An optimization decision method for power generation plan of coal-fired power plant

Abstract

The application discloses a kind of optimization decision method of coal-fired power plant power generation plan, belong to electric power production technical field, constructs migration learning model, predicts the carbon asset price of national carbon emission market;Build multi-source data fusion model, predict coal and electric power asset price;Based on the power generation production profit model of coal-fired power plant, based on the carbon emission cost and coal purchase cost of coal-fired power plant's electricity sale income, carbon quota asset, construct;With production rule as constraint, optimize and decide the carbon emission of power plant every week, coal purchasing quantity and power generation quantity with the maximum production profit as target.This application effectively avoids the problem that power plant cannot accurately evaluate production cost, and then cannot optimize power generation plan, proposes the prediction method of carbon, coal, electricity three types of asset prices, fully excavates the data characteristics of trading market, establishes comprehensive income calculation method for coal-fired power plant production activity, helps to establish power generation benefit evaluation and control measure, reduces the influence of centralized carbon trading and blind coal purchase.

Description

Technical Field

[0001] This invention belongs to the field of power production technology, specifically relating to an optimization decision-making method for power generation planning in coal-fired power plants. Background Technology

[0002] With the official launch of the national carbon emissions trading market, the power industry has become one of the first pilot industries to be included in the market. As high-carbon emission power generation enterprises, the carbon emission costs of coal-fired power plants have become an important factor affecting their production revenue.

[0003] Under the "dual carbon" target framework, coal-fired power plants are facing increasingly difficult survival conditions due to pressure from new energy sources. Furthermore, they lack experience in managing diversified market asset portfolios. With market risks expected to increase in the future, the difficulty of production management will inevitably rise. To ensure sustainable development in a multi-market trading environment, optimizing power generation decisions, improving carbon emission management methods, and formulating coal procurement strategies are crucial issues that coal-fired power plants urgently need to address.

[0004] Current technologies only combine coal purchase costs and electricity sales revenue to calculate the production profit of coal-fired power plants, using this as the ultimate goal for optimizing their production. However, these technologies do not consider the carbon emission costs of coal-fired power plant production in their power generation plans, nor do they take into account government-issued carbon allowances, which are traded centrally before the carbon emission compliance period. This prevents power plants from accurately and comprehensively assessing their production costs, thus hindering their ability to optimally schedule power generation, reducing their profitability and viability, and preventing coal-fired power from effectively fulfilling its fundamental role in ensuring basic supply and regulating the system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an optimization decision-making method for power generation planning of coal-fired power plants. This method can comprehensively evaluate the production costs of coal-fired power plants and determine the optimal weekly power generation, carbon emissions, and coal purchase volume, thereby guiding coal-fired power enterprises to make scientific and reasonable arrangements and plans for their production activities.

[0006] The technical solution to achieve the above objective is: an optimization decision-making method for power generation planning in coal-fired power plants, characterized by the following steps:

[0007] Step S1: Construct a transfer learning model to predict the price of carbon assets in the national carbon emissions market;

[0008] Step S2: Construct a multi-source data fusion model to predict coal and electricity asset prices;

[0009] Step S3: Based on the electricity sales revenue, carbon quota assets, carbon emission costs, and coal purchase costs of coal-fired power plants, construct a profit model for power generation production of coal-fired power plants.

[0010] Step S4: With the goal of maximizing production profits and constrained by production rules, optimize the decision-making regarding the power plant's weekly carbon emissions, coal procurement, and power generation.

[0011] Furthermore, in step S1, the carbon price prediction steps based on the transfer learning model are as follows:

[0012] S11: Training a prototype carbon price prediction model based on a large amount of data from mature carbon trading markets;

[0013] Let W = {w1, w2, ..., w n1}, D={d1, d2, …, d n2} represent the set of weights and the set of offsets in the neural network, respectively, T={T1, T2, …,T m Let} be the set of neural network functions, and define the error δ = Σ[y t – T W,D (A t )] 2 Minimize this error to obtain the optimal network model parameters W. * D * ;in, Let A be the carbon price in period t. t L is the set of carbon valences spanning T before period t. W,D It is a set of functions with randomly given initial values ​​of model parameters W and D;

[0014] S12: Training a target model for carbon price prediction based on limited data from emerging carbon trading markets;

[0015] The model parameters W obtained from step S11 are optimized. * D * The neural network model is transferred to the target carbon market, inputting sample data and calculating the error Δ=Σ[z]. r – T W*,D* (B r )] 2 Minimize this error Δ to obtain the target model for carbon price prediction, where T W*,D* W * D * A set of functions for initial values ​​of model parameters. It is the carbon price in period r, B r It is the set of carbon valences spanning T before period r.

