CN112686693A - Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market - Google Patents
Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market Download PDFInfo
- Publication number
- CN112686693A CN112686693A CN202011555954.0A CN202011555954A CN112686693A CN 112686693 A CN112686693 A CN 112686693A CN 202011555954 A CN202011555954 A CN 202011555954A CN 112686693 A CN112686693 A CN 112686693A
- Authority
- CN
- China
- Prior art keywords
- prediction
- quotation
- power
- day
- load
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the field of power spot market power price prediction, and discloses a method, a system, equipment and a storage medium for predicting marginal power price of a power spot market, wherein the method comprises the steps of obtaining a load influence parameter of a predicted day of the power spot market, and obtaining a power load prediction curve of the predicted day according to the load influence parameter; acquiring quotation influence parameters of the forecast day of the electric power spot market, and acquiring simulated quotation curves of power generation enterprises on the forecast day according to the quotation influence parameters; and obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day. The forecasting result of the marginal electricity price of each time node in each day is obtained through the power load forecasting curve and the simulated quotation curve, the accuracy of the forecasting result is effectively improved, and the deviation between the forecasting result and the actual result is reduced.
Description
Technical Field
The invention belongs to the field of power price prediction of a power spot market, and relates to a method, a system, equipment and a storage medium for predicting marginal power price of the power spot market.
Background
The electric power marketization is a great trend of a global electric power system, domestic electric power market innovation is also promoted gradually and orderly, the electricity price is the core of the market, the fluctuation of the electricity price influences the flowing and distribution of various resources in the electric power market, and the strong economic leverage is provided, so that the accurate electricity price prediction has very important significance for each participant in the market under the electric power market environment. The method is characterized in that under the condition that market supply and demand relations are fully considered, market participants implement market force, electric power cost, electric power market system structure, social and economic situation and other electric power price influence factors, relevant historical data are analyzed and researched by using a mathematical tool, internal relation between the electric power price and the influence factors and the development and change rule of the electric power price are explored, and under the condition that certain precision and speed are met, the electric power trading price in the future electric power market is predicted. The predicted objects can be the uniform clearing price of the system, the clearing price of each region and the clearing price of each node. At present, the marginal electricity price of the current electric power market not only becomes an economic tie for connecting users (power consumers), market supervisors (power supervision committees) and power generators (power generation enterprises), but also is an important factor for the economic benefits of the power consumers and the power generators, so that all participants of the electric power market are concerned about the development and change trend of the marginal electricity price of the system, and the prediction of the marginal electricity price of the market is particularly important for all the participants in the market. For the whole power system, the marginal electricity price prediction reduces the peak-valley difference of the system by guiding the electricity utilization behavior of users and adjusting the electricity consumption and the electricity utilization time, improves the load rate of the system, reduces the operation cost of the system, ensures the operation stability of the system, and solves the problems of capacity shortage in certain specific time periods and large amount of surplus in certain time periods to a certain extent. The prediction of the marginal electricity price provides scientific basis for promoting healthy, stable and orderly competition and development of the market and formulation of various electricity price policies, along with the continuous deepening of electric power marketization, the importance of the marginal electricity price prediction is more and more prominent, and the more accurate the prediction result is, the more intelligent business decisions can be made by market participants such as electric power companies and the like in a competitive and changeable environment.
In the aspect of marginal electricity price prediction, the electricity price prediction at the present stage is mainly based on the basic ideas of price prediction of other commodities and power load prediction, and a proper prediction technology is selected according to the actual situation and the characteristics of the electricity price prediction to analyze and predict the electricity price. The methods currently available for electricity price prediction can be mainly classified into four categories: prediction methods based on time series analysis, such as time series analysis, averaging functions, etc.; a prediction method based on factor analysis, such as a linear regression method, an artificial neural network method, a fuzzy clustering and comprehensive evaluation method and the like; a prediction method based on the metrological economics, such as a three-time point model, a Holter-Winters equation, a Markov economic prediction theory and the like; and (3) a prediction method based on a combination idea, namely fusing the previous prediction ideas to perform comprehensive prediction. The basic idea of predicting the power load by using a time series method is as follows: a large amount of accurate historical data is collected, the change rule of the historical load data along with time is revealed through the similarity of future and past time sequences, a scientific model is established by adopting a simple and effective algorithm, a large amount of tests are carried out, the model is continuously perfected, and the optimal prediction result is finally obtained. The change rule and behavior of the time sequence data are described, and the influence of comprehensive factors such as trend change, seasonal change and random fluctuation is allowed to be contained in the model. The main idea of the prediction method of the linear regression method based on the factor analysis is as follows: after the model is subjected to parameter estimation and inspection through the existing historical data, if the regression model is confirmed to have practical value, the model can be used for prediction. However, the linear regression analysis method has 2 difficulties, one is that the selection of regression variables should select major factors and ignore minor factors; the second is the quantification of the variable factor. Artificial Neural Network (ANN) technology is widely used to predict variables (such as power system load prediction) affected by many complex factors, with its characteristics of self-adaptation, information memory, autonomous learning, knowledge reasoning and optimized computation for a large number of non-structural