A virtual power plant hierarchical trading method and system based on multi-task learning

Abstract

The present invention relates to the field of virtual power plant technology, and more specifically to a multi-task learning-based hierarchical trading method and system for virtual power plants, comprising the following steps: obtaining transaction data for settlement between a virtual power plant and multiple trading markets; utilizing a neural network to separately arrange transaction tasks for settlement between the virtual power plant and the multiple trading markets based on the transaction data; and performing multi-task learning on each transaction task to obtain a joint transaction task for hierarchical settlement between the virtual power plant and the multiple trading markets. The present invention achieves balanced hierarchical transactions when a virtual power plant conducts transactions and settlements with each trading market, preventing transactions from being dominated by a single market, and ultimately balancing resource allocation among virtual power plants.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual power plants, and in particular to a multi-task learning-based virtual power plant hierarchical trading method and system. Background Art

[0002] A virtual power plant (VPP) is a power coordination and management system that leverages advanced information and communication technologies and software systems to aggregate and coordinate DER (Decentralized Generators), such as DG, energy storage systems, controllable loads, and electric vehicles, to participate in electricity market and grid operations as a single power plant. The core concepts of a VPP can be summarized as "communication" and "aggregation." Key VPP technologies include coordinated control, smart metering, and information and communication.

[0003] At present, in order to promote the enthusiasm of virtual power plant operators, increase the level of new energy consumption, and form a stable coordinated interaction capability between virtual power plants and power grids, tiered trading of virtual power plants is carried out, which greatly optimizes the resource allocation between virtual power plants.

[0004] Existing technologies mostly consider the cost structure of virtual power plants, the characteristics of their internal resources, or the extent of clean energy consumption, but rarely consider the settlement and profit distribution plans for virtual power plants. However, virtual power plants can participate in electricity markets, natural gas markets, carbon trading markets, and peak-shaving markets as a whole. However, when conducting transactions and settlements with each market, it is difficult to achieve balanced tiered trading. This can easily lead to transactions being dominated by a single market, ultimately affecting the effectiveness of tiered trading and causing imbalanced resource allocation among virtual power plants. Summary of the Invention

[0005] The purpose of the present invention is to provide a virtual power plant hierarchical trading method based on multi-task learning, so as to solve the technical problem in the existing technology that it is difficult to achieve balanced hierarchical transactions when virtual power plants conduct transaction settlements with various trading markets, which easily leads to transactions being dominated by a certain market, ultimately affecting the effect of hierarchical transactions and causing unbalanced resource allocation among virtual power plants.

[0006] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0007] A multi-task learning-based virtual power plant hierarchical trading method includes the following steps:

[0008] Obtain transaction data for virtual power plants to settle with multiple trading markets;

[0009] Based on the transaction data, using a neural network, respectively arrange transaction tasks for settlement between the virtual power plant and the multiple trading markets;

[0010] Multi-task learning is performed on each trading task to obtain a joint trading task for hierarchical settlement between the virtual power plant and multiple trading markets.

[0011] As a preferred solution of the present invention, the trading market includes: an electric energy market, a natural gas market, a carbon trading market, and a peak-shaving market.

[0012] As a preferred solution of the present invention, the transaction data of the electric energy market includes the scale of aggregatable resources, the total load scale, and the electricity clearing price of the electric energy market;

[0013] The transaction data of the peak-shaving market includes the peak-shaving declared quantity and electricity clearing price generated by the peak-shaving demand in the peak-shaving market;

[0014] The transaction data of the carbon trading market includes the carbon trading volume and carbon clearing price of the carbon trading market;

[0015] The transaction data of the natural gas market include natural gas trading volume and natural gas clearing price.

[0016] As a preferred solution of the present invention, the method for constructing the transaction task includes:

[0017] The mathematical relationship between the transaction data of the electric energy market and the income of the virtual power plant in the electric energy market is encapsulated into the BP neural network, and the electric energy market transaction task of obtaining the income of the virtual power plant in the electric energy market according to the transaction data of the electric energy market is obtained. The electric energy market transaction task is:

[0018] M1 = BP (data1), where M1 is the revenue of the virtual power plant in the electric energy market, data1 is the transaction data of the electric energy market, and BP is the BP neural network;

[0019] The mathematical relationship between the transaction data of the natural gas market and the revenue of the virtual power plant in the natural gas market is encapsulated into the BP neural network, and a natural gas market transaction task is obtained to obtain the revenue of the virtual power plant in the natural gas market according to the transaction data of the natural gas market. The natural gas market transaction task is:

