Power parameter auxiliary model decision-making method and device based on multi-agent system

Abstract

The invention relates to an electric power parameter auxiliary model decision-making method and device based on a multi-agent system. The method comprises the following steps: acquiring sample data of historical days, summarizing the sample data into a sample data set, and preprocessing the sample data set; analyzing the sample data set to obtain an adjustment proportion range; obtaining an original feature data set through a hot coding mode and expanding the original feature data set to obtain an expanded feature data set; taking the sample data set, the adjustment proportion range and the extended feature data set as input data, and training to obtain a multi-agent deep reinforcement learning model; and training the multi-agent deep reinforcement learning model for multiple times and selecting to obtain an optimal model, and calculating to obtain an optimal declaration parameter according to the optimal model. Through multi-source data fusion, comprehensive and accurate information support can be provided, a multi-agent deep reinforcement learning model is combined, internal relations between different declaration strategies and power generation benefits are sought, multi-element and multi-period power transaction decision optimization is performed, and decision optimization efficiency and final decision accuracy are improved.

Description

Technical Field

[0001] The present application relates to the technical field of multi-agent system application in power market transactions, and in particular to a method and device for power parameter auxiliary model decision-making based on a multi-agent system. Background Art

[0002] At present, according to the trading rules of the electricity spot market, when general new energy power generation enterprises participate in electricity spot trading, they must make a day-ahead declaration in accordance with the market rules. That is, market participants declare their power generation and consumption plans, quotations and other information on the operation day to the power trading agency or dispatching agency a day before (generally the day before the operation day), which provides market participants with an opportunity to plan in advance and participate in market competition, and helps to improve the declaration efficiency of the power market and the rationality of resource allocation.

[0003] At present, there are three main ways for most new energy power generation companies to apply for market declaration: the first is to directly use the original power forecast data of the wind power forecast system as the market application plan; the second is to use manual methods to adjust the original power forecast data to form a market application plan; the third is to use software systems to generate market application plans. However, the first method relies entirely on the original power forecast data, without comprehensively considering various factors such as the medium- and long-term contract situation, market supply and demand, and market assessment rules. In addition, due to the large deviation in power forecast results, its application plan does not consider comprehensive factors and lacks in-depth analysis and mining of data, resulting in insufficient comprehensiveness of the generated market application plan; the second method, although the original power forecast data has been adjusted to a certain extent, the adjustment operation is overly dependent on personal ability and experience, and there is no clear quantitative processing intermediate process. Some factors in the actual market application plan are artificially random and inefficient. The third method, although it makes use of information technology, is based on historical transaction data, uses statistical analysis methods to analyze price trends, and conducts error analysis on short-term power forecast data and actual power data. The error results are used as the basis for short-term power adjustment to generate market declaration plans. This method only uses conventional mathematical statistics methods, with simple logic, and does not fully consider the correlation between data. It lacks in-depth analysis and mining of all data, resulting in low accuracy of the generated market declaration plans, making it difficult to maximize the reduction of power generation costs and improve power generation benefits, which is not conducive to improving the declaration efficiency and result accuracy of the power market. Summary of the invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, the present application provides a method and device for power parameter auxiliary model decision-making based on a multi-agent system, which specifically adopts the following technical solutions:

[0005] A power parameter auxiliary model decision method based on a multi-agent system comprises the following steps:

[0006] Step 1: respectively obtain sample data of the power market corresponding to the tth collection time in the dth day, summarize them into a sample data set, and preprocess the sample data set; the sample data set at least includes the day-ahead price, the real-time price, the market day-ahead clearing price, the market real-time clearing price, the clearing quantity data, the short-term power data, the adjustment ratio data and the environmental parameters; wherein d=1, 2, ..., D, D is the total number of days in the sample data;

[0007] Step 2: Analyze the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range;

[0008] Step 3: Calculate the difference between the day-ahead price and the real-time price in the electricity market based on the sample data set, and process the difference between the day-ahead price and the real-time price into an original feature data set by hot coding, and then perform window sliding processing on the original feature data set to obtain an expanded feature data set;

[0009] Step 4: Using the sample data set, the adjusted ratio range, and the expanded feature data set as input data, train a multi-agent deep reinforcement learning model with settlement income and profit recovery loss as reward functions;

[0010] Step 5: Conduct multiple trainings based on the multi-agent deep reinforcement learning model, select the best multi-agent deep reinforcement learning model, obtain the optimal adjustment ratio based on the best multi-agent deep reinforcement learning model, and obtain the optimal declared parameters based on the optimal adjustment ratio and short-term power.

