A pre-training method for unit reporting agents in power markets

Abstract

The present invention proposes a pre-training method for a unit declaration agent in an electric power market. The method performs pre-training processing on the unit declaration agent, updates and optimizes the parameters in the declaration agent model of the unit declaration agent, and applies the unit declaration agent to the actual declaration link after pre-training, thereby avoiding the phenomenon of drastic fluctuations in declaration data and revenue data in the unit declaration link, and the simulation results can be directly used as data support for the unit declaration of market entities.

Description

Technical field:

[0001] The invention relates to a pre-training method for a unit declaration agent in an electric power market, and belongs to the field of electric power systems. Background technology:

[0002] The units in the power system can be divided into market units and non-market units according to whether they are connected to the power market. Market units generally set up corresponding unit declaration agents in the power market and obtain declaration data through the declaration agent model of the unit declaration agent. However, almost all existing unit declaration agents have the following two problems:

[0003] First, the existing unit declaration agent is directly applied to the actual unit declaration link. The declaration agent model of the unit declaration agent needs to optimize parameters according to the real data of the actual declaration link. This leads to abnormal fluctuations in the declaration data and income data of the unit declaration agent over a long period of time. The simulated results cannot be used as data support for actual unit declarations.

[0004] Second, the existing unit declaration agent model is trained for parameter optimization based on real data from the actual declaration process, which may lead to overfitting, and the simulated results cannot reflect the dynamic response of market players to the operation of the electricity market. Summary of the invention:

[0005] In order to solve the problems existing in the prior art, the present invention proposes a pre-training method for a unit declaration agent in the power market. The technical solution of the method is as follows:

[0006] A pre-training method for unit declaration agents in an electric power market, wherein each market unit in an electric power system has a corresponding unit declaration agent in the electric power market, and each unit declaration agent obtains declaration data through a declaration agent model, including:

[0007] Establish a data set related to node load, tie-line power and planned output of non-market units in power system operation scenarios;

[0008] The declaration agent model obtains the declaration data through the data set;

[0009] Perform clearing calculations based on the declared data to obtain the revenue data of the declared agent of each unit;

[0010] Using a genetic algorithm and based on the revenue data, the parameters in the unit declaration agent model are updated and optimized to obtain new declaration data;

[0011] The iteration of generating corresponding new income data from the new declaration data;

[0012] The iteration is terminated if the declared data and income data are the same for three consecutive iterations or the number of iterations reaches the set value.

[0013] Preferably, the declaration agent model obtains the declaration data with the goal of maximizing the unit profit.

[0014] Preferably, the data set includes a node load data set Q, a tie line power data set X, and a non-market unit planned output data set Y; wherein,

[0015] The node load data set Q: Q = {Q1, Q2, ..., Q M}, Q1, Q2, ..., Q M are the average node loads in the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; where the average node load Q m :

[0016] m=1,2,…,M;

[0017] In the formula, Q m，d is the node load in the mth period of the dth day in the operation scenario; D is the total number of days in the operation scenario;

[0018] The tie line power data set X: X = {X1, X2, ..., X M}, X1, X2, ..., X M are the tie line powers of the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; where the average tie line power X in the mth period in the operation scenario is m :

[0019] m=1,2,…,M;

[0020] Where, X m，d is the tie line power in the mth period of the dth day in the operation scenario; D is the total number of days in the operation scenario;

[0021] The non-market unit planned output data set Y: Y = {Y1, Y2, ..., Y M}, Y1, Y2, ..., Y M are the average non-market unit planned outputs in the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; among which, the average non-market unit planned output Y in the mth period in the operation scenario m :

[0022] m=1,2,…,M;

[0023] Where Y m，dis the planned output of non-market units in the mth period on the dth day in the operation scenario; D is the total number of days in the operation scenario.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention proposes a pre-training method for a unit declaration agent in an electric power market. The method performs pre-training processing on the unit declaration agent, updates and optimizes the parameters in the declaration agent model of the unit declaration agent, and applies the unit declaration agent to the actual declaration link after pre-training, thereby avoiding the phenomenon of drastic fluctuations in declaration data and revenue data in the unit declaration link, and the simulation results can be directly used as data support for municipal-level unit declaration.