[0020] M2 = BP(data2), where M2 is the revenue of the virtual power plant in the natural gas market, data2 is the transaction data of the natural gas market, and BP is the BP neural network;

[0021] The mathematical relationship between the transaction data of the carbon trading market and the income of the virtual power plant in the carbon trading market is encapsulated into the BP neural network, and a carbon trading market transaction task is obtained to obtain the income of the virtual power plant in the carbon trading market according to the transaction data of the carbon trading market. The carbon trading market transaction task is:

[0022] M3 = BP (data3), where M3 is the revenue of the virtual power plant in the carbon trading market, data3 is the transaction data of the carbon trading market, and BP is the BP neural network;

[0023] The mathematical relationship between the transaction data of the peak-shaving market and the revenue of the virtual power plant in the peak-shaving market is encapsulated into the BP neural network, and the peak-shaving market transaction task is obtained to obtain the revenue of the virtual power plant in the peak-shaving market according to the transaction data of the peak-shaving market. The peak-shaving market transaction task is:

[0024] M4=BP(data4), where M4 is the revenue of the virtual power plant in the peak-shaving market, data4 is the transaction data of the peak-shaving market, and BP is the BP neural network.

[0025] As a preferred solution of the present invention, the method for constructing the joint transaction task includes:

[0026] Set the task weight of each transaction task;

[0027] Jointly training each transaction task using the task weights to obtain the joint transaction task;

[0028] The joint transaction tasks are:

[0029]

[0030] Where A1, A2, A3 and A4 are the task weights of the electricity market trading task, natural gas market trading task, carbon trading market trading task and peak-shaving market trading task, respectively; M1 is the revenue of the virtual power plant in the electricity market; data1 is the trading data of the electricity market; M2 is the revenue of the virtual power plant in the natural gas market; data2 is the trading data of the natural gas market; M3 is the revenue of the virtual power plant in the carbon trading market; data3 is the trading data of the carbon trading market; M4 is the revenue of the virtual power plant in the peak-shaving market; and data4 is the trading data of the peak-shaving market.

[0031] As a preferred solution of the present invention, the method for setting the task weight includes:

[0032] Determine the transaction progress of each transaction task through the loss function of each transaction task;

[0033] Setting task weights using the transaction progress;

[0034] The task weights are:

[0035]

[0036] Among them, C i (t-1) = Li (t-1) / L i (t-2), i∈[1,2,3,4];

[0037] Where A i (t) is A i The task weight at the tth transaction time, C i (t-1) is A i The transaction progress at the t-1th transaction time, L i (t-1) is A i The loss function at the t-1th transaction time, L i (t-2) is A i The loss function at the t-2th transaction time, N is the total number of transaction tasks, T is the total length of the transaction time, i, j, t are all counting variables.

[0038] As a preferred solution of the present invention, the loss function is a function that quantifies the error between the predicted value of the revenue of the transaction task and the true value of the revenue.

[0039] As a preferred embodiment of the present invention, the present invention provides a virtual power plant hierarchical trading system based on multi-task learning, which is applied to the aforementioned virtual power plant hierarchical trading method based on multi-task learning. The system includes:

[0040] A data acquisition unit, used to obtain transaction data used for settlement between the virtual power plant and multiple trading markets;

[0041] a data processing unit, configured to utilize a joint transaction task to perform hierarchical settlement between the virtual power plant and multiple trading markets based on the transaction data;

[0042] The data storage unit is used to store joint transaction tasks.

[0043] As a preferred solution of the present invention, the method for constructing a joint transaction task in the data processing unit includes:

[0044] Set the task weight of each transaction task;

[0045] Jointly training each transaction task using the task weights to obtain the joint transaction task;

[0046] The joint transaction tasks are:

[0047]

[0048] Where A1, A2, A3 and A4 are the task weights of the electricity market trading task, natural gas market trading task, carbon trading market trading task and peak-shaving market trading task, respectively; M1 is the revenue of the virtual power plant in the electricity market; data1 is the trading data of the electricity market; M2 is the revenue of the virtual power plant in the natural gas market; data2 is the trading data of the natural gas market; M3 is the revenue of the virtual power plant in the carbon trading market; data3 is the trading data of the carbon trading market; M4 is the revenue of the virtual power plant in the peak-shaving market; and data4 is the trading data of the peak-shaving market.

