A novel optimization method for coal blending considering coal and carbon trading costs

Abstract

The present invention discloses a coal blending optimization method considering coal and carbon trading costs, comprising the following steps: obtaining coal blending related parameters, wherein the related parameters at least include historical power generation information; constructing a power generation prediction model based on the historical power generation information; obtaining a power generation prediction value based on the power generation prediction model, and then obtaining a power generation prediction error; constructing fuzzy sets of power generation prediction errors respectively; constructing corresponding DRO‑KL coal blending models and DRO‑W coal blending models respectively based on fuzzy sets under KL divergence and fuzzy sets under Wasserstein distance; then solving the optimal solutions of the DRO‑KL coal blending model and the DRO‑W coal blending model respectively; and obtaining the corresponding optimal coal blending scheme based on the optimal solution. The coal blending optimization method provided by the present invention not only reduces the total cost of power generation, including coal combustion cost and carbon trading cost, but also reduces carbon emissions and carbon trading costs.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal-fired power generation, and in particular relates to a coal blending optimization method taking into account coal and carbon trading costs. Background Art

[0002] Coal-fired power plants are one of the largest sources of carbon emissions, exacerbating global warming and climate change. Reducing carbon emissions from the coal-fired power generation industry has become an urgent issue that needs to be addressed.

[0003] In addition to the "hard technologies" of technological innovation or investment such as CCUS technology, "soft technologies" such as mixed coal blending and coal-biomass fuel co-firing have become more acceptable options for reducing carbon emissions from power plants. At present, there are a large number of experience-oriented coal blending methods for coal-fired power plants, that is, many power plants rely on the experience of employees to mix different types of coal, and a few use intelligent algorithms to complete automatic coal blending. However, in the coal blending process, the cost of coal is focused on while the carbon emissions in the combustion process and its corresponding carbon trading costs are ignored, resulting in high carbon trading costs and uneconomical. In addition, the required power generation plays an important guiding role in the coal blending process, but due to natural factors such as weather, temperature, and wind, new energy power generation has uncertainty, and the power demand of thermal power generation is uncertain in practice. Therefore, the current coal blending scheme does not take into account the uncertainty of power, resulting in the uneconomical selection of coal types in the coal blending scheme.

[0004] Under the premise of meeting uncertain electricity demand, how to formulate the optimal coal blending plan and find the appropriate coal type and quantity to meet the safety constraints and environmental emission requirements of coal burning while reducing the cost of coal-fired power generation and carbon emissions of coal-fired power generation is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] The present invention proposes a coal blending optimization method taking into account coal and carbon trading costs to solve the technical problems existing in the above-mentioned prior art.

[0006] To achieve the above object, the present invention provides a coal blending optimization method considering coal and carbon trading costs, comprising the following steps:

[0007] Acquire coal blending related parameters, wherein the related parameters at least include historical power generation information;

[0008] Based on the historical power generation information, a power generation prediction model based on LSTM network and machine learning is constructed;

[0009] Based on the power generation prediction model, a power generation prediction value is obtained, and then a power generation prediction error is obtained;

[0010] Based on KL divergence and Wasserstein distance, fuzzy sets of power generation forecast error are constructed respectively;

[0011] Based on the fuzzy sets under KL divergence and Wasserstein distance, the corresponding DRO-KL coal blending model and DRO-W coal blending model are constructed respectively.

[0012] With the goal of minimizing coal burning and carbon trading costs, the optimal solutions of the DRO-KL coal blending model and the DRO-W coal blending model are solved respectively;

[0013] A corresponding optimal coal blending scheme is obtained based on the optimal solution.

[0014] Optionally, the relevant parameters also include: coal type and its characteristic information and boiler safe combustion information; wherein, the coal type and its characteristic information include coal type, mining location, coal price, coal inventory, low heat content, received basis sulfur, dry ash-free basis volatile matter, received basis ash content, fixed carbon, high heat content and coal supply conditions; the boiler safe combustion information includes the boiler's calorific value requirements, sulfur content requirements, volatile content requirements and ash melting point requirements for burning mixed coal.