[0011] Optional: When obtaining the sample data of the electricity market in the step 1, the day is taken as a unit and each day is divided into a number of equally spaced collection moments, and the sample data of the corresponding moment is obtained at each collection moment.

[0012] Optionally: In step 1, a data normalization method is used to preprocess the sample data set:

[0013] First, obtain the maximum and minimum values ​​of different sample data in the sample data set, and classify the sample data according to different times of the same day to obtain the sample unit of the dth day;

[0014] Obtain the sample data difference corresponding to the tth acquisition time based on the maximum value and the minimum value of the corresponding sample data;

[0015] The sample data at the tth collection time on the dth day is obtained based on the sample unit on the dth day, and the normalized sample data corresponding to the tth collection time on the dth day is obtained according to the difference between the sample data at the tth collection time on the dth day and the sample data corresponding to the tth collection time.

[0016] Optionally, the step of analyzing the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range in step 2 includes:

[0017] The cleared electricity data corresponding to the tth collection time on the dth day in the sample data set is converted into power system output data;

[0018] Calculate the adjustment ratio of the power system output data and the short-term power data corresponding to the t-th collection time on the d-th day;

[0019] The sample data set is statistically analyzed using statistical methods to obtain the maximum and minimum values ​​of the adjustment ratio, and the maximum and minimum values ​​of the adjustment ratio are used as the adjustment ratio range;

[0020] The probability distribution of the adjustment ratio corresponding to each collection time in the historical days is calculated through the probability density curve function.

[0021] Optional: The step of processing the difference between the day-ahead price and the real-time price as feature data by hot coding in step 3 includes:

[0022] Get the day-ahead price and real-time price corresponding to the tth collection time on the dth day respectively;

[0023] Calculate the difference between the day-ahead price and the real-time price corresponding to the tth collection time on the dth day;

[0024] Compare the difference between the day-ahead price and the real-time price corresponding to the t-th collection time on the d-th day with the threshold 0 to determine the corresponding price difference direction: when the difference between the day-ahead price and the real-time price is less than 0, the price difference direction is -1; when the difference between the day-ahead price and the real-time price is equal to 0, the price difference direction is 0; when the difference between the day-ahead price and the real-time price is greater than 0, the price difference direction is 1;

[0025] The obtained price difference directions are summarized to obtain characteristic data.

[0026] Optionally, the step of performing window sliding processing on the original feature data set to expand the original feature data set to obtain an expanded feature data set in step 3 includes:

[0027] The window width W and the sliding step length ΔW of the time window are set, and the feature data of the original feature data set are arranged in chronological order to obtain the arranged original feature data set;

[0028] The starting position of the arranged original feature data set is shifted back by a sliding step size ΔW according to the window width W of the time window, and the feature data at the time window position is set to 0 to obtain new feature data.

[0029] Optional: The steps of using settlement income and profit recovery loss as reward functions when establishing the multi-agent deep reinforcement learning model in step 4 include:

[0030] Obtain the actual power generation and clearing power data and the real-time market clearing price corresponding to the tth collection time on the dth day;

[0031] Based on the cleared electricity data, a deviation range of the cleared electricity data at a corresponding moment is obtained by calculation;

[0032] Determine whether the actual power generation at the tth collection time in the dth day is within the deviation range of the cleared power data: if the actual power generation is within the deviation range of the cleared power data, set the corresponding deviation recovery loss to 0; if the actual power generation is not within the deviation range of the cleared power data, calculate the deviation recovery loss corresponding to the tth collection time;

[0033] The deviation recovery loss on the dth day is obtained based on the deviation recovery loss corresponding to different collection moments;

[0034] Calculate the settlement income at the tth collection time on the dth day based on the actual power generation, cleared power data, market day-ahead clearing price and market real-time clearing price at the tth collection time on the dth day;

[0035] The settlement income for the dth day is calculated based on the settlement income at the tth collection time on the dth day;

[0036] The reward function value on day d is obtained based on the difference between the deviation recovery loss and the settlement income on day d.