[0049] As a preferred solution of the present invention, the method for constructing each transaction task includes:

[0050] The mathematical relationship between the transaction data of the electric energy market and the income of the virtual power plant in the electric energy market is encapsulated into the BP neural network, and the electric energy market transaction task of obtaining the income of the virtual power plant in the electric energy market according to the transaction data of the electric energy market is obtained. The electric energy market transaction task is:

[0051] M1 = BP (data1), where M1 is the revenue of the virtual power plant in the electric energy market, data1 is the transaction data of the electric energy market, and BP is the BP neural network;

[0052] The mathematical relationship between the transaction data of the natural gas market and the revenue of the virtual power plant in the natural gas market is encapsulated into the BP neural network, and a natural gas market transaction task is obtained to obtain the revenue of the virtual power plant in the natural gas market according to the transaction data of the natural gas market. The natural gas market transaction task is:

[0053] M2 = BP(data2), where M2 is the revenue of the virtual power plant in the natural gas market, data2 is the transaction data of the natural gas market, and BP is the BP neural network;

[0054] The mathematical relationship between the transaction data of the carbon trading market and the income of the virtual power plant in the carbon trading market is encapsulated into the BP neural network, and a carbon trading market transaction task is obtained to obtain the income of the virtual power plant in the carbon trading market according to the transaction data of the carbon trading market. The carbon trading market transaction task is:

[0055] M3 = BP (data3), where M3 is the revenue of the virtual power plant in the carbon trading market, data3 is the transaction data of the carbon trading market, and BP is the BP neural network;

[0056] The mathematical relationship between the transaction data of the peak-shaving market and the revenue of the virtual power plant in the peak-shaving market is encapsulated into the BP neural network, and the peak-shaving market transaction task is obtained to obtain the revenue of the virtual power plant in the peak-shaving market according to the transaction data of the peak-shaving market. The peak-shaving market transaction task is:

[0057] M4=BP(data4), where M4 is the revenue of the virtual power plant in the peak-shaving market, data4 is the transaction data of the peak-shaving market, and BP is the BP neural network.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] The present invention utilizes a neural network to separately arrange transaction tasks for settlement between a virtual power plant and multiple trading markets; and performs multi-task learning on each transaction task to obtain a joint transaction task for hierarchical settlement between the virtual power plant and multiple trading markets, thereby achieving balanced hierarchical transactions when the virtual power plant conducts transactions and settlements with each trading market, avoiding the dominance of transactions by a certain market, and ultimately balancing resource allocation among virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0061] Figure 1 A flow chart of the virtual power plant hierarchical trading method provided by an embodiment of the present invention;

[0062] Figure 2 A block diagram of the virtual power plant hierarchical trading system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, the present invention provides a virtual power plant hierarchical trading method based on multi-task learning, comprising the following steps:

[0065] Obtain transaction data for virtual power plants to settle with multiple trading markets;

[0066] Based on transaction data, a neural network is used to arrange transaction tasks for settlement between the virtual power plant and multiple trading markets.

[0067] Multi-task learning is performed on each trading task to obtain a joint trading task for hierarchical settlement between the virtual power plant and multiple trading markets.

[0068] The present invention encapsulates transactions between virtual power plants and the electricity energy market, natural gas market, carbon trading market, and peak-shaving market into independent trading tasks through neural networks, so that the profit settlement results of the virtual power plant can be obtained based on the transaction data of each trading market.

[0069] In order to balance the trading balance between virtual power plants and the electricity energy market, natural gas market, carbon trading market, and peak-shaving market, that is, to achieve balanced resource allocation of virtual power plants among the electricity energy market, natural gas market, carbon trading market, and peak-shaving market, the present invention balances the progress of various transaction tasks during the transaction process through multi-task learning to avoid the dominance of a single task.

[0070] Specifically, by monitoring the loss function of each transaction task, the progress of each transaction task can be monitored. The loss function is used to calculate the rate of decrease of the loss function. The faster the rate of decrease of the loss function, the faster the progress of the transaction task and the stronger the transaction dominance.

[0071] Therefore, the present invention sets task weights so that trading tasks with faster task progress and stronger dominance have low task weights, thereby lowering their task progress or dominance, and trading tasks with slow task progress and weak dominance have high task weights, thereby improving their task progress or dominance, thereby achieving balance in the progress of trading tasks, that is, each trading task is settled and executed with similar dominance, and ultimately achieving balance in resource allocation of virtual power plants among the electricity energy market, natural gas market, carbon trading market, and peak-shaving market.