[0015] Optionally, the process of constructing a power generation prediction model includes: constructing an LSTM prediction model containing several layers of neural units, preprocessing the historical power generation information, inputting the preprocessed historical power generation information into the LSTM prediction model for processing, and obtaining the LSTM model power generation prediction value; based on the LSTM model power generation prediction value, constructing a random forest prediction model, a gradient boosting decision tree prediction model and an extreme gradient boosting prediction model; comparing the prediction accuracy of the LSTM prediction model, the random forest prediction model, the gradient boosting decision tree prediction model and the extreme gradient boosting prediction model, and taking the gradient boosting decision tree prediction model with the highest prediction accuracy as the final power generation prediction model.

[0016] Optionally, the fuzzy set under the KL divergence is:

[0017]

[0018]

[0019] Where P is the true distribution of power generation forecast error, is the empirical distribution of power generation forecast error, ε is the true value of power generation forecast error, is the empirical value of power generation forecast error, is the true distribution P and the empirical distribution of power generation forecast error The KL divergence between them, δ is the empirical distribution of the distance The farthest KL divergence (radius), is the set of all probability distributions, is a fuzzy set under KL divergence. Optionally, the DRO-KL coal blending model is as follows:

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031] Among them, CP id is the cost of the i-th type of coal on day d, x id is the decision variable for coal quantity selection, Car i is the carbon content per unit calorific value of the i-th coal, OF i is the carbon oxidation rate of the i-th coal, Q de is the power supply of the unit on day d, B e is the power supply reference value of the unit category, F1 is the unit cooling method correction coefficient, F r is the unit heat supply correction factor, F f is the unit load factor correction factor, l d is the actual power generation value, is the predicted power generation value, γ is the conversion coefficient between power generation and calorific value, η is the power generation efficiency of the unit, and q id is the calorific value of the i-th type of coal on the d-th day, MU s is the upper limit of coal type characteristics s, ML s is the lower limit of coal type characteristics s, M ids is the value of the characteristic s of the i-th type of coal on the d-th day, NL i is the upper limit of the quantity of the i-th type of coal, NU i is the lower limit of the quantity of the i-th type of coal, PL i is the upper limit of the proportion of the i-th coal, PU iis the lower limit of the proportion of the i-th type of coal, r is the upper limit of the total number of selected coal types, is the lower limit of the total number of selected coal types, N is the minimum limit of the amount of each selected coal type, M is a large number, y id is the decision variable for coal type selection, ε d is the power generation forecast error on day d.

[0032] Optionally, the constraints of the DRO-KL coal blending model are:

[0033]

[0034] Among them, δ is the radius in the fuzzy set that defines the KL divergence, and β is the Lagrange multiplier.

[0035] Optionally, the fuzzy set under the Wasserstein distance is:

[0036]

[0037]

[0038] Where P is the true distribution of power generation forecast error, is the empirical distribution of power generation forecast error, ε is the true value of power generation forecast error, is the empirical value of power generation forecast error, is the Wasserstein distance between the true distribution and the empirical distribution of power generation forecast error, is the set of all probability distributions, τ is the radius of the fuzzy set based on Wasserstein distance, is a fuzzy set under Wasserstein distance, which is based on A sphere in the probability distribution space with centered and τ as radius.