and non-precise laws. In the artificial neural network method, the most applied is a feedforward artificial neural network with a hidden layer, and the network can realize a complex nonlinear mapping relation from input to output. Therefore, several factors which influence the power load to be larger and further influence the real-time marginal electricity price to be larger are used as input, such as air temperature, wind speed, load at each time interval, peak-valley load, humidity, previous day electricity price and the like, the BP algorithm is utilized to train the network, and then electricity price prediction is carried out. The mathematical model in the field of the metrological economics is adopted to analyze and predict 96-point real-time marginal electricity prices the next day, and the method is of practical significance for improving the market competitiveness of independent power generation enterprises. The typical characteristic is a three-time model for marginal electricity price prediction, the most easily obtained data for predicting the electricity price is a historical value of the marginal electricity price, information such as power supply quantity, electricity consumption quantity and the like is hidden in the historical value of the marginal electricity price, and the electricity price guides increase and decrease of the power supply quantity and the electricity consumption quantity in each time period. Therefore, a mathematical model describing the relationship among the power supply quantity, the power consumption quantity and the electricity price can be established to predict the marginal electricity price based on the theoretical average value-macroscopic market price relationship and the relationship among the supply-price relationship and the demand-price relationship, so that the purpose of predicting the electricity price from the perspective of the relationship among the supply, the demand and the electricity price is realized, and the complex demand prediction is avoided.
Although the above prediction methods can predict the marginal electricity price in a certain period of time to some extent, the accuracy of the prediction result is often great. For example, a pure time series analysis method is taken as an example, the method is that historical data is based on characteristic similarity of a time series, power price data in external prediction time is obtained, data with the highest time similarity in the historical data is fitted to generate predicted data, but due to complexity of a marginal power price forming process, certain core influence factors in the marginal power price forming process cannot be fitted comprehensively by the method, particularly, the stability of the method is poor and large deviation is easy to occur in a quotation decision process of each power generation enterprise. In addition, a prediction method based on factor analysis, such as an electric power market quotation prediction method based on adaptive filtering, is characterized in that the method calculates the average of the discharge prices of each transaction time k in one day by selecting the discharge price data of a period of time before the electric power market in a prediction region, and fits the average with the time t to form a state transition matrix; and estimating the covariance of the observation noise, and substituting the estimated covariance of the observation noise into the power market to obtain the clear electricity price prediction model, but the accurate observation noise of the system is very difficult to obtain in practice, and the accurate prediction model is not easy to obtain. The marginal electricity price prediction method directly based on the neural network needs to be based on huge membership data, and because the quotation process of a power generation enterprise has the characteristic of game, an accurate training model is difficult to obtain, so that the accuracy of the prediction result is not high.
Disclosure of Invention
The invention aims to overcome the defect that the accuracy of a prediction result of the conventional marginal price prediction method is not high in the prior art, and provides a marginal price prediction method, a system, equipment and a storage medium for the electric power spot market.
In order to achieve the purpose, the invention adopts the following technical scheme to realize the purpose:
in a first aspect of the present invention, a method for predicting marginal electricity price of an electric power spot market includes the following steps:
acquiring load influence parameters of a forecast day of the power spot market, and acquiring a power load forecast curve of the forecast day according to the load influence parameters;
acquiring quotation influence parameters of the forecast day of the electric power spot market, and acquiring simulated quotation curves of power generation enterprises on the forecast day according to the quotation influence parameters;
and obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
The invention further improves the marginal electricity price forecasting method of the electric power spot market:
the specific method for obtaining the power load prediction curve of the prediction day according to the load influence parameters comprises the following steps: according to the load influence parameters, obtaining a power load prediction curve of a prediction day through a power load prediction model; the power load prediction model is established in the following mode:
acquiring historical best similar days of the forecast days of the electric power spot market; and constructing an initial power load prediction model based on a deep neural network, taking load data and load influence parameters of the historical best similar day as samples, training and testing the initial power load prediction model, and obtaining the power load prediction model.
The specific method for acquiring the historical best similarity date of the forecast date of the electric power spot market comprises the following steps:
acquiring Manhattan distances between load influence parameters of the forecast days of the power spot market and load influence parameters of the historical days of the power spot market; and taking the historical day with the shortest Manhattan distance as the historical best similarity day of the forecast day of the power spot market.
The specific method for obtaining the simulated quotation curves of the power generation enterprises in the forecast day according to the quotation influence parameters comprises the following steps: according to the quotation influence parameters, obtaining a simulated quotation curve of each power generation enterprise on a forecast day through a deep reinforcement learning quotation decision model of each power generation enterprise in the electric power spot market; the deep reinforcement learning quotation decision model of each power generation enterprise is constructed in the following mode:
constructing an initial deep reinforcement learning offer decision model of each power generation enterprise based on a deep reinforcement learning algorithm; and training the initial depth reinforcement learning quotation decision model of each power generation enterprise according to a preset return function of the initial depth reinforcement learning quotation decision model by using the historical data of each power generation enterprise and aiming at the minimum quotation deviation to obtain the depth reinforcement learning quotation decision model of each power generation enterprise.