[0072] Trading markets include: electricity market, natural gas market, carbon trading market, and peak-shaving market.

[0073] The transaction data of the electric energy market includes the scale of the energy market's aggregable resources, the total load scale, and the electricity-clearing price. The revenue of a virtual power plant in the energy market is (scale of the energy market's aggregable resources - total load scale) * electricity-clearing price.

[0074] The transaction data of the peak-shaving market includes the peak-shaving declared quantity and electricity-clearing price generated by the peak-shaving demand in the peak-shaving market. The revenue of the virtual power plant in the peak-shaving market is the peak-shaving declared quantity * electricity-clearing price.

[0075] The transaction data of the carbon trading market includes the carbon trading volume and carbon clearing price of the carbon trading market. The income of the virtual power plant in the carbon trading market is carbon trading volume * carbon clearing price;

[0076] The transaction data included in the natural gas market include the natural gas trading volume and natural gas clearing price in the natural gas market. The income of the virtual power plant in the natural gas market is the natural gas trading volume * natural gas clearing price.

[0077] The present invention encapsulates transactions between virtual power plants and the electricity energy market, natural gas market, carbon trading market, and peak-shaving market into independent transaction tasks through a neural network, so that the revenue settlement results of the virtual power plant can be obtained based on the transaction data of each trading market. The details are as follows:

[0078] The construction method of transaction tasks includes:

[0079] The mathematical relationship between the transaction data of the electric energy market and the income of the virtual power plant in the electric energy market is encapsulated into the BP neural network, and the electric energy market transaction task of obtaining the income of the virtual power plant in the electric energy market according to the transaction data of the electric energy market is obtained. The electric energy market transaction task is:

[0080] M1 = BP (data1), where M1 is the revenue of the virtual power plant in the electric energy market, data1 is the transaction data of the electric energy market, and BP is the BP neural network;

[0081] The mathematical relationship between the transaction data of the natural gas market and the revenue of the virtual power plant in the natural gas market is encapsulated into the BP neural network, and the natural gas market transaction task of obtaining the revenue of the virtual power plant in the natural gas market based on the transaction data of the natural gas market is obtained. The natural gas market transaction task is:

[0082] M2 = BP(data2), where M2 is the revenue of the virtual power plant in the natural gas market, data2 is the transaction data of the natural gas market, and BP is the BP neural network;

[0083] The mathematical relationship between the transaction data of the carbon trading market and the income of the virtual power plant in the carbon trading market is encapsulated into the BP neural network, and the carbon trading market transaction task of obtaining the income of the virtual power plant in the carbon trading market according to the transaction data of the carbon trading market is obtained. The carbon trading market transaction task is:

[0084] M3 = BP (data3), where M3 is the revenue of the virtual power plant in the carbon trading market, data3 is the transaction data of the carbon trading market, and BP is the BP neural network;

[0085] The mathematical relationship between the transaction data of the peak-shaving market and the revenue of the virtual power plant in the peak-shaving market is encapsulated into the BP neural network, and the peak-shaving market transaction task is obtained, which is to obtain the revenue of the virtual power plant in the peak-shaving market according to the transaction data of the peak-shaving market. The peak-shaving market transaction task is:

[0086] M4=BP(data4), where M4 is the revenue of the virtual power plant in the peak-shaving market, data4 is the transaction data of the peak-shaving market, and BP is the BP neural network.

[0087] BP (back propagation) neural network is a concept proposed by scientists led by Rumelhart and McClelland. It is a multi-layer feedforward neural network trained according to the error back propagation algorithm and is one of the most widely used neural network models.

[0088] BP neural networks have the ability to classify arbitrarily complex patterns and map multidimensional functions. Through self-training, they learn specific rules to produce the output value closest to the desired value for a given input, thereby determining the mapping relationship between input and output. The BP algorithm consists of two processes: forward propagation of the signal and backward propagation of the error. Specifically, the error output is calculated from input to output, while the weights and thresholds are adjusted from output to input. During forward propagation, the input signal passes through the hidden layer and acts on the output node, undergoing a nonlinear transformation to generate an output signal. If the actual output does not match the desired output, the error proceeds to the backward propagation process. Backward propagation propagates the output error layer by layer through the hidden layer to the input layer, distributing the error to all units in each layer. The error signal obtained from each layer serves as the basis for adjusting the weights of each unit. By adjusting the connection strengths between input and hidden layer nodes, the connection strengths between hidden layer nodes and output nodes, and the threshold, the error is reduced along the gradient. Through repeated training, the network parameters (weights and thresholds) that minimize the error are determined, and training ceases. At this point, the trained neural network can process the input information of similar samples and obtain the output information by itself after nonlinear transformation with the minimum output error.