[0039] Optionally, the DRO-W coal blending model is as follows:

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] Among them, Cp id is the cost of the i-th type of coal on day d, x id is the decision variable for coal quantity selection, Car i is the carbon content per unit calorific value of the i-th coal, OF i is the carbon oxidation rate of the i-th coal, Q de is the power supply of the unit on day d, B e is the power supply reference value of the unit category, F1 is the unit cooling method correction coefficient, F r is the unit heat supply correction factor, F f is the unit load factor correction factor, l d is the actual power generation value, is the predicted power generation value, γ is the conversion coefficient between power generation and calorific value, η is the power generation efficiency of the unit, and q id is the calorific value of the i-th type of coal on the d-th day, MU s is the upper limit of coal type characteristics s, ML s is the lower limit of coal type characteristics s, M ids is the value of the characteristic s of the i-th type of coal on the d-th day, NL i is the upper limit of the quantity of the i-th type of coal, NU i is the lower limit of the quantity of the i-th type of coal, PL i is the upper limit of the proportion of the i-th coal, PU i is the lower limit of the proportion of the i-th coal, r To select the upper limit of the total number of coal types, is the lower limit of the total number of selected coal types, N is the minimum limit of the amount of each selected coal type, M is a large number, y id is the decision variable for coal type selection, ε d is the power generation forecast error on day d.

[0052] Optionally, the constraints of the DRO-W coal blending model are:

[0053]

[0054] Among them, β is the Lagrange multiplier, and τ is the radius of the fuzzy set based on the Wasserstein distance.

[0055] Compared with the prior art, the present invention has the following advantages and technical effects:

[0056] The present invention adopts LSTM network and machine learning method to predict power generation. Compared with the traditional LSTM method, the prediction accuracy is improved and the prediction is more accurate.

[0057] The coal blending optimization method provided by the present invention not only reduces the total cost of power generation, including coal burning cost and carbon trading cost, but also reduces carbon emissions and carbon trading costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0059] Figure 1 A schematic diagram of the optimization method flow of an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of a method for predicting power generation according to an embodiment of the present invention;

[0061] Figure 3 The power generation prediction effect diagram of the embodiment of the present invention, wherein (a) is the power generation prediction effect diagram of the LSTM prediction model; (b) is the power generation prediction effect diagram of the random forest prediction model; (c) is the power generation prediction effect diagram of the gradient boosting decision tree prediction model; (d) is the power generation prediction effect diagram of the extreme gradient boosting prediction model;

[0062] Figure 4 is an empirical probability distribution diagram of power generation prediction error according to an embodiment of the present invention;

[0063] Figure 5 A total cost comparison diagram of the coal blending scheme according to an embodiment of the present invention;

[0064] Figure 6 The results of DRO-KL and DRO-M in the embodiment of the present invention are compared, wherein (a) is a comparison chart of coal costs; (b) is a comparison chart of carbon trading costs; and (c) is a comparison chart of carbon emissions;

[0065] Figure 7 Detailed flowchart of the optimization method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0067] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0068] Embodiment 1

[0069] like Figure 1 and Figure 7 As shown, the present invention provides a coal blending optimization method considering coal and carbon trading costs, the method comprising:

[0070] S1, i.e. step 1: obtaining parameters related to coal blending in thermal power plants by directly collecting from the database, the parameters including historical power generation information, coal types and their characteristics information, boiler safe combustion information, etc.;

[0071] The coal type and its characteristic information include coal type, mining location, coal price, coal inventory, low calorific value, received sulfur, dry ash-free volatile matter, received ash, fixed carbon, high calorific value, coal supply, etc.

[0072] Boiler safe combustion information includes the boiler's requirements for the calorific value, sulfur content, volatile matter, ash melting point, etc. of the mixed coal burned.

[0073] S2, i.e. step 2: Figure 2 As shown in the figure, based on the historical power generation data and combined with influencing factors such as weather and temperature, a power generation prediction method based on LSTM network and machine learning is constructed, which specifically includes the following parts:

[0074] (1) Preprocess the historical power generation data, and then normalize the historical power generation data (MinMaxScaler) to generate a regular input tensor;

[0075] (2) constructing an LSTM network containing several layers of neural units, the preprocessed data is propagated forward and backward through the LSTM network, and mapped through the ReLU linear rectification function to obtain the LSTM model prediction value; calculating the cost function of the LSTM model, the cost function of the LSTM model is the power generation prediction value output by the LSTM network The mean square error (MSE) of the corresponding true value (l) is:

[0076]

[0077] in, is the predicted value of power generation on the dth day by the LSTM network model, l d is the true value of power generation on day d.