[0026] The pre-training method of the present invention utilizes data sets related to node loads, interconnection line power, and planned output of non-market units in power system operation scenarios, and uses the data sets to train declaration agent model parameters of unit declaration agents. Compared with directly using real data from actual declaration links to optimize the declaration agent model parameters, the data set in the method of the present invention uses average data of each time period in the operation scenario to avoid being completely consistent with the actual operation scenario data, thereby avoiding overfitting. The simulated results can reflect the dynamic response of market entities to the operation of the power market. Description of the drawings:

[0027] Figure 1 is a flow chart of the pre-training method of the present invention;

[0028] Figure 2 It is a logic block diagram of the pre-training method of the present invention;

[0029] Figure 3 This is a screenshot of the system after 100 rounds of training using the pre-training method of the present invention;

[0030] Figure 4 This is a screenshot of the 86th unit declaration agent curve of the G1 unit trained by the pre-training method of the present invention;

[0031] Figure 5 This is a screenshot of the 87th unit declaration agent curve of the G1 unit trained by the pre-training method of the present invention;

[0032] Figure 6 This is a screenshot of the 88th unit declaration agent curve of the G1 unit trained by the pre-training method of the present invention;

[0033] Figure 7 This is a screenshot of the 89th unit declaration agent curve of the G1 unit trained by the pre-training method of the present invention. Specific implementation method:

[0034] To make the objectives, technical solutions and advantages of the present invention more clearly understood, embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0035] Embodiment 1:

[0036] This embodiment adopts a pre-training method of the present invention for unit declaration agents in the power market. Each market unit in the power system has a corresponding unit declaration agent in the power market. Each unit declaration agent obtains declaration data through a declaration agent model. In this embodiment, the declaration agent model adopts a power generation side alliance bidding strategy model based on SFAEGA. The declaration agent model is divided into an upper model and a lower model.

[0037] The upper model expression is as follows:

[0038] Assume that there are I market units bidding in the power market, among which I1 (I1 ≥ 1) units choose cooperative bidding, and I2 = I-I1 (I2 ≥ 0) units choose non-cooperative bidding. Therefore, the objective function and constraints can be constructed as follows:

[0039]

[0040] f i =ρ i q i -C i (2)

[0041]

[0042]

[0043]

[0044] Among them, fi(i∈I) is the revenue function of each unit; ρ i and q i are the node electricity price and the winning bid electricity, C i is the cost function of the i-th unit, a i 、b i 、c i is the cost coefficient; k i is the bidding strategy for the i-th unit, and is the upper and lower limits of the bid for the i-th unit; is is the quoted price of the i-th unit in the s-th quotation interval; φ i is the cost coefficient after single parameterization.

[0045] Lower model:

[0046] The node electricity price taking into account network loss is composed of fuel cost, grid congestion cost and network loss cost. According to the safety-constrained economic dispatch clearing model taking into account network loss, the electricity price of each load node can be solved, which can be specifically expressed as follows:

[0047]

[0048]

[0049] In formula (6) and formula (7), q is is the quotation quantity of the i-th unit in the s-th quotation interval; is is the quoted price of the i-th unit in the s-th quotation interval; q i is the winning bid power generation of the i-th unit; q b is the power load of load node b; p loss is the total network loss of the system; H is the network loss transfer coefficient; ψ l is the linear offset of network loss; T l is the power transfer coefficient of line l; F l is the flow into line l; and are the minimum and maximum power generation of the i-th unit respectively; F l min and F l max are the minimum and maximum power flows of line l respectively; L is the total number of line branches; B is the total number of load nodes.

[0050] The equality constraints include the winning bid power constraint of each unit, system load balance constraint, network loss constraint and line flow constraint.

[0051] like Figure 1 and Figure 2 As shown, the pre-training method in this embodiment includes the following steps:

[0052] Establish a data set related to node load, tie-line power and planned output of non-market units in power system operation scenarios;

[0053] The declaration agent model obtains declaration data by taking the data set as environmental parameters and maximizing the unit benefit;

[0054] Perform clearing calculations based on the declared data to obtain the revenue data of the declared agent of each unit;

[0055] Using a genetic algorithm and based on the revenue data, the parameters in the unit declaration agent model are updated and optimized to obtain new declaration data; the new declaration data is used to generate an iteration corresponding to the new revenue data;

[0056] The iteration is terminated if the declared data and income data are the same for three consecutive iterations or the number of iterations reaches the set value.