[0089] By utilizing the function mapping properties of the BP neural network, the present invention applies the BP neural network to the transaction settlement process, trains the BP neural network to learn the mapping relationship between transaction data and revenue, and thereby obtains the revenue of the virtual power plant in the electric energy market based on the transaction data of the electric energy market, and puts it into use in various transaction tasks.

[0090] Multi-task learning (MTL) is a promising field in machine learning. Its goal is to leverage the useful information contained in multiple learning tasks to help develop more accurate learners (or deep learning algorithms, neural networks) for each task. By leveraging the correlations between all tasks, both experimentally and theoretically, it has been found that jointly learning multiple tasks can achieve better performance than learning them individually.

[0091] In multi-task learning, information is shared between tasks, and knowledge is transferred between different tasks. Multi-task learning methods improve overall learning performance through multi-task information sharing, which is particularly effective for learning with small samples. Assuming there are a large number of small-sample learning tasks, multi-task learning methods can fully utilize the information from multiple small samples to improve the overall learning performance of multiple tasks.

[0092] By utilizing the performance-enhancing characteristics of multi-task learning, the present invention conducts multi-task learning on multiple trading tasks. Joint learning enables each trading task to achieve better performance, balances the progress of each trading task during the trading process, avoids the dominance of a single task, and ultimately achieves trading balance between the virtual power plant and the electricity energy market, natural gas market, carbon trading market, and peak-shaving market.

[0093] In order to balance the transactions between virtual power plants and the electricity market, natural gas market, carbon trading market, and peak-shaving market, that is, to achieve balanced resource allocation among virtual power plants in the electricity market, natural gas market, carbon trading market, and peak-shaving market, the present invention uses multi-task learning to balance the progress of each transaction task during the transaction process and avoid the dominance of a single task. The details are as follows:

[0094] The construction method of the joint transaction task includes:

[0095] Set the task weight of each transaction task;

[0096] Jointly train each transaction task using task weights to obtain a joint transaction task;

[0097] The joint transaction tasks are:

[0098]

[0099] Where A1, A2, A3 and A4 are the task weights of the electricity market trading task, natural gas market trading task, carbon trading market trading task and peak-shaving market trading task, respectively; M1 is the revenue of the virtual power plant in the electricity market; data1 is the trading data of the electricity market; M2 is the revenue of the virtual power plant in the natural gas market; data2 is the trading data of the natural gas market; M3 is the revenue of the virtual power plant in the carbon trading market; data3 is the trading data of the carbon trading market; M4 is the revenue of the virtual power plant in the peak-shaving market; and data4 is the trading data of the peak-shaving market.

[0100] The methods for setting task weights include:

[0101] Determine the transaction progress of each transaction task through the loss function of each transaction task;

[0102] Use transaction progress to set task weights;

[0103] The task weights are:

[0104]

[0105] Among them, C i (t-1) = L i (t-1) / L i (t-2), i∈[1,2,3,4];

[0106] Where A i (t) is A i The task weight at the tth transaction time, C i (t-1) is A i The transaction progress at the t-1th transaction time, L i (t-1) is A i The loss function at the t-1th transaction time, L i (t-2) is A i The loss function at the t-2th transaction time, N is the total number of transaction tasks, T is the total length of the transaction time, i, j, t are all counting variables.

[0107] In the present invention, joint training is carried out through various trading tasks, and the task weights are adjusted during the training process to balance the task progress, ensure the trading balance between the virtual power plant and the electric energy market, natural gas market, carbon trading market, and peak-shaving market, and avoid the generation of task dominance. For example, when the task weights of the electric energy market trading task, the natural gas market trading task, the carbon trading market trading task, and the peak-shaving market trading task appear, the task weight of the electric energy market trading task is 0.6, the task weight of the natural gas market trading task is 0.2, the task weight of the carbon trading market trading task is 0.1, and the task weight of the peak-shaving market trading task is 0.1. (The values ​​here are used as examples, the actual values ​​are (Calculations based on actual scenarios shall prevail). At this time, the electric energy market trading tasks are dominant, which will cause the virtual power plant to focus on trading in the electric energy market and ignore the other three trading markets, resulting in a trading imbalance. The joint trading tasks will actively reduce the task weights of the electric energy trading market trading tasks in the subsequent trading process, and then increase the task weights of the other three trading markets, thereby reducing the dominance of the electric energy trading market and shifting the virtual power plant's trading attention to the other three trading markets, thereby balancing the virtual power plant's resource allocation between the electric energy market, natural gas market, carbon trading market, and peak-shaving market, balancing the progress of each trading task during the trading process, and avoiding the dominance of a single task. The natural gas market, carbon trading market, and peak-shaving market have high task weights and are dominant, similar to the electric energy market, so they will not be elaborated here.