The deep reinforcement learning algorithm comprises the following steps: a depth-deterministic policy gradient algorithm.
And a return function of the initial deep reinforcement learning offer decision model is obtained based on demonstration reasoning learning.
The specific method for obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day is as follows:
obtaining quotation data of each power plant at each time node of a forecast day according to a simulated quotation curve of each power generation enterprise on the forecast day, wherein the quotation data comprises a plurality of quotation groups, and each quotation group comprises electric quantity and electricity price; obtaining the power load of each time node of the forecast day according to the power load forecast curve of the forecast day; and aiming at each time node of the forecast day, superposing the electric quantity of each power plant according to the sequence of the electric quantity from low to high until the electric quantity is not less than the electric power load of each time node, and taking the electric quantity in the quotation group of the last superposed electric quantity of each time node as the forecasting result of the marginal electric quantity of each time node.
In a second aspect of the present invention, a system for predicting marginal electricity prices of electric power spot markets includes:
the power load prediction module is used for acquiring load influence parameters of the forecast day of the power spot market and obtaining a power load prediction curve of the forecast day according to the load influence parameters;
the quotation prediction module is used for acquiring quotation influence parameters of the electric power spot market prediction day and obtaining simulated quotation curves of power generation enterprises on the prediction day according to the quotation influence parameters; and
and the marginal electricity price prediction module is used for obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
In a third aspect of the present invention, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned electric power spot market marginal electricity price prediction method when executing the computer program.
In a fourth aspect of the present invention, a computer readable storage medium stores a computer program, which when executed by a processor, implements the steps of the above-mentioned electric power spot market marginal electricity price prediction method.
Compared with the prior art, the invention has the following beneficial effects:
according to the method for predicting the marginal electricity price of the electric power spot market, the load influence parameter and the quotation influence parameter of the predicted day of the electric power spot market are obtained, and then the prediction result of the marginal electricity price of each time node of the predicted day is obtained according to the electric power load prediction curve of the measured day and the simulated quotation curves of each power generation enterprise of the predicted day. Compared with a neural network prediction model directly based on marginal electricity price historical data, the method has the advantages that the accuracy and the stability of power load prediction are much higher, the deviation of an obtained power load prediction curve and an actual power load curve is small, the structure of the whole algorithm is stable, the prediction result of the marginal electricity price of each time node in each prediction day is obtained through the more accurate power load prediction curve and the simulated quotation curve of each power generation enterprise based on the actual marginal electricity price forming process, the accuracy of the prediction result is effectively improved, and the deviation of the prediction result and the actual result is reduced. Meanwhile, after an accurate marginal electricity price prediction result is obtained, the method can help the power dispatching department to better evaluate the future electricity price change trend, make a decision and deploy, so that the electricity price is kept stable, the benefit balance of the power generation end and the power utilization end is achieved, and the social benefit is maximized.
Drawings
FIG. 1 is a flowchart of a method for predicting marginal electricity price of a power spot market according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a power load prediction model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a deep reinforcement learning offer decision model according to an embodiment of the present invention;
fig. 4 is a block diagram of a system for predicting marginal electricity prices of a power spot market according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
First, some technical terms in the embodiments of the present invention are explained as follows.
Marginal electricity price: in the spot electric energy transaction, the final price quoted by the last electric energy supplier which makes the submitted electric power meet the load demand is called as the 'marginal electricity price' of the system according to the sequence of the price quoted from low to high. The power of the generator set with the quoted price higher than the marginal electricity price cannot be paid, and the bidding fails; and the generator set with the price lower than the marginal electricity price settles the current electric quantity according to the price instead of the price and the electric power market, and settles the current electric quantity according to the marginal electricity price of the system.
Electric load: the sum of the electrical power taken by the consumers of the electrical energy users to the electrical power system at a certain moment.
Deep reinforcement learning: the perception capability of deep learning and the decision capability of reinforcement learning are combined, control can be directly carried out according to input data, and the method is an artificial intelligence method which is closer to a human thinking mode.
The invention is described in further detail below with reference to the accompanying drawings:
referring to fig. 1, in an embodiment of the present invention, a method for predicting marginal electricity prices of an electricity spot market is provided, where a predicted daily electricity load prediction curve and a predicted daily simulated price quotation curve of each electricity generating enterprise are predicted according to historical load data and historical price quotation data of a power plant, and a prediction result is generated based on an actual marginal electricity price forming process according to the predicted electricity load prediction curve and the predicted simulated price quotation curves of each electricity generating enterprise, so that accuracy of the prediction result is greatly improved. Specifically, the method for predicting the marginal electricity price of the electric power spot market comprises the following steps:
s1: and acquiring load influence parameters of the forecast day of the power spot market, and acquiring a power load forecast curve of the forecast day according to the load influence parameters.