[0037] Optionally: The calculation method of the deviation recovery loss includes:

[0038]

[0039] Among them J t is the deviation recovery loss at the tth collection time; T is the total number of collection times; λ is the deviation ratio coefficient allowed by the market; Q_zhong t is the total power of the medium- and long-term contracts at the tth collection time; t is the weighted electricity price of the medium- and long-term contract at the tth collection time; t is the market day-ahead clearing price at the tth collection time; Pzr t is the weighted electricity price in the medium and long term and the day before at the tth collection time; Q_shishi t is the actual power generation at the tth collection time; t is the real-time market clearing price at the tth collection time; Q_riqian t The electricity consumption data reported at the tth collection time; Q_shortt is the short-term power data at the tth acquisition moment; α t It is the adjustment ratio data reported at the tth collection time.

[0040] Optional: The calculation method of the settlement income includes:

[0041]

[0042] Among them J jiesuan is the total settlement income; T is the total number of collection moments input into the model; t is the settlement income at the tth collection time; Q_zhong t is the total power of the medium- and long-term contracts at the tth collection time; t is the weighted electricity price of the medium- and long-term contract at the tth collection time; t is the market day-ahead clearing price at the tth collection time; Q_shishi t is the actual power generation at the tth collection time; t is the real-time market clearing price at the tth collection time; Q_riqian t The electricity consumption data reported at the tth collection time; Q_short t is the short-term power data at the tth acquisition moment; α t It is the adjustment ratio data reported at the tth collection time.

[0043] Furthermore, the present application also discloses a power parameter auxiliary model decision-making device based on a multi-agent system, the device comprising:

[0044] The sample data acquisition module is used to respectively acquire the sample data of the power market corresponding to the t-th collection time in the d-th day and summarize them into a sample data set, and preprocess the sample data set; the sample data set at least includes the day-ahead price, the real-time price, the market day-ahead clearing price, the market real-time clearing price, the clearing quantity data, the short-term power data, the adjustment ratio data and the environmental parameters; wherein d=1, 2, ..., D, D is the total number of days in the sample data;

[0045] An adjustment range determination module is used to analyze the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range;

[0046] A feature data expansion module is used to calculate the difference between the day-ahead price and the real-time price of the power market based on the sample data set, and process the difference between the day-ahead price and the real-time price into an original feature data set by hot coding, and then perform window sliding processing on the original feature data set to expand it to obtain an expanded feature data set;

[0047] The model building module is used to build a multi-agent deep reinforcement learning model with settlement income and profit recovery loss as reward functions, using the sample data set, adjustment ratio range, and expanded feature data set as input data;

[0048] The declaration strategy determination module is used to conduct multiple trainings based on the multi-agent deep reinforcement learning model, select the best multi-agent deep reinforcement learning model, obtain the optimal adjustment ratio based on the best multi-agent deep reinforcement learning model, and obtain the optimal declaration parameters based on the optimal adjustment ratio and short-term power.

[0049] Beneficial Effects

[0050] The technical solution of this application has the following beneficial effects:

[0051] The power parameter auxiliary model decision method of the present application provides comprehensive and accurate information support through multi-source data fusion technology, and combines with a multi-agent deep reinforcement learning model to seek the intrinsic connection between different declaration strategies and power generation benefits. It can optimize multi-dimensional and multi-period power trading decisions, flexibly respond to complex and changing market environments, quantify the day-ahead declaration issues, eliminate the influence of human factors, and more deeply explore the correlation between data. The auxiliary decision model constructed can determine the optimal operating strategy to improve the declaration efficiency and result accuracy of the power market. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Flow chart of the power parameter auxiliary model decision method in an embodiment of the present application.

[0053] Figure 2 4 is a structural diagram of the power parameter auxiliary model decision-making device in an embodiment of the present application.

[0054] Figure 3 Schematic diagram of the changes in adjustment ratio, short-term power and optimal declared parameters in the embodiment of the present application.

[0055] Figure 4 This is a structural diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and cannot be used to limit the scope of protection of the present application. It should be noted that the following detailed descriptions are all exemplary and are intended to provide further explanation of the present application.