[0108] The present invention uses the rate of decrease of the loss function as the transaction progress. By monitoring the loss function of each transaction task, the progress of each transaction task can be monitored. The rate of decrease of the loss function is calculated through the loss function. The faster the rate of decrease of the loss function, the faster the progress of the transaction task and the stronger the transaction dominance.

[0109] Therefore, the present invention sets task weights so that trading tasks with faster task progress and stronger dominance have low task weights, thereby lowering their task progress or dominance, and trading tasks with slow task progress and weak dominance have high task weights, thereby improving their task progress or dominance, thereby achieving balance in the progress of trading tasks, that is, each trading task is settled and executed with similar dominance, and ultimately achieving balance in resource allocation of virtual power plants among the electricity energy market, natural gas market, carbon trading market, and peak-shaving market.

[0110] The loss function is a function that quantifies the error between the predicted value of the profit of the trading task and the true value of the profit, such as the mean square error function, L1 norm and other error quantification functions.

[0111] like Figure 2As shown, the present invention provides a virtual power plant hierarchical trading system based on multi-task learning, which is applied to a virtual power plant hierarchical trading method based on multi-task learning. The system includes:

[0112] A data acquisition unit, used to obtain transaction data used for settlement between the virtual power plant and multiple trading markets;

[0113] A data processing unit, configured to utilize joint transaction tasks to conduct hierarchical settlement between the virtual power plant and multiple trading markets based on transaction data;

[0114] The data storage unit is used to store joint transaction tasks.

[0115] As a preferred solution of the present invention, a method for constructing a joint transaction task in a data processing unit includes:

[0116] Set the task weight of each transaction task;

[0117] Jointly train each transaction task using task weights to obtain a joint transaction task;

[0118] The joint transaction tasks are:

[0119]

[0120] Where A1, A2, A3 and A4 are the task weights of the electricity market trading task, natural gas market trading task, carbon trading market trading task and peak-shaving market trading task, respectively; M1 is the revenue of the virtual power plant in the electricity market; data1 is the trading data of the electricity market; M2 is the revenue of the virtual power plant in the natural gas market; data2 is the trading data of the natural gas market; M3 is the revenue of the virtual power plant in the carbon trading market; data3 is the trading data of the carbon trading market; M4 is the revenue of the virtual power plant in the peak-shaving market; and data4 is the trading data of the peak-shaving market.

[0121] The construction methods of each transaction task include:

[0122] The mathematical relationship between the transaction data of the electric energy market and the income of the virtual power plant in the electric energy market is encapsulated into the BP neural network, and the electric energy market transaction task of obtaining the income of the virtual power plant in the electric energy market according to the transaction data of the electric energy market is obtained. The electric energy market transaction task is:

[0123] M1 = BP (data1), where M1 is the revenue of the virtual power plant in the electric energy market, data1 is the transaction data of the electric energy market, and BP is the BP neural network;

[0124] The mathematical relationship between the transaction data of the natural gas market and the revenue of the virtual power plant in the natural gas market is encapsulated into the BP neural network, and the natural gas market transaction task of obtaining the revenue of the virtual power plant in the natural gas market based on the transaction data of the natural gas market is obtained. The natural gas market transaction task is:

[0125] M2 = BP(data2), where M2 is the revenue of the virtual power plant in the natural gas market, data2 is the transaction data of the natural gas market, and BP is the BP neural network;

[0126] The mathematical relationship between the transaction data of the carbon trading market and the income of the virtual power plant in the carbon trading market is encapsulated into the BP neural network, and the carbon trading market transaction task of obtaining the income of the virtual power plant in the carbon trading market according to the transaction data of the carbon trading market is obtained. The carbon trading market transaction task is:

[0127] M3 = BP (data3), where M3 is the revenue of the virtual power plant in the carbon trading market, data3 is the transaction data of the carbon trading market, and BP is the BP neural network;

[0128] The mathematical relationship between the transaction data of the peak-shaving market and the revenue of the virtual power plant in the peak-shaving market is encapsulated into the BP neural network, and the peak-shaving market transaction task is obtained, which is to obtain the revenue of the virtual power plant in the peak-shaving market according to the transaction data of the peak-shaving market. The peak-shaving market transaction task is:

[0129] M4=BP(data4), where M4 is the revenue of the virtual power plant in the peak-shaving market, data4 is the transaction data of the peak-shaving market, and BP is the BP neural network.

