gain (gain); max_depth represents the maximum depth of the decision tree, and a reasonable tree depth can avoid overfitting of the decision tree; the splitter parameter determines the strategy for dividing nodes in the decision tree.

[0097] b) Construct hyperparameter combinations based on the hyperparameter adjustment range.

[0098] c) Divide the training set data set into k disjoint subsets.

[0099] d) Select a set of decision tree hyperparameters, use any k-1 subsets in the training set to train the decision tree, and calculate the evaluation index of the decision tree based on the remaining subset. After k iterations, the evaluation index E of k decision trees under this hyperparameter combination can be obtained. i , and finally take its average value as the evaluation value of the decision tree under this hyperparameter combination;

[0100] e) Traversing all hyperparameter combinations, selecting a set of parameters with the highest decision tree evaluation value in step d) to construct a decision tree, and obtaining the unit market power evaluation model with the best classification effect.

[0101] In S4, the optimal power generation unit market power decision tree classification model is visualized. The indicator with the largest information gain will be used as the root node of the decision tree, and the remaining indicators with decreasing information gain will be used as child nodes of the decision tree. When evaluating whether a unit is suspected of market power, first determine whether the root node indicator value of the unit in the decision tree exceeds the limit. If it exceeds the limit, the unit can be directly judged as a non-compliant unit. Otherwise, it is necessary to combine other indicators to further determine the violation of the unit, and so on, until the unit is classified.

[0102] In the next clearing period, the exercise of the unit's market power is evaluated through the constructed decision tree evaluation model. Each unit traverses the decision tree evaluation model to complete the evaluation of the unit's market power, and a new bidding unit market power feature database is formed. The unit's market power evaluation model is corrected to improve the accuracy of the unit's market power evaluation in subsequent periods, thereby realizing closed-loop management of the unit's market power evaluation.

[0103] In summary, the method of the present invention proposes a set of evaluation index systems from two aspects: unit quotation strategy and market clearing results, to help market operators establish a market power evaluation perspective; based on grid search and decision tree algorithms, the strong relationship characteristics between unit index values ​​and potential results are discovered and a market power evaluation model is established to complete the evaluation of unit market power. The dynamic evaluation method of the market power of the power generating unit of the present invention can also learn from labeled unit transaction data sets and adapt itself as the market power behavior of the unit changes; it realizes closed-loop management of market power evaluation, can help market operators quickly find illegal units, and provides a good basis for market operators to effectively identify market power in the power market, which is conducive to improving the level of risk management and control in the power market and has promotion and application value.

[0104] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for dynamic evaluation of market power of power generating units based on grid search and decision tree, characterized in that: include: 1) Collect bidding unit data including basic unit parameters, unit transaction behavior data, market settlement data and market management data through the power market technical support system; 2) Based on the various data collected in step 1) and the power market power risk monitoring indicator system, the indicator values ​​of the computer groups are calculated to form an indicator value vector describing the market behavior characteristics of the units, and the types of units in the set are labeled using an expert system to construct a market power characteristic data set of the bidding units; 3) Based on the market power characteristic data set of the bidding unit in step 2), the optimal parameters of the unit market power evaluation model are found through grid search and cross-validation algorithms, and a unit market power evaluation decision tree is created according to the optimal parameters; 4) In the next clearing period, the market power of the generating units is evaluated through the constructed decision tree evaluation model to form a new market power characteristic database of the bidding generating units, which is used to correct the market power evaluation model of the generating units; In step 1), the basic parameters of the unit include the unit type, the rated capacity of the unit, the minimum technical output of the unit, the unit status and the maximum climbing power of the unit; the transaction behavior data of the unit include the quantity and price declaration data of the unit participating in the market and the unit clearing data; the market settlement data include the market supply and demand data and the market settlement electricity price; the market management data include the deviation between the actual output of the unit and the clearing electricity, the payment of electricity fees and the performance of the credit guarantee letter; Step 3) specifically includes the following steps: 31) Based on the power market power risk monitoring indicator system and the historical transaction data of the power market, the computer group indicator values ​​are calculated to form a complete sample, and the units that use market power in the set are marked through the expert system to form a market power characteristic data set for the bidding units; 32) Construct a decision tree evaluation model, find the hyperparameters that make the decision tree classification effect the best through grid search and cross-validation, and establish a unit market power evaluation model based on the optimal hyperparameters; The specific steps of building a decision tree evaluation model include: a) Determine the hyperparameters and value ranges that need to be adjusted for the model based on the size of the market and the number of indicators, and preset evaluation indicators as evaluation criteria for the decision tree; the hyperparameters of the decision tree model include the method for selecting the optimal indicator, the maximum depth of the decision tree, and the strategy for dividing nodes in the decision tree; b) constructing a hyperparameter combination according to the hyperparameter adjustment range; c) Divide the training set data set into k disjoint subsets; d) Select a set of decision tree hyperparameters, use any k-1 subsets in the training set to train the decision tree, and calculate the evaluation index of the decision tree based on the remaining subset. After k iterations, the evaluation index E of k decision trees under this hyperparameter combination can be obtained. i , and finally take its average value as the evaluation value of the decision tree under this hyperparameter combination; e) Traversing all hyperparameter combinations, selecting a set of parameters with the highest decision tree evaluation value in step d) to construct a decision tree, and obtaining the unit market power evaluation model with the best classification effect.