Specifically, the load impact parameters generally include: the method comprises the steps of determining load influence parameters of a forecast day of the power spot market by analyzing various parameters influencing the load, and combining all the load influence parameters to obtain a load influence characteristic vector of the forecast day. And inputting the load influence characteristic vector of the forecast day into the power load forecasting model, processing the load influence characteristic vector by the power load forecasting model, and outputting a power load forecasting curve of the forecast day.
The specific method for obtaining the power load prediction curve of the prediction day according to the load influence parameters comprises the following steps: according to the load influence parameters, obtaining a power load prediction curve of a prediction day through a power load prediction model; the power load prediction model is established in the following mode:
s101: and acquiring the historical best similar day of the forecast day of the power spot market.
Specifically, acquiring Manhattan distances between load influence parameters of the forecast days of the power spot market and load influence parameters of each historical day of the power spot market; and taking the historical day with the shortest Manhattan distance as the historical best similarity day of the forecast day of the power spot market.
First, for time series payload data { xiEvery data sample xiThere are m load influence parameters, and those load influence parameters with a larger influence proportion are considered preferentially, which can be expressed as:
xi＝(xi1，xi2，…，xim)
wherein x isijThe jth load impact parameter representing the ith sample. The determination of the influence proportion determines the time series load influence parameter which has the largest influence on the power load based on practical experience and the summary of daily data, such as the change of weekend power consumption data, the change of holiday power consumption data and the like.
The load impact parameter matrix for all samples is then:
the load influencing parameters are then normalized as follows.
Calculating mean values of load influencing parameters
And standard deviation Sk：
Then, the load impact parameters are transformed based on the mean and standard deviation as follows:
and normalizing the load influence parameters to be in a range of [0,1] through translation transformation:
wherein, x'ikIs normalized to [0,1]The load within the range affects the parameter.
The load influence parameters of the forecast days of the power spot market and the load influence parameters of each historical day are normalized according to the method, and then the Manhattan distance of the load influence parameters is calculated according to the following formula:
wherein r isijIs a load influence parameter Manhattan distance, x 'between sample i and sample j'jkIs the normalized value of the kth load influence parameter of sample j, c is the adjustment coefficient, let rijNormalized to [0,1]]Within the range.
Establishing a characteristic similarity matrix based on the load influence parameters of the prediction days and the Manhattan distance of the load influence parameters of each historical day, and obtaining a similarity relation matrix R:
and obtaining the optimal similar day through ordered arrangement based on the overall similarity, specifically, searching the historical day with the load influence parameters most similar to the load influence parameters of the forecast day in the historical day based on the similarity relation matrix, namely, the historical day with the shortest Manhattan distance between the load influence parameters of the historical day and the load influence parameters of the forecast day, and obtaining the optimal similar day in the historical day.
S102: and constructing an initial power load prediction model based on a deep neural network, taking load data and load influence parameters of the historical best similar day as samples, training and testing the initial power load prediction model, and obtaining the power load prediction model.
Specifically, referring to FIG. 2, a basic framework based on a deep neural network model is shown, where t-1, t, t +1 are time series, U, V, W are weights, X is input, O is output, and X istRepresenting the input at time t, OtRepresenting the output at the time t, establishing an initial power load prediction model based on a deep neural network on the basis of the output, taking load data and load influence parameters of the historical best similar day as samples, dividing the samples into a training set and a testing set, training the initial power load prediction model through the training set, inputting the load influence parameters of the samples in the training set into the initial power load prediction model to obtain the predicted load data of the training samples, optimizing the model parameters of the initial power load prediction model according to the error between the predicted load data and the actual load data of the training samples, and finishing the training of the initial power load prediction model.
Then, inputting the load influence parameters of the test sample in the test set into the trained initial power load prediction model to obtain the predicted load data of the test sample, and when the deviation between the predicted load data of the test sample and the actual load data is smaller than a set threshold, testing to be qualified, for example, the deviation is smaller than 10%, so as to obtain the power load prediction model; otherwise, the training set and the test set are redistributed and the training is carried out again.
By collecting a large amount of accurate historical data, according to the similarity of future and past time sequences, the change rule of the historical data over time is disclosed through the historical data, a scientific model is established by adopting a simple and effective algorithm, a large amount of tests are carried out, so that the model is continuously perfected, and the optimal prediction result is finally obtained. The change rule and behavior of the time series data are described, and comprehensive influence factors such as trend change, seasonal change and random fluctuation are allowed to be contained in the model, so that the prediction result is closer to the actual situation.
S2: and acquiring quotation influence parameters of the forecast day of the electric power spot market, and acquiring simulated quotation curves of power generation enterprises on the forecast day according to the quotation influence parameters.