[0130] The present invention utilizes a neural network to separately arrange transaction tasks for settlement between a virtual power plant and multiple trading markets; and performs multi-task learning on each transaction task to obtain a joint transaction task for hierarchical settlement between the virtual power plant and multiple trading markets, thereby achieving balanced hierarchical transactions when the virtual power plant conducts transactions and settlements with each trading market, avoiding the dominance of transactions by a certain market, and ultimately balancing resource allocation among virtual power plants.

[0131] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.

Claims

1. A virtual power plant hierarchical trading method based on multi-task learning, characterized in that: The following steps are involved: Obtain transaction data for virtual power plants to settle with multiple trading markets; Based on the transaction data, using a neural network, respectively arrange transaction tasks for settlement between the virtual power plant and the multiple trading markets; Perform multi-task learning on each transaction task to obtain a joint transaction task for hierarchical settlement between the virtual power plant and multiple trading markets; The construction method of transaction tasks includes: The mathematical relationship between the transaction data of the electric energy market and the income of the virtual power plant in the electric energy market is encapsulated into the BP neural network, and the electric energy market transaction task of obtaining the income of the virtual power plant in the electric energy market according to the transaction data of the electric energy market is obtained. The electric energy market transaction task is: M1 = BP (data1), where M1 is the revenue of the virtual power plant in the electric energy market, data1 is the transaction data of the electric energy market, and BP is the BP neural network; The mathematical relationship between the transaction data of the natural gas market and the revenue of the virtual power plant in the natural gas market is encapsulated into the BP neural network, and the natural gas market transaction task of obtaining the revenue of the virtual power plant in the natural gas market based on the transaction data of the natural gas market is obtained. The natural gas market transaction task is: M2 = BP(data2), where M2 is the revenue of the virtual power plant in the natural gas market, data2 is the transaction data of the natural gas market, and BP is the BP neural network; The mathematical relationship between the transaction data of the carbon trading market and the income of the virtual power plant in the carbon trading market is encapsulated into the BP neural network, and the carbon trading market transaction task of obtaining the income of the virtual power plant in the carbon trading market according to the transaction data of the carbon trading market is obtained. The carbon trading market transaction task is: M3 = BP (data3), where M3 is the revenue of the virtual power plant in the carbon trading market, data3 is the transaction data of the carbon trading market, and BP is the BP neural network; The mathematical relationship between the transaction data of the peak-shaving market and the revenue of the virtual power plant in the peak-shaving market is encapsulated into the BP neural network, and the peak-shaving market transaction task is obtained, which is to obtain the revenue of the virtual power plant in the peak-shaving market according to the transaction data of the peak-shaving market. The peak-shaving market transaction task is: M4 = BP (data4), where M4 is the revenue of the virtual power plant in the peak-shaving market, data4 is the transaction data of the peak-shaving market, and BP is the BP neural network; The construction method of the joint transaction task includes: Set the task weight of each transaction task; Jointly train each transaction task using task weights to obtain a joint transaction task; The joint transaction tasks are: Where A1, A2, A3 and A4 are the task weights of the electricity market trading task, natural gas market trading task, carbon trading market trading task and peak-shaving market trading task respectively; M1 is the revenue of the virtual power plant in the electricity market; data1 is the trading data of the electricity market; M2 is the revenue of the virtual power plant in the natural gas market; data2 is the trading data of the natural gas market; M3 is the revenue of the virtual power plant in the carbon trading market; data3 is the trading data of the carbon trading market; M4 is the revenue of the virtual power plant in the peak-shaving market; data4 is the trading data of the peak-shaving market; The methods for setting task weights include: Determine the transaction progress of each transaction task through the loss function of each transaction task; Use transaction progress to set task weights; The task weights are: Among them, C i (t-1) = L i (t-1) / L i (t-2), Where A i (t) is A i The task weight at the tth transaction time, C i (t-1) is A i The transaction progress at the t-1th transaction time, L i (t-1) is A i The loss function at the t-1th transaction time, L i (t-2) is A i The loss function at the t-2th transaction time, N is the total number of transaction tasks, T is the total length of the transaction time, i, j, t are all counting variables.