[0057] This method pre-trains the unit declaration agent, updates and optimizes the parameters in the declaration agent model of the unit declaration agent, and then applies the unit declaration agent to the actual declaration link after pre-training, avoiding the phenomenon of drastic fluctuations in declaration data and revenue data during the unit declaration link. The simulation results can be directly used as data support for municipal-level unit declarations.

[0058] Embodiment 2:

[0059] This embodiment is made on the basis of the first embodiment. In this embodiment, the process of establishing the data set related to the node load, the tie line power and the planned output of the non-market units is as follows:

[0060] The data sets include node load data set I, tie line power data set X, and non-market unit planned output data set Y;

[0061] By averaging the node loads in the same period of the multi-day operation scenario, the node load data of each period in the operation scenario can be obtained to form a node load data set Q: Q = {Q1, Q2, ..., Q M}, Q1, Q2, ..., Q M are the average node loads in the 1st, 2nd, …, and Mth periods in the operation scenario respectively; among which, the average node load Qm in the mth period in the operation scenario is:

[0062] m=1,2,…,M;

[0063] In the formula, Q m，d is the node load in the mth period of the dth day in the operation scenario; D is the total number of days in the operation scenario;

[0064] By averaging the tie line power of the same period in the multi-day operation scenario, the tie line power data of each period in the operation scenario can be obtained to form a tie line power data set X: X = {X1, X2, ..., X M}, X1, X2, ..., X M are the tie line powers of the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; where the average tie line power X in the mth period in the operation scenario is m :

[0065] m=1,2,…,M;

[0066] Where, X m，d is the tie line power in the mth period of the dth day in the operation scenario; D is the total number of days in the operation scenario;

[0067] By averaging the planned output of non-market units in the same period of multi-day operation scenarios, the planned output data of non-market units in each period of the operation scenario can be obtained to form a non-market unit planned output data set Y: Y = {Y1, Y2, ..., Y M}, Y1, Y2, ..., Y M are the average non-market unit planned outputs in the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; among which, the average non-market unit planned output Y in the mth period in the operation scenario m :

[0068] m=1,2,…,M;

[0069] Where Y m，d is the planned output of non-market units in the mth period on the dth day in the operation scenario; D is the total number of days in the operation scenario.

[0070] Compared with directly using real data from the actual declaration process to optimize the parameters of the declaration agent model, the above design can avoid overfitting, and the simulated results can reflect the dynamic response of market players to the operation of the electricity market.

[0071] Application examples:

[0072] This application example implements the pre-training method for unit declaration agents in a power market of the present invention on a unit declaration agent in a simulated power market. The simulated power market is provided with three nodes and three units. In this example, the safety constrained unit commitment (SCUC) and safety constrained economic dispatch (SCED) algorithms are used for clearing calculations. The number of iterations is set to 100. The data set used is the operation scenario data set extracted from the actual simulation project. The specific data is as follows:

[0073] Node load data set for node 1 The specific values ​​are as follows:

[0074]

[0075] Node load data set for node 2 The specific values ​​are as follows:

[0076]

[0077] Node load data set for node 3 The specific values ​​are as follows:

[0078]

[0079] Tie line power data set X = {X1, X2, ..., X24}, the specific values ​​are as follows:

[0080] <![CDATA[X1]]> <![CDATA[X2]]> <![CDATA[X3]]> <![CDATA[X4]]> <![CDATA[X5]]> <![CDATA[X6]]> 3.14 3.14 3.14 3.14 3.14 3.14 <![CDATA[X7]]> <![CDATA[X8]]> <![CDATA[X9]]> <![CDATA[X 10 ]]> <![CDATA[X 11 ]]> <![CDATA[X 12 ]]> 4.56 4.56 4.56 4.56 4.56 4.56 <![CDATA[X 13 ]]> <![CDATA[X 14 ]]> <![CDATA[X 15 ]]> <![CDATA[X 16 ]]> <![CDATA[X 17 ]]> <![CDATA[X 18 ]]> 5.11 5.11 5.11 5.11 5.11 5.11 <![CDATA[X 19 ]]> <![CDATA[X 20 ]]> <![CDATA[X 21 ]]> <![CDATA[X 22 ]]> <![CDATA[X 23 ]]> <![CDATA[X 24 ]]> 4.68 4.68 4.68 4.68 4.68 4.68