[0078] (3) Calculate the mean absolute error (MAE) and the mean absolute percentage error (MAPE);

[0079]

[0080]

[0081] (4) Obtain data on weather, maximum temperature, and minimum temperature, factors that affect photovoltaic and wind power generation;

[0082] (5) Based on the power generation prediction value of LSTM and combined with influencing factors such as weather, maximum temperature, and minimum temperature, random forest regression (RF), gradient boosting decision tree (GBDT), and eXtreme Gradient Boosting (XGBoost) prediction models are constructed to predict power generation.

[0083] All LSTM models and RF, GBDT, and XGBoost models are implemented in the Python programming language. The LSTM-based power generation prediction model is implemented using the Keras deep learning library. The random forest (RF)-based power generation prediction model is implemented using the RandomForestRegression machine learning library; the gradient boosting decision tree (GBDT)-based power generation prediction model is implemented using the GradientBoostingRegressor machine learning library; the extreme gradient boosting (XGBoost)-based power generation prediction model is implemented using the XGBRegressor machine learning library.

[0084] Table 1 shows the prediction accuracy of power generation prediction by four models: LSTM prediction model, random forest prediction model (LSTM+RF), gradient boost decision tree prediction model (LSTM+GBDT), and extreme gradient boost prediction model (LSTM+XGBoost), mainly including power generation prediction value The mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) of the corresponding true value (l). Table 1 shows that the power generation forecasting method based on LSTM network and machine learning has higher prediction accuracy and more accurate prediction than the traditional LSTM method. Among them, LSTM+GBDT has the best prediction effect. Figure 3It can also be observed in the power generation prediction effect diagram. Therefore, in the subsequent DRO coal blending method of the present invention, the LSTM+GBDT model is used to predict the power generation prediction value.

[0085] Table 1

[0086]

[0087] S3, i.e. step 3: constructing a DRO coal blending optimization method considering coal and carbon trading costs, specifically including the following parts:

[0088] (1) The present invention establishes a deterministic coal blending model that takes into account coal combustion and carbon trading costs.

[0089] Decision variables: There are two decision variables involved in the coal blending model, namely x id and id Among them, y id is the decision variable for coal type selection, y id =1 means that the i-th coal is selected on the d-th day, otherwise y id =0. x id is the decision variable for coal quantity selection, that is, the quantity of the i-th type of coal selected on the d day.

[0090] Objective function: The objective function of the coal blending model is to minimize the total cost, which includes two parts: coal burning cost and carbon trading cost. (i) Cp id is the cost of the i-th type of coal on the d-th day, is the coal cost on day d. (ii) A de is the carbon quota for the dth day, calculated according to the Implementation Plan for the Total Setting and Allocation of National Carbon Emission Trading Quotas in 2021 and 2022 (Power Generation Industry), that is, A de =Q de ×B e ×F1×F r ×F f . Where Q de is the power supply of the unit on day d, B e is the power supply reference value of the unit category, F1 is the unit cooling method correction factor, F r is the unit heat supply correction factor, F f It is the unit load (output) factor correction factor. d is the carbon emission on day d, Car i is the carbon content per unit calorific value of the i-th type of coal, in tons of carbon per gigajoules (tC / GJ), OF i is the carbon oxidation rate of the ith coal, expressed in %.

[0091] It is expressed by the following formula:

[0092]

[0093] Constraints:

[0094] According to the actual power generation demand d , build burning x id The calorific value of coal meets the constraints of the calorific value corresponding to the power generation demand, and the constraints are: Among them, γ is the conversion coefficient between power generation and calorific value, η is the power generation efficiency of the unit, and q id is the calorific value of the i-th type of coal on the d-th day.