2. The method for dynamic evaluation of market power of power generating units based on grid search and decision tree according to claim 1, characterized in that: In step 2), the power market power risk monitoring index system includes bidding strategy indicators and clearing result indicators. The unit categories in the set are marked by an expert system to form a bidding unit market power feature data set, where: The bidding strategy indicators include the unit's declared electricity market share, the unit's average bidding relative level indicator, the unit's high bidding quantity and price index, the unit's high bidding quantity and price index relative level, the unit's high bidding ratio and the unit's high bidding ratio relative level indicator; The clearing result indicators include the market share of the winning electricity volume, the relative level of the unit's winning rate, and the unit's out-of-sequence capacity index; The expert system consists of an expert knowledge database and an inference engine, wherein: The expert knowledge database includes reasoning rules based on transaction behavior, reasoning rules based on risk assessment, and reasoning rules based on market performance; The inference engine analyzes the unit's behavior in participating in the market according to the inference rules in the expert knowledge base. When the unit reaches the risk threshold, it is determined that the generator unit is very likely to use market power to disrupt the market, and the unit is marked as a risky unit that uses market power to make a profit.

3. The method for dynamic evaluation of market power of power generating units based on grid search and decision tree according to claim 2, characterized in that: The expression of the relative level index of the average quotation of the unit is: In the formula, is the average relative price of unit i; y i,t is the profit margin of unit i participating in the market at time t; is the mean profit margin of unit i during the statistical period.

4. The method for dynamic evaluation of market power of power generating units based on grid search and decision tree according to claim 2, characterized in that: The unit high bid quantity price index is a relative level index of the unit high bid quantity price index, and its expression is: In the formula, and L i,t,h are the high bid quantity-price index and the relative level index of the high bid quantity-price index of unit i at time t, and N is the number of all bidding units in the market.

5. The method for dynamic evaluation of market power of power generating units based on grid search and decision tree according to claim 2, characterized in that: The expression of the relative level index of the unit high bid ratio is: In the formula, is the relative level of the high bid ratio of unit i at time t; R i,t,h is the high bid ratio of unit i at time t; N is the number of all bidding units in the market.

6. The method for dynamic evaluation of market power of power generating units based on grid search and decision tree according to claim 1, characterized in that: The method for selecting the optimal index includes two parameters: the Gini coefficient and the information gain.

7. The method for dynamic evaluation of market power of power generating units based on grid search and decision tree according to claim 1, characterized in that: The specific content of step 4) is: In the next clearing period, the exercise of the unit's market power is evaluated through the constructed decision tree evaluation model. Each unit traverses the decision tree evaluation model to complete the evaluation of the unit's market power and form a new bidding unit market power feature database to correct the unit's market power evaluation model and achieve closed-loop management of the unit's market power evaluation.

Images