[0081] Non-market unit planned output data set Y = {Y1, Y2, ..., Y 24}, the specific values ​​are as follows:

[0082] <![CDATA[Y1]]> <![CDATA[Y2]]> <![CDATA[Y3]]> <![CDATA[Y4]]> <![CDATA[Y5]]> <![CDATA[Y6]]> 4.19 4.19 4.19 4.19 4.19 4.19 <![CDATA[Y7]]> <![CDATA[Y8]]> <![CDATA[Y9]]> <![CDATA[Y 10 ]]> <![CDATA[Y 11 ]]> <![CDATA[Y 12 ]]> 4.19 4.19 4.19 4.19 4.19 4.19 <![CDATA[Y 13 ]]> <![CDATA[Y 14 ]]> <![CDATA[Y 15 ]]> <![CDATA[Y 16 ]]> <![CDATA[Y 17 ]]> <![CDATA[Y 18 ]]> 4.19 4.19 4.19 4.19 4.19 4.19 <![CDATA[Y 19 ]]> <![CDATA[Y 20 ]]> <![CDATA[Y 21 ]]> <![CDATA[Y 22 ]]> <![CDATA[Y 23 ]]> <![CDATA[Y 24 ]]> 4.19 4.19 4.19 4.19 4.19 4.19

[0083] After pre-training, the revenue data curves of the three units are as follows: Figure 3 As shown, Figure 3 The three curves in the figure correspond to the three units. Figure 3 It can be seen that the profit data of the three units fluctuated abnormally from round 1 to round 71, tended to be flat in round 86, and the profit data of three consecutive iterations in round 87 were the same.

[0084] Figure 4 , Figure 5 , Figure 6 and Figure 7 They are the declared data curves of unit G1 in rounds 86 to 89, and it can be seen that the declared data of unit G1 in rounds 87 to 89 are the same. Since the declared data curves of units G2 and G3 are almost the same as those of unit G1, they will not be repeated here.

Claims

1. A pre-training method for unit declaration agents in a power market, wherein each market unit in a power system has a corresponding unit declaration agent in the power market, and each unit declaration agent obtains declaration data through a declaration agent model, characterized in that: include, Establish a data set related to node load, tie-line power and planned output of non-market units in power system operation scenarios; The declaration agent model obtains the declaration data through the data set; Perform clearing calculations based on the declared data to obtain the revenue data of the declared agent of each unit; Using a genetic algorithm and based on the revenue data, the parameters in the unit declaration agent model are updated and optimized to obtain new declaration data; The iteration of generating corresponding new income data from the new declaration data; The iteration is terminated if the declared data and income data are the same for three consecutive iterations or the number of iterations reaches the set value.

2. The pre-training method for unit reporting agents in the power market according to claim 1 is characterized in that: The declaration agent model obtains declaration data with the goal of maximizing unit profit.

3. The pre-training method for unit reporting agents in the power market according to claim 1 or 2, characterized in that: The data set includes a node load data set Q, a tie line power data set X, and a non-market unit planned output data set Y; wherein, The node load data set Q: Q = {Q1, Q2, ..., Q M }, Q1, Q2, …, Q M are the average node loads in the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; where the average node load Q m : In the formula, Q m，d is the node load in the mth period of the dth day in the operation scenario; D is the total number of days in the operation scenario; The tie line power data set X: X={X1, X2, ......, X M }, X1, X2, …, X M are the tie line powers of the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; where the average tie line power X in the mth period in the operation scenario is m : In the formula, X m，d is the tie line power in the mth period of the dth day in the operation scenario; D is the total number of days in the operation scenario; The non-market unit planned output data set Y: Y = {Y1, Y2, ..., Y M }, Y1, Y2, …, Y M are the average non-market unit planned outputs in the 1st, 2nd, …, and Mth periods in the operation scenario, respectively; among which, the average non-market unit planned output Y in the mth period in the operation scenario m : Where Y m，d is the planned output of non-market units in the mth period on the dth day in the operation scenario; D is the total number of days in the operation scenario.

Images