[0016] Furthermore, the neural network in steps S11 and S12 is a deep learning network model, which includes an input layer, a hidden layer, and an output layer. First, the deep learning network model is trained, and then the result is calculated using the trained model.

[0017] Further, in step S2, the price of the coal or electricity asset is denoted as y. p The market trading volume and transaction amount related to price are denoted as y. v , y a The trader sentiment value related to price is denoted as y. s Then the multi-source data is denoted as the vector Y=[y] formed by fusing the above information. p , y v , y a , y s The steps for asset price prediction based on multi-source data fusion are as follows:

[0018] S21: Calculate the mean vector of multi-source data information for one week, i.e., five consecutive trading days. :

[0019] ,

[0020] Where n is a certain trading day, i is the 5 trading days before trading day n, and Y i This is the multi-source data vector for day i; in the sample dataset, starting from the 6th trading day, the mean vector of the multi-source data is calculated daily to form the input data vector. The corresponding output data vector is m is the total number of samples;

[0021] S22: Input the input and output sets of the sample data into the intelligent learning model E, whose learning function set is E={E1,E2, …,E2}. t}, where t is the number of functions; obtain the learning error γ=1 / (m-5)|O – E(I)|, minimize this error, and obtain the optimal learning model weight parameters P. * and offset parameter Q * ; Input data vector of the asset whose price is to be predicted , where n is the amount of data to be predicted, The data was calculated using the formula in S21 from the multi-source data of the last 5 days of the sample set, and then the calculation was performed day by day using this method. ;Will Input the intelligent learning model E to obtain asset price predictions. ,in With P * Q * This is a set of functions for model parameters.

[0022] Furthermore, the price, trading volume, and transaction amount data for coal or electricity assets come from publicly available data on official trading websites; the price-related trader sentiment data comes from trader communities and public platforms like Weibo.

[0023] Furthermore, among them, the trader sentiment value y s The calculation method is as follows:

[0024]

[0025] Where, n t e represents the number of phrases in the keyword text of the public platform. i This represents the score of a phrase i in a natural language dictionary. Natural language dictionaries include commonly used Chinese and English sentiment analysis dictionaries such as SnowNLP, BosonNLP, CoreNLP, pyLTP, or pyNLPir.

[0026] Furthermore, in step S3, the profit model for power generation from a coal-fired power plant is constructed as follows:

[0027] ,

[0028] This model is used to calculate the profit Y generated by a coal-fired power plant, where V p For the annual revenue from electricity sales, V a The value of carbon allowance assets allocated to the government, V e For carbon emission costs, V c For coal purchase costs, This is a conversion factor for water consumption, desulfurization and denitrification, and labor costs in coal-fired power plants.

[0029] Furthermore, the annual revenue V from electricity sales is calculated using the following method. p :

[0030] ,

[0031] in, The agreed-upon long-term contract electricity price is a fixed value. Let $ be the auction price for week i, and $ be the arithmetic average of the predicted electricity prices for the five trading days within week i. For the long-term contract electricity volume in week i, Let m be the bidding volume for week i, and m be the number of weeks in the year.

[0032] Carbon allowance asset V is calculated using the following method. a :

[0033] ,

[0034] in, The peak carbon price is predicted for the carbon emission compliance period. CEA carbon allowances are distributed free of charge by the government.

[0035] The carbon emission cost V is calculated using the following method.e :

[0036] ,

[0037] in, This represents the predicted trough value of carbon prices during the carbon emission compliance period. Let be the carbon emissions for week i. ρ is the annual agreement transfer price for CCER, which is a fixed value; ρ is the maximum percentage of CCER allowed to offset CEA; and m is the number of weeks in a year.