[0095] The boiler has certain requirements on the characteristics of coal, mainly including the restrictions on the calorific value, volatile matter and sulfur content of the mixed coal. The constraints are as follows: Among them, MU s ,ML s Respectively represent the upper and lower limits of coal type characteristics s, M ids is the value of the characteristic s of the i-th type of coal on the d-th day. s=1, 2, 3 represent the calorific value, volatile matter, and sulfur content respectively.

[0096] The boiler has a coal quantity requirement for each type of coal. The constraints are Among them NL i NU i are the upper and lower limits of the quantity of the i-th type of coal respectively.

[0097] The boiler has requirements for the ratio of each type of coal. The constraints are: Among them, PL i PU i They are the upper and lower limits of the proportion of the i-th type of coal respectively.

[0098] Due to the limitation of coal bunkers in power plants, the total number of coal types is generally limited. in r , To select the upper and lower limits of the total number of coal types.

[0099] The model needs to select the coal type y id With x id Get in touch, And there is a minimum limit on the type and quantity of coal selected, i.e. N is the minimum quantity limit for each selected coal type, and M is a very large number.

[0100] In summary, the deterministic coal blending model established by the present invention taking into account coal combustion and carbon trading costs is as follows:

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107]

[0108]

[0109]

[0110]

[0111] (2) The present invention establishes a DRO-KL coal blending model that takes into account coal combustion and carbon trading costs.

[0112] like Figure 4 As shown in the figure, the above deterministic coal allocation model considering carbon trading costs is constructed under the deterministic actual power generation demand, but the actual power generation demand l d Unknown, distribution is also unknown. The power generation value predicted by the LSTM+GBDT model in step 2 The power generation prediction error (ε) is the difference between the actual power generation value (l) and the predicted value The difference is Due to the influence of natural factors such as weather, temperature, and wind, there is uncertainty in the generation of new energy such as photovoltaic power generation and wind power generation, which leads to uncertainty in the load instructions issued to thermal power plants when the power grid performs load dispatch. The present invention proposes a data-driven fuzzy set of power generation prediction errors. Since KL divergence shows great advantages in predicting the uncertainty associated with weather changes, the present invention adopts Define the true distribution P and empirical distribution of power generation forecast error The KL divergence between .

[0113]

[0114] The present invention defines the fuzzy set of power generation prediction error under KL divergence:

[0115]

[0116] Where P is the true distribution of power generation forecast error, is the empirical distribution of power generation forecast error, ε is the true value of power generation forecast error, is the empirical value of power generation forecast error, is the true distribution P and the empirical distribution of power generation forecast error The KL divergence between them, δ is the empirical distribution of the distance The farthest KL divergence (radius), is the set of all probability distributions, is a fuzzy set under KL divergence.

[0117] Under the KL divergence fuzzy set of power generation forecast error, all distributed power generation forecast errors in the fuzzy set satisfy the worst case calorific value of the burned coal to meet the power generation demand. The constraints on power generation in the above deterministic coal blending model are satisfied. Replace the component blue stick constraint (DRO-KL constraint), that is but:

[0118]

[0119] The DRO-KL coal blending model established by the present invention considering coal burning and carbon trading costs is as follows:

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131] Among them, Cp id is the cost of the i-th type of coal on day d, x id is the decision variable for coal quantity selection, y id is the decision variable for coal type selection, Car i is the carbon content per unit calorific value of the i-th coal, OF i is the carbon oxidation rate of the i-th coal, Q de is the power supply of the unit on day d, Be is the power supply reference value of the unit category, F1 is the unit cooling method correction coefficient, F r is the unit heat supply correction factor, F f is the unit load factor correction factor, l d is the actual power generation value, is the predicted power generation value, γ is the conversion coefficient between power generation and calorific value, η is the power generation efficiency of the unit, and q id is the calorific value of the i-th type of coal on the d-th day, MU s is the upper limit of coal type characteristics s, ML s is the lower limit of coal type characteristics s, M ids is the value of the characteristic s of the i-th type of coal on the d-th day, NL i is the upper limit of the quantity of the i-th type of coal, NU i is the lower limit of the quantity of the i-th type of coal, PL i is the upper limit of the proportion of the i-th coal, PU i is the lower limit of the proportion of the i-th type of coal, r is the upper limit of the total number of selected coal types, is the lower limit of the total number of selected coal types, N is the minimum limit of the amount of each selected coal type, M is a large number, ε d is the power generation forecast error on day d.

[0132] The present invention uses the strong duality theory to re-describe the DRO-KL constraints established above as:

[0133]

[0134] According to historical data, the LSTM+GBDT model is used to predict power generation, and the empirical probability distribution of the prediction error is approximately normal distribution. Assuming that the power generation prediction error follows the normal distribution N(μ, σ 2 ), the DRO-KL constraint is equivalently converted to the following formula:

[0135] (3) The present invention establishes a DRO-W coal blending model that takes into account coal combustion and carbon trading costs.

[0136] Since the fuzzy set based on KL divergence only considers samples in historical data and does not consider situations outside historical data, and the fuzzy set based on Wasserstein distance better considers samples outside historical data, the present invention establishes a DRO-W coal blending model that considers carbon trading costs.

[0137] use Define the true distribution P and empirical distribution of power generation forecast error The Wasserstein distance between them.

[0138]

[0139] The present invention defines the fuzzy set of power generation prediction error under Wasserstein distance:

[0140]

[0141] Where P is the true distribution of power generation forecast error, is the empirical distribution of power generation forecast error, ε is the true value of power generation forecast error, is the empirical value of power generation forecast error, is the Wasserstein distance between the true distribution and the empirical distribution of power generation forecast error, is the set of all probability distributions, τ is the radius of the fuzzy set based on Wasserstein distance, is a fuzzy set under Wasserstein distance, which is based on A sphere in the probability distribution space with centered and τ as radius.

[0142] Similarly, the constraints on power generation in the deterministic coal blending model are Replace the component blue stick constraint (DRO-W constraint), that is

[0143]

[0144] The DRO-W coal blending model established by the present invention taking into account coal burning and carbon trading costs is as follows:

[0145]

[0146]

[0147]

[0148]

[0149]

[0150]

[0151]

[0152]

[0153]

[0154]

[0155]

[0156] Among them, Cp id is the cost of the i-th type of coal on day d, x id is the decision variable for coal quantity selection, Car i is the carbon content per unit calorific value of the i-th coal, OF i is the carbon oxidation rate of the i-th coal, Q de is the power supply of the unit on day d, B e is the power supply reference value of the unit category, F1 is the unit cooling method correction coefficient, F r is the unit heat supply correction factor, F f is the unit load factor correction factor, l d is the actual power generation value, is the predicted power generation value, γ is the conversion coefficient between power generation and calorific value, η is the power generation efficiency of the unit, and q id is the calorific value of the i-th type of coal on the d-th day, MU s is the upper limit of coal type characteristics s, ML s is the lower limit of coal type characteristics s, M ids is the value of the characteristic s of the i-th type of coal on the d-th day, NL i is the upper limit of the quantity of the i-th type of coal, NU i is the lower limit of the quantity of the i-th type of coal, PL i is the upper limit of the proportion of the i-th coal, PU i is the lower limit of the proportion of the i-th type of coal, r is the upper limit of the total number of selected coal types, is the lower limit of the total number of selected coal types, N is the minimum limit of the amount of each selected coal type, M is a large number, y id is the decision variable for coal type selection, ε d is the power generation forecast error on day d.

[0157] The present invention uses the strong duality theory to re-describe the DRO-W constraint established above as:

[0158]

[0159] Among them, β is the Lagrange multiplier, and τ is the radius of the fuzzy set based on the Wasserstein distance.

[0160] make Assume support set is convexly closed. For any fixed x, H(·, ε): = max k≤K H k (·, ε), forming the function H k (·, ε) is upper semicontinuous and concave in ε. The DRO-W constraint can be further transformed into the following deterministic constraint form.

[0161]

[0162]

[0163]

[0164]

[0165]

[0166] S4, i.e. step four: the present invention needs to calculate the above-mentioned coal blending optimization model.

[0167] The DRO-KL coal blending model and the DRO-W coal blending model are solved by algorithms to obtain the optimal solutions of the objective functions of the two models under constraints; the optimal solutions of the two models are compared to obtain the optimal coal blending scheme with low total cost, including the selected coal type and coal quantity.

[0168] The following is described using specific embodiments:

[0169] This embodiment uses the SIS system database of A coal-fired power plant to obtain parameters related to coal blending in thermal power plants, including historical power generation information, coal types and their characteristics information, and boiler safe combustion information. This embodiment uses the acquired historical power generation data to predict power generation, and substitutes parameter information such as coal types and their characteristics information and boiler safe combustion into the coal blending optimization model to obtain a coal blending scheme. The algorithm of the present invention optimizes the coal blending of the power plant for 21 days. The total cost comparison table of the blending scheme given by the present invention and the original blending scheme of the power plant is shown in Table 2. It can be found that the average daily cost savings of the scheme given by the DRO-KL model is about 402,000 yuan, and the average daily cost savings of the scheme given by the DRO-W model is about 476,000 yuan.

[0170] The solution results of the DRO-KL and DRO-W models proposed in this invention are compared with the current coal blending scheme of the enterprise. The total cost results of the coal blending scheme are as follows: Figure 5 As shown in the figure, the total cost (the sum of coal cost and carbon dioxide cost) of the coal blending schemes of the DRO-KL model and the DRO-W model is lower than the cost of the company's current scheme.

[0171] In addition, this embodiment found that the DRO-KL model has a higher cost than the DRO-W model, that is, the DRO-KL model is more conservative than the DRO-W model. The present invention further analyzes the performance of the DRO-KL and DRO-W models in terms of coal cost, carbon trading cost, and carbon emissions. Figure 6 , which also verifies that the DRO-KL model is more conservative than the DRO-W model.

[0172] Table 2

[0173]