Specifically, the offer impact parameters generally include: the method comprises the steps of analyzing parameters influencing quoted prices of various power generation enterprises, determining quoted price influencing parameters of a forecast day of a power spot market, combining all quoted price influencing parameters, and obtaining a quoted price influencing characteristic vector of the forecast day, wherein the generated cost (including fixed cost and variable cost), the current power and electric quantity balance information of a power grid, the highest and lowest limit prices of the power market (generally, the highest and lowest limit prices and the lowest limit prices float in a certain range on the basis of a standard price approved by a government) and the like of the power plant. Inputting the quotation influence characteristic vector of the forecast day into a deep reinforcement learning quotation decision model of each power generation enterprise in the electric power spot market, processing through the deep reinforcement learning quotation decision model of each power generation enterprise, and outputting a simulation quotation curve of each power generation enterprise of the forecast day.
The specific method for obtaining the simulated quotation curves of the power generation enterprises in the forecast day according to the quotation influence parameters comprises the following steps: according to the quotation influence parameters, obtaining a simulated quotation curve of each power generation enterprise on a forecast day through a deep reinforcement learning quotation decision model of each power generation enterprise in the electric power spot market; the deep reinforcement learning quotation decision model of each power generation enterprise in the electric power spot market is established in the following mode: constructing an initial deep reinforcement learning offer decision model of each power generation enterprise based on a deep reinforcement learning algorithm; and training the initial deep reinforcement learning quotation decision model of each power generation enterprise by using the historical data of each power generation enterprise and aiming at the minimum quotation deviation to obtain the deep reinforcement learning quotation decision model of each power generation enterprise.
Specifically, referring to fig. 3, each power generation enterprise in the electric power spot market is modeled as an agent (agent), the whole quotation decision process of the agent is modeled based on a deep reinforcement learning algorithm, the generated data is used for modifying the action strategy of the agent, the agent interacts with the environment to generate new data, the new data is used for further improving the behavior of the agent, and after several times of iterative learning, the optimal action (optimal strategy) for completing the corresponding task is finally learned.
agent represents strategy parameterization as a random strategy gradient function shown in the following formula, calculates the strategy function gradient related to the action, continuously adjusts the action and approaches to the optimal strategy:
wherein, piθIn order to be a strategy for the price quote,
q is a cost function for the gradient of the quotation strategy. And a larger Q value is obtained by adjusting the parameters of the random strategy gradient function.
Furthermore, in order to reduce the deviation between the simulated quotation strategy and the actual power plant quotation strategy, a deep reinforcement learning method is adopted to apply historical quotation data of each power plant as training data of each agent, and an initial deep reinforcement learning quotation decision model of each power generation enterprise is trained according to a preset return function of the initial deep reinforcement learning quotation decision model with the minimum quotation deviation as a target. The state transition probability matrix P of each initial deep reinforcement learning quotation decision model is obtained based on a statistical principle according to actual historical quotation data of each power plant, and is continuously adjusted and corrected through the actual quotation data in the subsequent prediction learning. Meanwhile, the collected samples are firstly put into a sample pool, and then one sample is randomly selected from the sample pool to be used for training an initial deep reinforcement learning offer decision model.
The state transition probability matrix P is used for generating future predicted quotations of each power plant based on a reinforcement learning principle, namely, transition conditions of different quotation strategies are generated based on the load demands of the power spot market under different time sequences, and selection probabilities of different quotations under different load demands are obtained by counting historical sample data.
In the embodiment, initial depth-enhanced Learning quotation decision models of each power plant are generated based on a depth certainty strategy gradient algorithm (DDPG), specifically, firstly, a Q-Learning algorithm in the depth certainty strategy gradient algorithm is used for constructing the initial depth-enhanced Learning quotation decision model, the Q value of a Q network in the initial depth-enhanced Learning quotation decision model is determined by simulating depth through a convolutional neural network, a Q network initial target value is provided for the initial depth-enhanced Learning quotation decision model through the Q-Learning algorithm and is updated, the updated basis is a time sequence difference formula, and a new strategy is obtained according to the updated Q value.
The Q-Learning algorithm is a value-based algorithm in a reinforcement Learning algorithm, Q is Q (S, a), namely in the S State (S belongs to S) at a certain moment, an Action a (a belongs to A) is taken to obtain the expectation of profit, the environment feeds back corresponding rewardr according to the Action of agent, so the main idea of the Q-Learning algorithm is to construct State and Action into a Q-table to store a Q value, and then the Action capable of obtaining the maximum profit is selected according to the Q value.
And determining the quotation scheme in each time interval according to the similar day parameters of the quotation period based on the trained quotation strategy. The quotation strategy model generates different similar day quotation strategy models based on historical similar day clustering data, prediction is carried out according to the similar day models of the prediction dates when the quotation strategy of a future power plant is formulated, and prediction accuracy is improved.
Preferably, in order to improve the accuracy of simulating the quotation data of each power plant by the agent, the return function of the initial deep reinforcement learning quotation decision model is obtained based on demonstration reasoning learning:
wherein R(s)t,at) For teaching price(s) by expertt,at) Reward for a policy, s is the current state, a is the action taken, π is the quote policy, E is the mathematical expectation.