[0038] The coal purchase cost V is calculated using the following method. c :

[0039] ,

[0040] in, The annual contract price for coal is a fixed value. The price of coal in the spot market during week i is denoted as , and the arithmetic average of the predicted coal prices for the five trading days within week i is denoted as . This represents the breakdown volume of coal purchased under the annual contract for the i-th week. Let m represent the coal purchase volume in the spot market during week i, and m be the number of weeks in a year.

[0041] Furthermore, in step S4, production rule constraints are added to the power generation production profit model to establish an optimization mathematical model with multiple equality and inequality constraints. By solving the model to maximize production profit, the power plant's weekly carbon emissions, coal purchases, and power generation are optimized.

[0042] Furthermore, production rules constraints include carbon emission constraints, planned power generation constraints, annual contracted total power generation constraints, weekly coal consumption constraints, coal inventory constraints, weekly coal purchase constraints, and generator unit output constraints.

[0043] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0044] This invention establishes a complete method for calculating the production cost, profit, and production rules of coal-fired power generation, effectively avoiding the problem that coal-fired power plants cannot accurately assess production costs and thus cannot optimize power generation plans. It proposes a method for predicting the prices of three types of assets: carbon, coal, and electricity, fully explores the data characteristics of the trading market, establishes a comprehensive revenue calculation method for the production activities of coal-fired power plants, and helps to establish power generation benefit evaluation and control measures, reducing the impact of centralized carbon trading and blind coal purchases. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the optimization decision-making method for power generation planning in a coal-fired power plant according to the present invention. Detailed Implementation

[0046] The present invention will be further illustrated below with reference to specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0047] An optimization decision-making method for power generation planning in coal-fired power plants includes the following steps:

[0048] Step S1: Construct a transfer learning model to predict the price of carbon assets in the national carbon emissions market.

[0049] Step S2: Construct a multi-source data fusion model to predict coal and electricity asset prices.

[0050] Step S3: Based on the electricity sales revenue, carbon quota assets, carbon emission costs, and coal purchase costs of coal-fired power plants, construct a profit model for power generation production of coal-fired power plants.

[0051] Step S4: With the goal of maximizing production profits and constrained by production rules, optimize the decision-making regarding the power plant's weekly carbon emissions, coal procurement, and power generation.

[0052] In step S1, a transfer learning model is constructed to predict the price of carbon assets in the national carbon emissions market. The specific implementation method is as follows:

[0053] make This is a sample from the t-th period of a mature carbon trading market, in which It is the set of carbon valences spanning T before period t. It is the carbon price within this set. It is the carbon price in period t.

[0054] make This is the r-th period sample of emerging target markets, where It is the set of carbon valences spanning T before period r. It is the carbon price within this set. It is the carbon price in period r.

[0055] make This is the s-th expected forecast data for emerging target markets, among which It is the set of carbon valences spanning T before period s. It is the carbon price within this set. It is the carbon price for the s-th period to be predicted.

[0056] The steps for carbon price prediction based on a transfer learning model are as follows:

[0057] S11: Train a prototype carbon price prediction model based on a large amount of data from mature carbon trading markets.

[0058] Let W = {w1, w2, ..., w n1}, D={d1, d2, …, d n2 Let L = {L1, L2, …, L} be the set of weights and the set of offsets of the neural network, respectively. m Let} be the set of neural network functions, and define the error δ = Σ[y t – L W,D (A t )] 2 L W,D Given a set of functions with randomly assigned initial model parameters W and D, the optimal network model parameters W are obtained by minimizing the error δ. * D * The specific implementation process is as follows:

[0059] The neural network used in this embodiment is a deep learning neural network, which consists of an input layer, a hidden layer, and an output layer.

[0060] The input layer nodes are carbon prices spanning T years prior to period t. The number of nodes is T; the number of hidden layer nodes is 2. n n is a natural number (in this implementation, n=8 is used as an example, then the number of nodes is 256), and the activation function is the Sigmoid function σ(x)=1 / (1+e -x The output layer has one node. The set of weights is W = {w1, w2, ..., w...}. 256T+256} and offset D = {d1, d2, …, d 257 The initial value is random. The training process of the deep learning neural network is as follows:

[0061] Calculate the value of the first node in the hidden layer: h1 = σ(L1( ))=σ(L1( ))= σ( );

[0062] Calculate the value of the second node in the hidden layer: h2 = σ(L2( ))=σ(L2( ))= σ( ).