2. The multi-task learning-based virtual power plant hierarchical trading method according to claim 1 is characterized by: The trading markets include: electricity market, natural gas market, carbon trading market, and peak-shaving market.

3. The multi-task learning-based virtual power plant hierarchical trading method according to claim 2 is characterized by: The transaction data of the electric energy market includes the scale of aggregated resources, total load scale, and electricity clearing price in the electric energy market; The transaction data of the peak-shaving market includes the peak-shaving declared quantity and electricity clearing price generated by the peak-shaving demand in the peak-shaving market; The transaction data of the carbon trading market includes the carbon trading volume and carbon clearing price of the carbon trading market; The transaction data of the natural gas market include natural gas trading volume and natural gas clearing price.

4. The multi-task learning-based virtual power plant hierarchical trading method according to claim 1 is characterized by: The loss function is a function that quantifies the error between the predicted value of the transaction task's profit and the true value of the profit.

5. A virtual power plant hierarchical trading system based on multi-task learning, characterized by: A multi-task learning-based virtual power plant hierarchical trading method according to any one of claims 1 to 4, the system comprising: A data acquisition unit, used to obtain transaction data used for settlement between the virtual power plant and multiple trading markets; a data processing unit, configured to utilize a joint transaction task to perform hierarchical settlement between the virtual power plant and multiple trading markets based on the transaction data; The data storage unit is used to store joint transaction tasks.

6. The multi-task learning-based virtual power plant hierarchical trading system according to claim 5 is characterized by: The method for constructing a joint transaction task in the data processing unit includes: Set the task weight of each transaction task; Jointly training each transaction task using the task weights to obtain the joint transaction task; The joint transaction tasks are: Where A1, A2, A3 and A4 are the task weights of the electricity market trading task, natural gas market trading task, carbon trading market trading task and peak-shaving market trading task, respectively; M1 is the revenue of the virtual power plant in the electricity market; data1 is the trading data of the electricity market; M2 is the revenue of the virtual power plant in the natural gas market; data2 is the trading data of the natural gas market; M3 is the revenue of the virtual power plant in the carbon trading market; data3 is the trading data of the carbon trading market; M4 is the revenue of the virtual power plant in the peak-shaving market; and data4 is the trading data of the peak-shaving market.

7. The multi-task learning-based virtual power plant hierarchical trading system according to claim 6 is characterized by: The construction methods of each transaction task include: The mathematical relationship between the transaction data of the electric energy market and the income of the virtual power plant in the electric energy market is encapsulated into the BP neural network, and the electric energy market transaction task of obtaining the income of the virtual power plant in the electric energy market according to the transaction data of the electric energy market is obtained. The electric energy market transaction task is: M1 = BP (data1), where M1 is the revenue of the virtual power plant in the electric energy market, data1 is the transaction data of the electric energy market, and BP is the BP neural network; The mathematical relationship between the transaction data of the natural gas market and the revenue of the virtual power plant in the natural gas market is encapsulated into the BP neural network, and a natural gas market transaction task is obtained to obtain the revenue of the virtual power plant in the natural gas market according to the transaction data of the natural gas market. The natural gas market transaction task is: M2 = BP(data2), where M2 is the revenue of the virtual power plant in the natural gas market, data2 is the transaction data of the natural gas market, and BP is the BP neural network; The mathematical relationship between the transaction data of the carbon trading market and the income of the virtual power plant in the carbon trading market is encapsulated into the BP neural network, and a carbon trading market transaction task is obtained to obtain the income of the virtual power plant in the carbon trading market according to the transaction data of the carbon trading market. The carbon trading market transaction task is: M3 = BP (data3), where M3 is the revenue of the virtual power plant in the carbon trading market, data3 is the transaction data of the carbon trading market, and BP is the BP neural network; The mathematical relationship between the transaction data of the peak-shaving market and the revenue of the virtual power plant in the peak-shaving market is encapsulated into the BP neural network, and the peak-shaving market transaction task is obtained to obtain the revenue of the virtual power plant in the peak-shaving market according to the transaction data of the peak-shaving market. The peak-shaving market transaction task is: M4=BP(data4), where M4 is the revenue of the virtual power plant in the peak-shaving market, data4 is the transaction data of the peak-shaving market, and BP is the BP neural network.

Images