Preferably, the obtained deep reinforcement learning quotation decision model of each power generation enterprise continuously adjusts the deep reinforcement learning quotation decision model of each power generation enterprise through the actual quotation information of each power generation enterprise in the actual power spot market in the later prediction, so that the quotation result of the deep reinforcement learning quotation decision model of each power generation enterprise and the quotation deviation of the actual power generation enterprise are minimum.
S3: and obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
Specifically, according to a simulated quotation curve of each power generation enterprise on a forecast day, obtaining quotation data of each power plant on each time node on the forecast day, wherein the quotation data comprises a plurality of quotation groups, and each quotation group comprises electric quantity and electricity price; obtaining the power load of each time node of the forecast day according to the power load forecast curve of the forecast day; and aiming at each time node of the forecast day, superposing the electric quantity of each power plant according to the sequence of the electric quantity from low to high until the electric quantity is not less than the electric power load of each time node, and taking the electric quantity in the quotation group of the last superposed electric quantity of each time node as the forecasting result of the marginal electric quantity of each time node.
In this embodiment, the prediction result of the marginal electricity price at each time node of the prediction day is obtained based on the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day. The specific process includes the steps that the power load of each time node of a forecast day is generated based on a power load forecast curve of the forecast day, the quoted price data of each power plant of each time node of the forecast day are arranged from low to high based on a simulated quoted price curve of each power generation enterprise of the forecast day, the quoted price data comprise a plurality of quoted price groups, each quoted price group comprises electric quantity and electricity price, and the quoted price groups are arranged from low to high during arrangement. Firstly, assuming that the power load of a predicted time node is Q, and arranging the quoted electricity quantity and the electricity price of each power plant from low to high according to the electricity price as follows:
(q1,v1)<(q2,v2)<…<(qm,vm)<…<(qn,vn)
wherein q ismRepresenting the electric quantity, v, of the m-th power plantmRepresents the electric quantity of the mth power plant as qmThe price of electricity in hours.
If:
and:
then, vmThe prediction result of the marginal electricity price of the electric power spot market at the time node is obtained. According to the principle, the prediction of the marginal electricity price at 96 time points on the day and each time node is completed.
In summary, compared with the traditional marginal electricity price prediction scheme, the method for predicting the marginal electricity price of the electricity spot market generates prediction data based on the actual marginal electricity price forming process through the prediction day power load prediction curve and the prediction day simulation quotation curve of each power generation enterprise, and compared with a neural network prediction model directly based on the marginal electricity price historical data, the accuracy and the stability of the power load prediction model are much higher. Through modeling of different power generation enterprises based on deep reinforcement learning, the quotation characteristics and decision processes of different power generation enterprises can be reflected more accurately, and the game process of the true quotation decision is simulated to a greater extent. Based on the power load prediction curve obtained by the method and the simulated quotation curves of various power generation enterprises, the marginal motor prediction data are generated based on the actual marginal motor forming principle, the algorithm stability is good, the accuracy of the obtained prediction result is higher, the online learning characteristic is realized, and the accuracy of the prediction result can be improved along with continuous correction of new data. After the accurate marginal electricity price prediction is obtained, the method can help the power dispatching department to better evaluate the future electricity price change trend, make decision deployment, keep the power price stable, enable the power generation end and the power utilization end to reach the benefit balance, and maximize the social benefit.
The following are embodiments of the apparatus of the present invention that may be used to perform embodiments of the method of the present invention. For details of non-careless mistakes in the embodiment of the apparatus, please refer to the embodiment of the method of the present invention.
Referring to fig. 4, in yet another embodiment of the present invention, an electric power spot market marginal electricity price prediction system is provided, which can be used to implement the electric power spot market marginal electricity price prediction method described above, and specifically, the electric power spot market marginal electricity price prediction system includes an electric power load prediction module, an offer prediction module, and a marginal electricity price prediction module.
The power load prediction module is used for acquiring load influence parameters of the forecast day of the power spot market and obtaining a power load prediction curve of the forecast day according to the load influence parameters; the quotation prediction module is used for acquiring quotation influence parameters of the electric power spot market prediction day and obtaining simulated quotation curves of power generation enterprises on the prediction day according to the quotation influence parameters; and the marginal electricity price prediction module is used for obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
In yet another embodiment of the present invention, a computer device is provided that includes a processor and a memory for storing a computer program comprising program instructions, the processor for executing the program instructions stored by the computer storage medium. The Processor may be a Central Processing Unit (CPU), or may be other general-purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable gate array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal, and is specifically adapted to load and execute one or more instructions in a computer storage medium to implement a corresponding method flow or a corresponding function; the processor provided by the embodiment of the invention can be used for the operation of the marginal electricity price prediction method of the electric power spot market, and comprises the following steps: acquiring load influence parameters of a forecast day of the power spot market, and acquiring a power load forecast curve of the forecast day according to the load influence parameters; acquiring quotation influence parameters of the forecast day of the electric power spot market, and acquiring simulated quotation curves of power generation enterprises on the forecast day according to the quotation influence parameters; and obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
In yet another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a Memory device in a computer device and is used for storing programs and data. It is understood that the computer readable storage medium herein can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also, one or more instructions, which may be one or more computer programs (including program code), are stored in the memory space and are adapted to be loaded and executed by the processor. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory.