[0063] By analogy, calculate the values ​​of all 256 hidden layer nodes to form the set of hidden layer node values ​​H. t ={h1, h2, h3, …h256}.

[0064] Calculate the output layer node values: o t =σ(L 257 ( ))=σ(L 257(h1, h2, h3, …h256))= σ( )

[0065] Calculate the training error between the output value at period t and the true value at period t: δ t =(y t - o t ) 2

[0066] Following the method described above, calculate the total training error δ = Σ(y) for all periods in the sample. t - o t ) 2 Using the least squares method, the neural network weights W={w1, w2,…,w... are calculated to minimize the total error. 256T+256} and offset D = {d1, d2, ..., d 257}, thereby determining the optimal parameters W of the deep learning neural network. * and D * .

[0067] S12: Train the target model for carbon price prediction based on limited data from emerging carbon trading markets; optimize the model parameters W obtained in step S11. * D * The neural network model is transferred to the target carbon market, inputting sample data and calculating the error Δ=Σ[z]. r – L W*,D* (B r )] 2 L W*,D* W * D * This is the set of functions for initial model parameters. Minimizing this error Δ optimizes the carbon price prediction target model parameters W. ** D ** The specific implementation process is as follows:

[0068] The neural network model is a deep learning neural network derived from S11, consisting of an input layer, hidden layers, and an output layer.

[0069] The input layer nodes are carbon prices spanning T before period r. The number of nodes is T; the number of hidden layer nodes is 2. n If n=8, then the number of nodes is 256, and the activation function is the Sigmoid function σ(x)=1 / (1+e^(-1 / 2)). -x The output layer has one node. The initial values ​​of the weight set W and the offset set D are derived from S11. * D * The training process of a deep learning neural network is as follows:

[0070] Calculate the value of the first node in the hidden layer: h1 = σ(L1( ))=σ(L1( ))= σ( ).

[0071] Calculate the value of the second node in the hidden layer: h2 = σ(L2( ))=σ(L2( ))= σ( ).

[0072] By analogy, calculate the values ​​of all 256 hidden layer nodes to form the set of hidden layer node values ​​H. r ={h1, h2, h3, …h256}.

[0073] Calculate the output layer node values: o r =σ(L 257 ( ))=σ(L 257 (h1, h2, h3, …h256))= σ( )

[0074] Calculate the training error between the output value at time r and the true value at time r: δ r =(z r - o r ) 2

[0075] Following the method described above, calculate the total training error δ = Σ(z) for all periods in the sample. r - o r ) 2 Using the least squares method, the neural network weights W={w1, w2,…,w... are calculated to minimize the total error. 256T+256} and offset D = {d1, d2, ..., d 257}, thereby determining the optimal parameters W of the deep learning neural network. ** and D ** .

[0076] S13: Based on the trained carbon price prediction target model, input the data to be predicted to obtain the price to be predicted. The specific implementation process is as follows:

[0077] The neural network model is a deep learning neural network trained by S12, consisting of an input layer, hidden layers, and an output layer.

[0078] The input layer nodes are carbon valence values ​​spanning T before period s. The number of nodes is T; the number of hidden layer nodes is 2. n If n=8, then the number of nodes is 256, and the activation function is the Sigmoid function σ(x)=1 / (1+e^(-1 / 2)).-x The output layer has one node. The weight set W and the offset set D are obtained from training S12. ** D ** The prediction process of a deep learning neural network is as follows:

[0079] Calculate the value of the first node in the hidden layer: h1 = σ(L1( ))=σ(L1( ))= σ( ).

[0080] Calculate the value of the second node in the hidden layer: h2 = σ(L2( ))=σ(L2( ))= σ( ).

[0081] By analogy, calculate the values ​​of all 256 hidden layer nodes to form the set of hidden layer node values ​​H. s ={h1, h2, h3, …h256}.

[0082] Calculate the output layer node values: o s =σ(L 257 ( ))=σ(L 257 (h1, h2, h3, …h256))= σ( )

[0083] Output layer node values This refers to the carbon price in the s-th period to be predicted. This enables carbon price prediction in emerging carbon trading markets.

[0084] In step S2, a multi-source data fusion model is constructed to predict coal and electricity asset prices. The specific implementation method is as follows:

[0085] The price of coal or electricity assets is denoted as y. p The market trading volume and transaction amount related to price are denoted as y. v ,y a The trader sentiment value related to price is denoted as y. s Then the multi-source data is denoted as the vector Y=[y] formed by fusing the above information. p ,y v , y a , y s Price, volume, and transaction value data for coal or electricity assets are sourced from publicly available data on official trading websites. Trader sentiment data related to prices comes from trader communities, Weibo, and other public platforms; among them, the trader sentiment value y... s The calculation method is as follows:

[0086]

[0087] Where, n t e represents the number of phrases in the keyword text of the public platform. i This represents the score of a phrase i in a natural language dictionary. Natural language dictionaries include commonly used Chinese and English sentiment analysis dictionaries such as SnowNLP, BosonNLP, CoreNLP, pyLTP, and pyNLPir.

[0088] The steps for asset price prediction based on multi-source data fusion are as follows:

[0089] S21: Calculate the mean vector of multi-source data information for one week, i.e., five consecutive trading days. :

[0090] ,

[0091] Where n is a certain trading day, i is the 5 trading days before trading day n, and Y i This is the multi-source data vector for day i. Starting from the 6th trading day, the mean vector of the multi-source data is calculated daily in the sample dataset to form the input data vector. The corresponding output data vector is , m is the total number of samples.

[0092] S22: Input the input and output sets of the sample data into the intelligent learning model E, which consists of a set of learning functions E={E1, E2, …,E…} t}, where t is the number of functions. Obtain the learning error γ = 1 / (m-5)|O – E(I)|, minimize this error, and obtain the optimal model weight parameters P. * and offset parameter Q * ; Input data vector of the asset whose price is to be predicted , where n is the amount of data to be predicted, The data was calculated using the formula in S21 from the multi-source data of the last 5 days of the sample set, and then the calculation was performed day by day using this method. .Will Input the intelligent learning model E to obtain asset price predictions. ,in With P * and Q * This is a set of functions for model parameters.

[0093] The intelligent learning model E is a deep learning network model, and the specific implementation process of model training is as follows:

[0094] The input layer nodes of a deep learning neural network are the mean vectors of multi-source data information corresponding to a certain trading day u. The number of nodes in the first layer is 4; the number of nodes in the hidden layer is 2. n n is a natural number (in this implementation, n=6 is used as an example, then the number of nodes is 64), and the activation function is the Sigmoid function σ(x)=1 / (1+e -x The output layer has one node. The set of weights is P = {p1, p2, ..., p...}. 320} and offset Q = {q1, q2, …, q 65 The initial values ​​are random. The training process of a deep learning neural network is as follows:

[0095] Calculate the value of the first node in the hidden layer: h1 = σ(E1( ))=σ(E1( ))= σ( ). These represent the average price of coal or electricity assets, the average market trading volume related to price, the average market trading amount related to price, and the average trader sentiment value related to price, respectively, for the five trading days preceding a given trading day u. The calculation method is the formula in step S21.

[0096] Calculate the value of the second node in the hidden layer: h2 = σ(E2( ))=σ(E2( ))= σ( ).

[0097] By analogy, calculate the values ​​of all 64 hidden layer nodes to form the set of hidden layer node values ​​H. u ={h1, h2, h3,…h64}.

[0098] Calculate the output layer node values: o u =σ(E 65 ( ))=σ(E 65 (h1, h2, h3, …h64))= σ( )

[0099] Calculate the training error between the output value and the true value of u on the trading day: δ u =(y pu - o u ) 2 y pu This represents the true price of coal or electricity assets on a given trading day.

[0100] Following the method described above, the total training error δ = Σ(y) for all trading days in the sample is calculated. pu – o u) 2 Using the least squares method, the neural network weights P={p1, p2,…,p} are calculated to minimize the total error. 320} and offset Q = {q1, q2, ..., q 65}, thereby determining the optimal parameters P of the deep learning neural network. * and Q * .

[0101] After the deep learning network model is trained, it is used to predict the prices of coal and electricity assets. The prediction process for coal and electricity asset prices is the same. Taking coal price prediction as an example, the specific implementation process is as follows:

[0102] The input layer nodes of the neural network model are the mean vectors of multi-source data information corresponding to a certain trading day v. The number of nodes in the first layer is 4; the number of nodes in the hidden layer is 2. n If n=6, then the number of nodes is 64, and the activation function is the Sigmoid function σ(x)=1 / (1+e^(-1 / 2)). -x The output layer has one node. The values ​​of the weight set P and the offset set Q are obtained from training. * Q * The prediction process is as follows:

[0103] Calculate the value of the first node in the hidden layer: h1 = σ(E1( ))=σ(E1( ))= σ( ). These represent the average price of coal or electricity assets, the average market trading volume related to the price, the average market trading amount related to the price, and the average trader sentiment value related to the price, respectively, for the five trading days preceding a certain trading day v to be predicted. The calculation method is the formula in step S21.

[0104] Calculate the value of the second node in the hidden layer: h2 = σ(E2( ))=σ(E2( ))= σ( ).

[0105] By analogy, calculate the values ​​of all 64 hidden layer nodes to form the set of hidden layer node values ​​H'. v ={h1', h2',h3', …h64'}.

[0106] Calculate the output layer node value: o' v =σ(E 65 ( ))=σ(E 65 (h1', h2', h3', …h64'))= σ( )

[0107] Output layer node values This means that the coal price on the v-th trading day to be predicted is calculated by performing the above calculation process on the input data for all trading days to be predicted, thus forming a set of predicted coal prices. This enables the prediction of coal asset prices.

[0108] In step S3, the profit model for power generation of coal-fired power plants is constructed based on their electricity sales revenue, carbon quota assets, carbon emission costs, and coal purchase costs as follows:

[0109] The annual revenue V from electricity sales is calculated using the following method. p :

[0110] ,

[0111] in, The agreed-upon long-term contract electricity price is a fixed value. Let $ be the auction price for week i, and $ be the arithmetic average of the predicted electricity prices for the five trading days within week i. For the long-term contract electricity volume in week i, Let m be the bidding volume for week i, and m be the number of weeks in the year.

[0112] Carbon allowance asset V is calculated using the following method. a :

[0113] ,

[0114] in, The peak carbon price is predicted for the carbon emission compliance period. CEA carbon allowances are distributed free of charge by the government.

[0115] The carbon emission cost V is calculated using the following method. e :

[0116] ,

[0117] in, This represents the predicted trough value of carbon prices during the carbon emission compliance period. Let be the carbon emissions for week i. ρ is the annual agreement transfer price for CCER, which is a fixed value; ρ is the maximum percentage of CCER allowed to offset CEA; and m is the number of weeks in a year.

[0118] The coal purchase cost V is calculated using the following method. c :

[0119] ,

[0120] in, The annual contract price for coal is a fixed value. The price of coal in the spot market during week i is denoted as , and the arithmetic average of the predicted coal prices for the five trading days within week i is denoted as . This represents the breakdown volume of coal purchased under the annual contract for the i-th week. Let m represent the coal purchase volume in the spot market during week i, and m be the number of weeks in a year.

[0121] Based on the above data, the following profit model for power generation is obtained:

[0122] ,

[0123] The profit Y from power generation at a coal-fired power plant is calculated using this model, where, This refers to the conversion factor for costs such as water, desulfurization and denitrification, and labor in coal-fired power plants, based on engineering experience. The value is 1.43.

[0124] In step S4, with the goal of maximizing production profit and constrained by production rules, the power plant's weekly carbon emissions, coal procurement volume, and power generation are optimized. The process is as follows:

[0125] The constraints on production rules include carbon emission constraints, planned power generation constraints, annual contracted total power generation constraints, weekly coal consumption constraints, coal inventory constraints, weekly coal purchase constraints, and generator unit output constraints.

[0126] Carbon emission constraints are: , where r s Carbon emission intensity for power supply.

[0127] The planned power generation constraint is: Q i,max, Q i,min These are the highest and lowest power generation in week i, respectively, determined by a 5% fluctuation above or below the planned power generation.

[0128] The total annual contracted electricity volume constraint is: Q l =ΣQ l,i Q l,max This represents the maximum total electricity volume under long-term (annual) contracts.

[0129] Weekly coal consumption constraints are: B i r represents the amount of standard coal consumed in week i. b This refers to the standard coal consumption rate per unit of electricity.

[0130] Coal inventory constraints are:

[0131] ,

[0132] Among them, R i Let R0 and R be the coal inventory at the end of week i. 52 These represent the initial and year-end coal inventories, R. min R max These represent the minimum and maximum coal inventories, respectively.

[0133] The weekly coal purchase volume constraint is: .

[0134] The generator set output constraint is: Among them, the maximum output P max The rated power and minimum output P of the unit. min It is generally 40% of the rated power, where j represents any unit in the power plant. Under normal circumstances, the unit operates for 24 hours a day.

[0135] By incorporating production rule constraints into the power generation profit model, an optimization mathematical model with multiple equality and inequality constraints is established as follows:

[0136]

[0137] ST

[0138]

[0139]

[0140]

[0141]

[0142]

[0143]

[0144] This embodiment uses the number of weeks as an example to construct an optimization model, where m is the number of weeks in a year. However, the method of this patent is not limited to weekly planning; it can also be applied to daily, monthly, or quarterly planning scenarios.

[0145] By solving the model to maximize production profit, the weekly carbon emissions S of the power plant can be optimized. i Weekly contract coal purchase volume Spot coal purchase volume Contracted power generation and bidding for power generation .

[0146] This invention establishes a complete method for calculating the production cost, profit, and production rules of coal-fired power generation, effectively avoiding the problem that power plants cannot accurately assess production costs and thus cannot optimize power generation plans. It proposes a method for predicting the prices of three types of assets: carbon, coal, and electricity, fully explores the data characteristics of the trading market, establishes a comprehensive revenue calculation method for the production activities of coal-fired power plants, and helps to establish power generation benefit evaluation and control measures, reducing the impact of centralized carbon trading and blind coal purchases.

[0147] Those skilled in the art should recognize that the above embodiments are merely illustrative of the invention and not intended to limit the invention. Any variations or modifications to the above embodiments that are within the spirit and essence of the invention will fall within the scope of the claims.

Claims

1. An optimization decision-making method for power generation planning in coal-fired power plants, characterized in that, The decision-making process for optimizing power generation plans in coal-fired power plants includes the following steps: Step S1: Construct a transfer learning model to predict the price of carbon assets in the national carbon emissions market; S11: Training a prototype carbon price prediction model based on a large amount of data from mature carbon trading markets; Let W = {w1, w2, ..., w n1 }, D={d1, d2, …, d n2 } represent the set of weights and the set of offsets in the neural network, respectively, T={T1, T2, …,T m Let} be the set of neural network functions, and define the error δ = Σ[y t – T W,D (A t )] 2 Minimize this error to obtain the optimal network model parameters W. * D * ;in, Let A be the carbon price in period t. t L is the set of carbon valences spanning T before period t. W,D It is a set of functions with randomly given initial values ​​of model parameters W and D; S12: Training a target model for carbon price prediction based on limited data from emerging carbon trading markets; The model parameters W obtained from step S11 are optimized. * D * The neural network model is transferred to the target carbon market, inputting sample data and calculating the error Δ=Σ[z]. r – T W*,D* (B r )] 2 Minimize this error Δ to obtain the target model for carbon price prediction, where T W*,D* W * D * A set of functions for initial values ​​of model parameters. This is the carbon price in period r, B r It is the set of carbon valences spanning T before period r; The neural network in steps S11 and S12 is a deep learning network model. The deep learning neural network includes an input layer, a hidden layer, and an output layer. First, the deep learning network model is trained, and then the result is calculated using the trained model. Step S2: Construct a multi-source data fusion model to predict coal and electricity asset prices; In step S2, the price of the coal or electricity asset is denoted as y. p The market trading volume and transaction amount related to price are denoted as y. v , y a The trader sentiment value related to price is denoted as y. s Then the multi-source data is denoted as the vector Y=[y] formed by fusing the above information. p , y v , y a , y s The steps for asset price prediction based on multi-source data fusion are as follows: S21: Calculate the mean vector of multi-source data information for one week, i.e., five consecutive trading days. : ， Where n is a certain trading day, i is the 5 trading days before trading day n, and Y i This is the multi-source data vector for day i; in the sample dataset, starting from the 6th trading day, the mean vector of the multi-source data is calculated daily to form the input data vector. The corresponding output data vector is m is the total number of samples; S22: Input the input and output sets of the sample data into the intelligent learning model E, which consists of a set of learning functions E={E1, E2, …,E…} t }, where t is the number of functions; obtain the learning error γ=1 / (m-5)|O – E(I)|, minimize this error, and obtain the optimal model weight parameters P. * and offset parameter Q * ; Input data vector of the asset whose price is to be predicted , where n is the amount of data to be predicted, The results were calculated using the formula in S21 from the multi-source data of the last 5 days of the sample set, and then the calculation was performed day by day using this method. ;Will Input the intelligent learning model E to obtain asset price predictions. ,in With P * Q * A set of functions for model parameters Step S3: Based on the electricity sales revenue, carbon quota assets, carbon emission costs, and coal purchase costs of coal-fired power plants, construct a profit model for power generation production of coal-fired power plants. Step S4: With the goal of maximizing production profits and constrained by production rules, optimize the decision-making regarding the power plant's weekly carbon emissions, coal procurement, and power generation.

2. The optimization decision-making method for power generation planning of coal-fired power plants according to claim 1, characterized in that, Price, volume, and transaction value data for coal or electricity assets are obtained from publicly available data on official trading websites; trader sentiment data related to prices comes from trader communities and public platforms such as Weibo.

3. The optimization decision-making method for power generation planning of coal-fired power plants according to claim 1, characterized in that, Trader sentiment value y s The implementation method is as follows: ， Where, n t e represents the number of phrases in the keyword text of the public platform. i This represents the score of a phrase i in a natural language dictionary. Natural language dictionaries include commonly used Chinese and English sentiment analysis dictionaries such as SnowNLP, BosonNLP, CoreNLP, pyLTP, or pyNLPir.

4. The optimization decision-making method for power generation planning of coal-fired power plants according to claim 1, characterized in that, In step S3, the profit model for power generation from a coal-fired power plant is constructed as follows: ， This model is used to calculate the profit Y generated by a coal-fired power plant, where V p For the annual revenue from electricity sales, V a The value of carbon allowance assets allocated to the government, V e For carbon emission costs, V c For coal purchase costs, This is a conversion factor for water consumption, desulfurization and denitrification, and labor costs in coal-fired power plants.

5. The optimization decision-making method for power generation planning of a coal-fired power plant according to claim 1, characterized in that, The annual revenue V from electricity sales is calculated using the following method. p : ， in, The agreed-upon long-term contract electricity price is a fixed value. Let $ be the auction price for week i, and $ be the arithmetic average of the predicted electricity prices for the five trading days within week i. For the long-term contract electricity volume in week i, Let m be the bidding volume for week i, and m be the number of weeks in the year. Carbon allowance asset V is calculated using the following method. a : ， in, The peak carbon price is predicted for the carbon emission compliance period. CEA carbon allowances are distributed free of charge by the government. The carbon emission cost V is calculated using the following method. e : ， in, This represents the predicted trough value of carbon prices during the carbon emission compliance period. Let be the carbon emissions for week i. ρ is the annual agreement transfer price for CCER, which is a fixed value; ρ is the maximum percentage of CCER allowed to offset CEA; and m is the number of weeks in a year. The coal purchase cost V is calculated using the following method. c : ， in, The annual contract price for coal is a fixed value. The price of coal in the spot market during week i is denoted as , and the arithmetic average of the predicted coal prices for the five trading days within week i is denoted as . This represents the breakdown volume of coal purchased under the annual contract for the i-th week. Let m represent the coal purchase volume in the spot market during week i, and m be the number of weeks in a year.

6. The optimization decision-making method for power generation planning of a coal-fired power plant according to claim 1, characterized in that, In step S4, production rule constraints are added to the power generation production profit model to establish an optimization mathematical model with multiple equality and inequality constraints. By solving the model to maximize production profit, the power plant's weekly carbon emissions, coal purchases, and power generation are optimized.

7. The optimization decision-making method for power generation planning of a coal-fired power plant according to claim 1, characterized in that: Production rules constraints include carbon emission constraints, planned power generation constraints, annual contracted total power generation constraints, weekly coal consumption constraints, coal inventory constraints, weekly coal purchase constraints, and generator unit output constraints.

Images