[0057] Combination Figure 1 As shown, the embodiment of the present application specifically discloses a power parameter auxiliary model decision method based on a multi-agent system, which includes the following steps:

[0058] Step 1: respectively obtain sample data of the power market corresponding to the tth collection time in the dth day, summarize them into a sample data set, and preprocess the sample data set; the sample data set at least includes the day-ahead price, the real-time price, the market day-ahead clearing price, the market real-time clearing price, the clearing quantity data, the short-term power data, the adjustment ratio data and the environmental parameters; wherein d=1, 2, ..., D, D is the total number of days in the sample data;

[0059] Step 2: Analyze the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range;

[0060] Step 3: Calculate the difference between the day-ahead price and the real-time price in the electricity market based on the sample data set, and process the difference between the day-ahead price and the real-time price into an original feature data set by hot coding, and then perform window sliding processing on the original feature data set to obtain an expanded feature data set;

[0061] Step 4: Using the sample data set, the adjusted ratio range, and the expanded feature data set as input data, train a multi-agent deep reinforcement learning model with settlement income and profit recovery loss as reward functions;

[0062] Step 5: Conduct multiple trainings based on the multi-agent deep reinforcement learning model, select the best multi-agent deep reinforcement learning model, obtain the optimal adjustment ratio based on the best multi-agent deep reinforcement learning model, and obtain the optimal declaration parameters based on the optimal adjustment ratio and short-term power.

[0063] Specifically, when obtaining the sample data of the power market in step 1 of the embodiment of the present application, the day is used as a unit and each day is divided into a number of equally spaced collection moments, and the sample data of the corresponding moment is obtained at each collection moment. Preferably, this embodiment uses the day as a unit and divides the collection moments into 15-minute intervals for collection and sorting, thereby obtaining data at 96 collection moments per day.

[0064] Furthermore, since the collected original sample data are in different data dimension standards and there are some dirty data, such as inaccurate, incomplete, inconsistent or non-standard data, it is necessary to clean the original sample data. In step 1 of this embodiment, a data normalization method is used to preprocess the sample data set to avoid the deviation caused by the difference between the two levels of data affecting the generalization ability of the model, and the data is standardized to a fixed range, such as [0,1] or [-1,1], to speed up the model training speed and improve the prediction accuracy. The specific process of the above preprocessing is:

[0065] First, the maximum and minimum values ​​of different sample data in the sample data set are obtained respectively, and the sample data are classified according to different times of the same day to obtain the sample unit of the dth day; it should be noted that the obtained maximum and minimum values ​​are selected within the range of sample data belonging to the same type in the sample data set. For example, when obtaining the maximum and minimum values ​​of a certain sample data, the sample data of the same type collected in historical days can be obtained respectively, and then the maximum and minimum values ​​are filtered out according to the sample data in the historical days.

[0066] Obtain the sample data difference corresponding to the tth acquisition time based on the maximum value and the minimum value of the corresponding sample data;

[0067] The sample data at the tth collection time on the dth day is obtained based on the sample unit on the dth day, and the normalized sample data corresponding to the tth collection time on the dth day is obtained according to the difference between the sample data at the tth collection time on the dth day and the sample data corresponding to the tth collection time:

[0068]

[0069] where X' d,t is the normalized sample data at the tth collection time on the dth day; X d,t is the sample data at the tth collection time on the dth day; is the minimum value of the sample data at the tth collection time in the sample data set; It is the maximum value of the sample data at the tth collection time in the sample data set.

[0070] More specifically, in step 2 of the embodiment of the present application, the adjustment ratio data already in the sample data set is analyzed to obtain the adjustment ratio range. The adjustment ratio is an important decision-making factor in the power market, which requires market participants to comprehensively consider market price fluctuations, changes in supply and demand, their own costs and benefits, changes in policies and regulations, and other factors in the hope of achieving maximum economic benefits and stable operation of the power system. In this embodiment, the allowable range of the adjustment ratio is analyzed by the collected historical sample data, and then adjustments are made within this range. The specific steps include:

[0071] Firstly, the cleared electricity data corresponding to the tth collection time on the dth day in the sample data set is converted into the power system output data;

[0072] Then, the adjustment ratios of the power system output data and the short-term power data corresponding to the t-th collection time on the d-th day are calculated respectively:

[0073]

[0074] Where R d,t is the adjustment ratio at the tth collection time on the dth day; Power_cleand,t The power system output data at the tth collection time on the dth day; Power_short d,t is the short-term power data at the tth collection time on the dth day;

[0075] Then, the sample data set was statistically analyzed using statistical methods to obtain the maximum value R of the adjustment ratio. max and the minimum value R min The maximum and minimum values ​​of the adjustment ratio are taken as the adjustment ratio range R range =[R min ,R max ];

[0076] Finally, the probability distribution of the adjustment ratio corresponding to each collection moment in the historical days is calculated by the probability density curve function to obtain the optimal adjustment ratio for each collection moment.

[0077] In detail, in step 3 of the embodiment of the present application, the difference between the day-ahead price and the real-time price is processed as feature data by hot coding. Hot coding is a coding method for converting categorical variables into numerical variables. It represents each category as a binary vector, which can avoid the problem of numerical comparison between different categories and enable the model to better process categorical variables. The specific steps include:

[0078] First, obtain the day-ahead price and real-time price corresponding to the tth collection time on the dth day respectively;

[0079] Then, the difference between the day-ahead price and the real-time price corresponding to the tth collection time on the dth day is calculated, that is:

[0080] Price_diff d,t =Price_riqian d,t -Price_shishi d,t ;

[0081] Then, the difference between the day-ahead price and the real-time price corresponding to the t-th collection time on the d-th day is compared with the threshold 0 to determine the corresponding price difference direction: when the difference between the day-ahead price and the real-time price is less than 0, the price difference direction is -1; when the difference between the day-ahead price and the real-time price is equal to 0, the price difference direction is 0; when the difference between the day-ahead price and the real-time price is greater than 0, the price difference direction is 1. The process of determining the price difference direction is shown in the following formula:

[0082]

[0083] Where Price_diff d,t is the price difference at the tth collection time on the dth day; Price_riqian d,tis the day-ahead price at the tth collection time on the dth day; Price_shishi d,t is the real-time price at the tth collection time on the dth day; Price_hot d,t is the price difference direction at the tth collection time on the dth day;

[0084] Finally, the obtained price difference directions are summarized to obtain the corresponding characteristic data.

[0085] Furthermore, in step three of the embodiment of the present application, the original feature data set is further subjected to window sliding processing to expand the original feature data set to obtain an expanded feature data set, and the specific steps include:

[0086] First, the window width W and the sliding length ΔW of the time window are set, and the feature data of the original feature data set are arranged in chronological order to obtain the arranged original feature data set;

[0087] The starting position of the arranged original feature data set is shifted back by a sliding step size ΔW according to the window width W of the time window, and the feature data at the time window position is set to 0 to obtain new feature data.

[0088] In this embodiment, the time window length is 96 points, and the existing feature data set is subjected to hysteresis processing, which uses the time window as the length reference to shift each feature data backward by a time window length as the new feature data. For example, as shown in the figure, feature feature1 = [1, 2, 3, 4, 5, 6, ...], assuming that the time window length W is 3, then feature1 is shifted backward by W points in sequence as a whole, that is, data point 1 is moved to the position of data 4, data point 2 is moved to the position of data 5, data point 3 is moved to the position of data 6, and so on, to obtain feature2 = [0, 0, 0, 1, 2, 3, ...], where the first T data points of feature2 are filled with 0.

[0089] In more detail, the steps of establishing a multi-agent deep reinforcement learning model in step 4 of the present application using settlement income, profit recovery and loss as reward functions include:

[0090] Obtain the actual power generation and clearing power data and the real-time market clearing price corresponding to the tth collection time on the dth day;

[0091] Based on the cleared electricity data, a deviation range of the cleared electricity data at a corresponding moment is obtained by calculation;

[0092] Determine whether the actual power generation at the tth collection time in the dth day is within the deviation range of the cleared power data: if the actual power generation is within the deviation range of the cleared power data, set the corresponding deviation recovery loss to 0; if the actual power generation is not within the deviation range of the cleared power data, calculate the deviation recovery loss corresponding to the tth collection time;

[0093] For example, when Q_riqian t >(1+λ)×Q_shishi t When the actual power generation Q_shishit is lower than the clearing power data Q_riqian t The lower limit of the allowed deviation, at this time the deviation recovery loss is:

[0094] J 11，t =max(Q_riqian t -(1+λ)×Q_shishi t ,0)×max(Pzr t -P_shishi t ,0);

[0095] When Q_riqian t <(1-λ)×Q_shishi t When the actual power generation Q_shishit is higher than the power clearing data Q_riqian t The upper limit of the allowed deviation, at this time the deviation recovery loss is:

[0096] J 12，t =max(Q_riqian t -(1+λ)×Q_shishi t ,0)×max(Pzr t -P_shishi t ,0);

[0097] The deviation recovery loss on the dth day is calculated based on the deviation recovery loss corresponding to different collection moments. The calculation method of the deviation recovery loss includes:

[0098]

[0099] Among them J t is the deviation recovery loss at the tth acquisition moment. In this embodiment, J t =J 11,t +J 12,t ; T is the total number of collection moments; λ is the deviation ratio coefficient allowed by the market; Q_zhong t is the total power of the medium- and long-term contracts at the tth collection time; tis the weighted electricity price of the medium- and long-term contract at the tth collection time; t is the market day-ahead clearing price at the tth collection time; Q_riqian t Pzr is the cleared power data at the tth collection time; t is the weighted electricity price in the medium and long term and the day before at the tth collection time; Q_shishi t is the actual power generation at the tth collection time; t is the real-time market clearing price at the tth collection time; Q_riqian t The electricity consumption data reported at the tth collection time; Q_short t is the short-term power data at the tth acquisition moment; α t It is the adjustment ratio data reported at the tth collection time.

[0100] The settlement income at the tth collection time on the dth day is calculated based on the actual power generation, cleared power data, market day-ahead clearing price and market real-time clearing price at the tth collection time on the dth day:

[0101]

[0102] The settlement income for day d is calculated based on the settlement income at the tth collection time on day d:

[0103]

[0104] Among them J jiesuan is the total settlement income; T is the total number of collection moments input into the model; t is the settlement income at the tth collection time; Q_zhong t is the total power of the medium- and long-term contracts at the tth collection time; t is the weighted electricity price of the medium- and long-term contract at the tth collection time; t is the market day-ahead clearing price at the tth collection time; Q_shishi t is the actual power generation at the tth collection time; t is the real-time market clearing price at the tth collection time; Q_riqian t The electricity consumption data reported at the tth collection time; Q_short t is the short-term power data at the tth acquisition moment; α t It is the adjustment ratio data reported at the tth collection time.

[0105] Finally, the reward function value J on day d is obtained based on the difference between the deviation recovery loss and the settlement income on day d:

[0106] J=J jiesuan -J chaoe .

[0107] Furthermore, step five of the embodiment of the present application is to obtain the optimal adjustment ratio based on the multi-agent deep reinforcement learning model. This process is to pre-set the number of multi-agent deep reinforcement learning model training times, save the corresponding model after each training, and input the verification data into the model saved after each training respectively. The benefit of each model is calculated, and the model with the largest benefit is selected as the optimal model to obtain the best adjustment ratio. In this embodiment, a single training uses half a year of data as training data, the data of the last N days of the target day as the verification set, and the T-point data of a single day as the sequence length for training, and the number of training steps is set to M*T steps. A total of 200 trainings are conducted, and the model of each training is saved; then, the verification set data is imported into the saved models in turn, the verification set benefit in each model is calculated, and the model with the largest benefit is selected as the optimal model;

[0108] Finally, the data of the target day is brought into the optimal model to obtain the adjustment ratio of the final target day, such as Figure 3 As shown, multiplying it by the short-term power gives the final declared strategy.

[0109] Further, such as Figure 2 As shown, the present application also discloses a power parameter auxiliary model decision device based on a multi-agent system, the device comprising:

[0110] The sample data acquisition module is used to respectively acquire the sample data of the power market corresponding to the t-th collection time in the d-th day and summarize them into a sample data set, and preprocess the sample data set; the sample data set at least includes the day-ahead price, the real-time price, the market day-ahead clearing price, the market real-time clearing price, the clearing quantity data, the short-term power data, the adjustment ratio data and the environmental parameters; wherein d=1, 2, ..., D, D is the total number of days in the sample data;

[0111] An adjustment range determination module is used to analyze the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range;

[0112] A feature data expansion module is used to calculate the difference between the day-ahead price and the real-time price of the power market based on the sample data set, and process the difference between the day-ahead price and the real-time price into an original feature data set by hot coding, and then perform window sliding processing on the original feature data set to expand it to obtain an expanded feature data set;

[0113] The model building module is used to build a multi-agent deep reinforcement learning model with settlement income and profit recovery loss as reward functions, using the sample data set, adjustment ratio range, and expanded feature data set as input data;

[0114] The declaration strategy determination module is used to conduct multiple trainings based on the multi-agent deep reinforcement learning model, select the best multi-agent deep reinforcement learning model, obtain the optimal adjustment ratio based on the best multi-agent deep reinforcement learning model, and obtain the optimal declaration parameters based on the optimal adjustment ratio and short-term power.

[0115] The device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented by the method embodiment will not be described again here.

[0116] like Figure 4 As shown, the embodiment of the present application also provides an electronic device, including a processor and a memory, a program or instruction stored in the memory and executable on the processor, and when the program or instruction is executed by the processor, the following is achieved: Figure 1 The various processes of the method embodiment shown in the figure can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0117] The embodiment of the present application also provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by a processor, the above Figure 1 The various processes of the method embodiments described above can achieve the same technical effect, and will not be described again here to avoid repetition.

[0118] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiments described above can achieve the same technical effect, and will not be described again here to avoid repetition.

[0119] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0120] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0121] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another device, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0122] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0123] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0124] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0125] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can essentially or in other words, the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a device (which can be a terminal or a platform, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0126] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A power parameter auxiliary model decision method based on a multi-agent system, characterized in that: The steps include: Step 1: respectively obtain sample data of the power market corresponding to the tth collection time in the dth day, summarize them into a sample data set, and preprocess the sample data set; the sample data set at least includes the day-ahead price, the real-time price, the market day-ahead clearing price, the market real-time clearing price, the clearing quantity data, the short-term power data, the adjustment ratio data and the environmental parameters; wherein d=1, 2, ..., D, D is the total number of days in the sample data; Step 2: Analyze the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range; Step 3: Calculate the difference between the day-ahead price and the real-time price in the electricity market based on the sample data set, and process the difference between the day-ahead price and the real-time price into an original feature data set by hot coding, and then perform window sliding processing on the original feature data set to obtain an expanded feature data set; Step 4: Using the sample data set, the adjusted ratio range, and the expanded feature data set as input data, train a multi-agent deep reinforcement learning model with settlement income and profit recovery loss as reward functions; Step 5: Conduct multiple trainings based on the multi-agent deep reinforcement learning model, select the best multi-agent deep reinforcement learning model, obtain the optimal adjustment ratio based on the best multi-agent deep reinforcement learning model, and obtain the optimal declared parameters based on the optimal adjustment ratio and short-term power.

2. The power parameter auxiliary model decision method according to claim 1, characterized in that: When acquiring the sample data of the power market in the step 1, the day is taken as a unit and each day is divided into a number of equally spaced collection moments, and the sample data of the corresponding moment is acquired at each collection moment.

3. The power parameter auxiliary model decision method according to claim 2, characterized in that: In the step 1, a data normalization method is used to preprocess the sample data set: First, obtain the maximum and minimum values ​​of different sample data in the sample data set, and classify the sample data according to different times of the same day to obtain the sample unit of the dth day; Obtain the sample data difference corresponding to the tth acquisition time based on the maximum value and the minimum value of the corresponding sample data; The sample data at the tth collection time on the dth day is obtained based on the sample unit on the dth day, and the normalized sample data corresponding to the tth collection time on the dth day is obtained according to the difference between the sample data at the tth collection time on the dth day and the sample data corresponding to the tth collection time.

4. The power parameter auxiliary model decision method according to claim 1, characterized in that: The step of analyzing the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range in step 2 includes: The cleared electricity data corresponding to the tth collection time on the dth day in the sample data set is converted into power system output data; Calculate the adjustment ratio of the power system output data and the short-term power data corresponding to the t-th collection time on the d-th day; The sample data set is statistically analyzed using statistical methods to obtain the maximum and minimum values ​​of the adjustment ratio, and the maximum and minimum values ​​of the adjustment ratio are used as the adjustment ratio range; The probability distribution of the adjustment ratio corresponding to each collection time in the historical days is calculated through the probability density curve function.

5. The power parameter auxiliary model decision method according to claim 1, characterized in that: The step of processing the difference between the day-ahead price and the real-time price into feature data by hot coding in step 3 includes: Get the day-ahead price and real-time price corresponding to the tth collection time on the dth day respectively; Calculate the difference between the day-ahead price and the real-time price corresponding to the tth collection time on the dth day; Compare the difference between the day-ahead price and the real-time price corresponding to the t-th collection time on the d-th day with the threshold 0 to determine the corresponding price difference direction: when the difference between the day-ahead price and the real-time price is less than 0, the price difference direction is -1; when the difference between the day-ahead price and the real-time price is equal to 0, the price difference direction is 0; when the difference between the day-ahead price and the real-time price is greater than 0, the price difference direction is 1; The obtained price difference directions are summarized to obtain characteristic data.