One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the method for predicting the marginal electricity price of the electricity spot market in the embodiment; one or more instructions in the computer-readable storage medium are loaded by the processor and perform the steps of: acquiring load influence parameters of a forecast day of the power spot market, and acquiring a power load forecast curve of the forecast day according to the load influence parameters; acquiring quotation influence parameters of the forecast day of the electric power spot market, and acquiring simulated quotation curves of power generation enterprises on the forecast day according to the quotation influence parameters; and obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (10)
1. A method for predicting marginal electricity price of an electric power spot market is characterized by comprising the following steps:
acquiring load influence parameters of a forecast day of the power spot market, and acquiring a power load forecast curve of the forecast day according to the load influence parameters;
acquiring quotation influence parameters of the forecast day of the electric power spot market, and acquiring simulated quotation curves of power generation enterprises on the forecast day according to the quotation influence parameters;
and obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
2. The method for predicting the marginal electricity price of the electric power spot market according to claim 1, wherein the specific method for obtaining the electric power load prediction curve of the prediction day according to the load influence parameters is as follows: according to the load influence parameters, obtaining a power load prediction curve of a prediction day through a power load prediction model; the power load prediction model is established in the following mode:
acquiring historical best similar days of the forecast days of the electric power spot market;
and constructing an initial power load prediction model based on a deep neural network, taking load data and load influence parameters of the historical best similar day as samples, training and testing the initial power load prediction model, and obtaining the power load prediction model.
3. The method for predicting the marginal electricity price of the power spot market according to claim 2, wherein the specific method for acquiring the historical best similar day of the predicted day of the power spot market is as follows:
acquiring Manhattan distances between load influence parameters of the forecast days of the power spot market and load influence parameters of the historical days of the power spot market;
and taking the historical day with the shortest Manhattan distance as the historical best similarity day of the forecast day of the power spot market.
4. The method for predicting the marginal electricity price of the electric power spot market according to claim 1, wherein the specific method for obtaining the simulated quotation curves of the power generation enterprises in the forecast day according to the quotation influence parameters comprises the following steps: according to the quotation influence parameters, obtaining a simulated quotation curve of each power generation enterprise on a forecast day through a deep reinforcement learning quotation decision model of each power generation enterprise in the electric power spot market; the deep reinforcement learning quotation decision model of each power generation enterprise is constructed in the following mode:
constructing an initial deep reinforcement learning offer decision model of each power generation enterprise based on a deep reinforcement learning algorithm;
and training the initial depth reinforcement learning quotation decision model of each power generation enterprise according to a preset return function of the initial depth reinforcement learning quotation decision model by using the historical data of each power generation enterprise and aiming at the minimum quotation deviation to obtain the depth reinforcement learning quotation decision model of each power generation enterprise.
5. The method for predicting the marginal electricity price of the electric power spot market according to claim 4, wherein the deep reinforcement learning algorithm is as follows: a depth-deterministic policy gradient algorithm.
6. The method for predicting the marginal electricity price of the electric power spot market according to claim 5, wherein the reward function of the initial deep reinforcement learning offer decision model is obtained based on demonstration inference learning.
7. The method for predicting the marginal electricity price of the electric power spot market according to claim 1, wherein the specific method for obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the electric power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day is as follows:
obtaining quotation data of each power plant at each time node of a forecast day according to a simulated quotation curve of each power generation enterprise on the forecast day, wherein the quotation data comprises a plurality of quotation groups, and each quotation group comprises electric quantity and electricity price;
obtaining the power load of each time node of the forecast day according to the power load forecast curve of the forecast day;
and aiming at each time node of the forecast day, superposing the electric quantity of each power plant according to the sequence of the electric quantity from low to high until the electric quantity is not less than the electric power load of each time node, and taking the electric quantity in the quotation group of the last superposed electric quantity of each time node as the forecasting result of the marginal electric quantity of each time node.
8. A system for predicting marginal electricity prices of electric power spot markets is characterized by comprising:
the power load prediction module is used for acquiring load influence parameters of the forecast day of the power spot market and obtaining a power load prediction curve of the forecast day according to the load influence parameters;
the quotation prediction module is used for acquiring quotation influence parameters of the electric power spot market prediction day and obtaining simulated quotation curves of power generation enterprises on the prediction day according to the quotation influence parameters; and
and the marginal electricity price prediction module is used for obtaining the prediction result of the marginal electricity price of each time node of the prediction day according to the power load prediction curve of the prediction day and the simulated quotation curve of each power generation enterprise of the prediction day.
9. A computer arrangement comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, carries out the steps of the electric power spot market marginal electricity price prediction method according to any one of claims 1 to 7.