[0174] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A coal blending optimization method considering coal and carbon trading costs, characterized in that: The following steps are involved: Acquire coal blending related parameters, wherein the related parameters at least include historical power generation information; Based on the historical power generation information, a power generation prediction model based on LSTM network and machine learning is constructed; Based on the power generation prediction model, a power generation prediction value is obtained, and then a power generation prediction error is obtained; Based on KL divergence and Wasserstein distance, fuzzy sets of power generation forecast error are constructed respectively; Based on the fuzzy sets under KL divergence and Wasserstein distance, the corresponding DRO-KL coal blending model and DRO-W coal blending model are constructed respectively. With the goal of minimizing coal burning and carbon trading costs, the optimal solutions of the DRO-KL coal blending model and the DRO-W coal blending model are solved respectively; Obtaining a corresponding optimal coal blending solution based on the optimal solution; The process of constructing a power generation prediction model includes: constructing an LSTM prediction model containing several layers of neural units, preprocessing the historical power generation information, inputting the preprocessed historical power generation information into the LSTM prediction model for processing, and obtaining the LSTM model power generation prediction value; based on the LSTM model power generation prediction value, constructing a random forest prediction model, a gradient boosting decision tree prediction model, and an extreme gradient boosting prediction model; comparing the prediction accuracy of the LSTM prediction model, the random forest prediction model, the gradient boosting decision tree prediction model, and the extreme gradient boosting prediction model, and using the gradient boosting decision tree prediction model with the highest prediction accuracy as the final power generation prediction model; The fuzzy set under the KL divergence is: in, is the true distribution of power generation forecast error, is the empirical distribution of power generation forecast error, is the true value of power generation forecast error, is the empirical value of power generation forecast error, is the true distribution of power generation forecast error Distribution with experience The KL divergence between The empirical distribution of distance The farthest KL divergence, is the set of all probability distributions, is a fuzzy set under KL divergence; The DRO-KL coal blending model is shown below: in, is the cost of the i-th type of coal on the d-th day, is the decision variable for coal quantity selection, is the decision variable for coal type selection, is the carbon content per unit calorific value of the i-th type of coal, is the carbon oxidation rate of the ith coal, is the power supply of the unit on day d, is the power supply reference value of the category to which the unit belongs. is the unit cooling method correction factor, is the unit heat supply correction factor, is the unit load factor correction factor, To predict the power generation value, is the conversion coefficient between power generation and calorific value, is the generating efficiency of the unit, is the calorific value of the i-th type of coal on the d-th day, Coal type characteristics The upper limit of Coal type characteristics The lower limit, is the value of the characteristic s of the i-th type of coal on the d-th day, is the upper limit of the quantity of the i-th type of coal, is the lower limit of the quantity of the i-th type of coal, is the upper limit of the proportion of the i-th coal, is the lower limit of the proportion of the i-th coal, To select the upper limit of the total number of coal types, To select the lower limit of the total number of coal types, The minimum coal quantity limit for each selected coal type, For very large numbers, is the power generation forecast error on day d; The fuzzy set under the Wasserstein distance is: in, is the true distribution of power generation forecast error, is the empirical distribution of power generation forecast error, is the true value of power generation forecast error, is the empirical value of power generation forecast error, is the Wasserstein distance between the true distribution and the empirical distribution of power generation forecast error, is the set of all probability distributions, is the radius of the Wasserstein distance fuzzy set, is a fuzzy set under Wasserstein distance, which is based on Centered on is a sphere in the probability distribution space with radius ; The DRO-W coal blending model is shown below: in, is the cost of the i-th type of coal on the d-th day, is the decision variable for coal quantity selection, is the carbon content per unit calorific value of the i-th type of coal, is the carbon oxidation rate of the ith coal, is the power supply of the unit on day d, is the power supply reference value of the category to which the unit belongs. is the unit cooling method correction factor, is the unit heat supply correction factor, is the unit load factor correction factor, To predict the power generation value, is the conversion coefficient between power generation and calorific value, is the generating efficiency of the unit, is the calorific value of the i-th type of coal on the d-th day, Coal type characteristics The upper limit of Coal type characteristics The lower limit, is the value of the characteristic s of the i-th type of coal on the d-th day, is the upper limit of the quantity of the i-th type of coal, is the lower limit of the quantity of the i-th type of coal, is the upper limit of the proportion of the i-th coal, is the lower limit of the proportion of the i-th coal, To select the upper limit of the total number of coal types, To select the lower limit of the total number of coal types, The minimum coal quantity limit for each selected coal type, For very large numbers, is the decision variable for coal type selection, is the power generation forecast error on day d.

2. The coal blending optimization method considering coal and carbon trading costs according to claim 1 is characterized in that: The relevant parameters also include: coal types and their characteristic information and boiler safe combustion information; wherein, the coal types and their characteristic information include coal types, mining locations, coal prices, coal inventory, low calorific value, received basis sulfur, dry ash-free basis volatile matter, received basis ash, fixed carbon, high calorific value and coal supply conditions; the boiler safe combustion information includes the boiler's requirements for the calorific value, sulfur content, volatile matter and ash melting point of the mixed coal.

3. The coal blending optimization method considering coal and carbon trading costs according to claim 1 is characterized in that: The constraints of the DRO-KL coal blending model are: in, To define the radius of the KL divergence fuzzy set, is the Lagrange multiplier, is the calorific value of the i-th type of coal on the d-th day.

4. The coal blending optimization method considering coal and carbon trading costs according to claim 1 is characterized in that: The constraints of the DRO-W coal blending model are: in, is the Lagrange multiplier, is the radius of the fuzzy set based on Wasserstein distance, is the calorific value of the i-th type of coal on the d-th day.

Images