6. The power parameter auxiliary model decision method according to claim 5, characterized in that: The step of performing window sliding processing on the original feature data set to obtain an expanded feature data set in step 3 includes: The window width W and the sliding step length ΔW of the time window are set, and the feature data of the original feature data set are arranged in chronological order to obtain the arranged original feature data set; The starting position of the arranged original feature data set is shifted back by a sliding step size ΔW according to the window width W of the time window, and the feature data at the time window position is set to 0 to obtain new feature data.

7. The power parameter auxiliary model decision method according to claim 1, characterized in that: The steps of establishing a multi-agent deep reinforcement learning model in step 4 using settlement income and profit recovery loss as reward functions include: Obtain the actual power generation and clearing power data and the real-time market clearing price corresponding to the tth collection time on the dth day; Based on the cleared electricity data, a deviation range of the cleared electricity data at a corresponding moment is obtained by calculation; Determine whether the actual power generation at the tth collection time in the dth day is within the deviation range of the cleared power data: if the actual power generation is within the deviation range of the cleared power data, set the corresponding deviation recovery loss to 0; if the actual power generation is not within the deviation range of the cleared power data, calculate the deviation recovery loss corresponding to the tth collection time; The deviation recovery loss on the dth day is obtained based on the deviation recovery loss corresponding to different collection moments; Calculate the settlement income at the tth collection time on the dth day based on the actual power generation, cleared power data, market day-ahead clearing price and market real-time clearing price at the tth collection time on the dth day; The settlement income for the dth day is calculated based on the settlement income at the tth collection time on the dth day; The reward function value on day d is obtained based on the difference between the deviation recovery loss and the settlement income on day d.

8. The electric power parameter auxiliary model decision method according to claim 7, characterized in that: The calculation method of the deviation recovery loss includes: Among them J t is the deviation recovery loss at the tth collection time; T is the total number of collection times; λ is the deviation ratio coefficient allowed by the market; Q_zhong t is the total power of the medium- and long-term contracts at the tth collection time; t is the weighted electricity price of the medium- and long-term contract at the tth collection time; t is the market day-ahead clearing price at the tth collection time; Pzr t is the weighted electricity price in the medium and long term and the day before at the tth collection time; Q_shishi t is the actual power generation at the tth collection time; t is the real-time market clearing price at the tth collection time; Q_riqian t Q_short is the electricity consumption data reported at the tth collection time; t is the short-term power data at the tth acquisition moment; α t It is the adjustment ratio data reported at the tth collection time.

9. The electric power parameter auxiliary model decision method according to claim 7, characterized in that: The calculation method of the settlement income includes: Among them J jiesuan is the total settlement income; T is the total number of collection moments input into the model; t is the settlement income at the tth collection time; Q_zhong t is the total power of the medium- and long-term contracts at the tth collection time; t is the weighted electricity price of the medium- and long-term contract at the tth collection time; t is the market day-ahead clearing price at the tth collection time; Q_shishi t is the actual power generation at the tth collection time; t is the real-time market clearing price at the tth collection time; Q_riqian t Q_short is the electricity consumption data reported at the tth collection time; t is the short-term power data at the tth acquisition moment; α t It is the adjustment ratio data reported at the tth collection time.

10. A power parameter auxiliary model decision device based on a multi-agent system, characterized in that: The device comprises: The sample data acquisition module is used to respectively acquire the sample data of the power market corresponding to the t-th collection time in the d-th day and summarize them into a sample data set, and preprocess the sample data set; the sample data set at least includes the day-ahead price, the real-time price, the market day-ahead clearing price, the market real-time clearing price, the clearing quantity data, the short-term power data, the adjustment ratio data and the environmental parameters; wherein d=1, 2, ..., D, D is the total number of days in the sample data; An adjustment range determination module is used to analyze the existing adjustment ratio data in the sample data set to obtain the adjustment ratio range; A feature data expansion module is used to calculate the difference between the day-ahead price and the real-time price of the power market based on the sample data set, and process the difference between the day-ahead price and the real-time price into an original feature data set by hot coding, and then perform window sliding processing on the original feature data set to expand it to obtain an expanded feature data set; The model building module is used to build a multi-agent deep reinforcement learning model with settlement income and profit recovery loss as reward functions, using the sample data set, adjustment ratio range, and expanded feature data set as input data; The declaration strategy determination module is used to conduct multiple trainings based on the multi-agent deep reinforcement learning model, select the best multi-agent deep reinforcement learning model, obtain the optimal adjustment ratio based on the best multi-agent deep reinforcement learning model, and obtain the optimal declaration parameters based on the optimal adjustment ratio and short-term power.

Images