10. A computer-readable storage medium, in which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for predicting the marginal electricity price in a power spot market according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011555954.0A CN112686693A (en) | 2020-12-24 | 2020-12-24 | Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011555954.0A CN112686693A (en) | 2020-12-24 | 2020-12-24 | Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112686693A true CN112686693A (en) | 2021-04-20 |
Family
ID=75453059
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011555954.0A Pending CN112686693A (en) | 2020-12-24 | 2020-12-24 | Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112686693A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115498629A (en) * | 2022-09-06 | 2022-12-20 | 清华大学 | Scene enhancement type power load operation control method and device |
| CN116109014A (en) * | 2023-04-11 | 2023-05-12 | 广东广宇科技发展有限公司 | Simulation fire-fighting evacuation method for urban rail transit large transfer station |
| CN116760122A (en) * | 2023-08-21 | 2023-09-15 | 国网浙江省电力有限公司宁波供电公司 | Virtual power plant resource management and control method and device, computer equipment and storage medium |
| CN117996747A (en) * | 2024-02-05 | 2024-05-07 | 中科聚(北京)能源科技有限公司 | Electric purchasing control method under electric power spot trade market |
-
2020
- 2020-12-24 CN CN202011555954.0A patent/CN112686693A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115498629A (en) * | 2022-09-06 | 2022-12-20 | 清华大学 | Scene enhancement type power load operation control method and device |
| CN116109014A (en) * | 2023-04-11 | 2023-05-12 | 广东广宇科技发展有限公司 | Simulation fire-fighting evacuation method for urban rail transit large transfer station |
| CN116760122A (en) * | 2023-08-21 | 2023-09-15 | 国网浙江省电力有限公司宁波供电公司 | Virtual power plant resource management and control method and device, computer equipment and storage medium |
| CN116760122B (en) * | 2023-08-21 | 2023-12-26 | 国网浙江省电力有限公司宁波供电公司 | Virtual power plant resource management and control method and device, computer equipment and storage medium |
| CN117996747A (en) * | 2024-02-05 | 2024-05-07 | 中科聚(北京)能源科技有限公司 | Electric purchasing control method under electric power spot trade market |
| CN117996747B (en) * | 2024-02-05 | 2024-07-23 | 中科聚(北京)能源科技有限公司 | Electric purchasing control method under electric power spot trade market |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Toubeau et al. | Deep learning-based multivariate probabilistic forecasting for short-term scheduling in power markets | |
| Agrawal et al. | Long term load forecasting with hourly predictions based on long-short-term-memory networks | |
| Wang et al. | A seasonal GM (1, 1) model for forecasting the electricity consumption of the primary economic sectors | |
| CN112686693A (en) | Method, system, equipment and storage medium for predicting marginal electricity price of electric power spot market | |
| Gao et al. | A multiagent competitive bidding strategy in a pool-based electricity market with price-maker participants of WPPs and EV aggregators | |
| CN109919658A (en) | A kind of duty control method and system based on game theory | |
| Pinto et al. | Adaptive entropy-based learning with dynamic artificial neural network | |
| CN113983646A (en) | Air conditioner interaction end energy consumption prediction method based on generation countermeasure network and air conditioner | |
| CN111563615A (en) | Load prediction method based on feature analysis and combination learning | |
| CN113919944A (en) | Stock trading method and system based on reinforcement learning algorithm and time series model | |
| CN116862551A (en) | New energy consumption price decision method considering user privacy protection | |
| Lincoln et al. | Comparing policy gradient and value function based reinforcement learning methods in simulated electrical power trade | |
| Wang et al. | Multi-agent simulation for strategic bidding in electricity markets using reinforcement learning | |
| CN115049102A (en) | Electricity price prediction method and device, mobile terminal and storage medium | |
| Boukas et al. | Intra-day bidding strategies for storage devices using deep reinforcement learning | |
| Liu et al. | A novel electricity load forecasting based on probabilistic least absolute shrinkage and selection operator-Quantile regression neural network | |
| CN117557375A (en) | Transaction evaluation method and related device based on virtual power plant | |
| Wang et al. | Risk-Averse Optimal Combining Forecasts for Renewable Energy Trading Under CVaR Assessment of Forecast Errors | |
| Ganesh et al. | Forecasting imbalance price densities with statistical methods and neural networks | |
| CN117117878A (en) | Power grid demand side response potential evaluation and load regulation method based on artificial neural network and multi-agent reinforcement learning | |
| Chen et al. | Reinforcement learning with expert trajectory for quantitative trading | |
| Cano-Martínez et al. | Dynamic energy prices for residential users based on Deep Learning prediction models of consumption and renewable generation | |
| Shendryk et al. | Short-term Solar Power Generation Forecasting for Microgrid | |
| Watanabe et al. | Agent-based simulation model of electricity market with stochastic unit commitment | |
| Alcántara et al. | Optimal day-ahead offering strategy for large producers based on market price response learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |