A Data-Driven Causal Inference-Based Decision Support Method for Low-Carbon Power Generation Operation

Abstract

This invention discloses a data-driven causal inference-based auxiliary decision-making method for low-carbon operation of power generation, comprising the following steps: (1) collecting and preprocessing data from thermal power plants; (2) constructing a causal graph reflecting the relationship between variables; (3) identifying control variables in the causal inference; (4) using a matching algorithm to calculate and filter matching values, thereby obtaining causal inference functions for intervention variables and outcome variables; and (5) evaluating the sensitivity of low-carbon operation control of thermal power plants based on the causal inference function obtained in step (4). This invention constructs a causal graph, fits the relationship between operation control quantities and carbon emissions based on data-driven causal inference, performs sensitivity analysis on carbon emission levels, and thus achieves reasonable control of carbon emissions, thereby realizing auxiliary decision-making for low-carbon operation of thermal power plants.

Description

Technical Field

[0001] This invention relates to a carbon emission monitoring and analysis technology, and more particularly to a data-driven causal inference-based auxiliary decision-making method for low-carbon operation of power generation. Background Technology

[0002] In 2021, the first compliance cycle for the power generation industry in the national carbon market officially commenced, with thermal power being the only industry subject to emission controls. Currently, emission control efforts are gradually intensifying, further amplifying the impact of low-carbon operation on the economic benefits of thermal power plants. However, numerous factors influence the carbon emission levels of thermal power plants. Besides objective factors such as technology, these include unit characteristics, fuel, and load factor. It is difficult for thermal power plant staff to identify the optimal operating point for unit emissions, hindering low-carbon operation and consequently increasing operating costs and reducing economic efficiency. Therefore, under given conditions, achieving reasonable control of carbon emissions from thermal power plants has become a crucial factor affecting their economic viability. Summary of the Invention

[0003] Purpose of the invention: The present invention aims to provide a data-driven causal inference-based auxiliary decision-making method for low-carbon operation of power generation, which can perform sensitivity analysis on carbon emission levels and thus achieve reasonable control of carbon emissions.

[0004] Technical solution: The present invention provides a data-driven causal inference-based auxiliary decision-making method for low-carbon operation of power generation, comprising the following steps:

[0005] (1) Collect data from thermal power plants and perform preprocessing;

[0006] (2) Construct a causal graph that reflects the relationships between variables;

[0007] (3) Identify control variables in causal inference;

[0008] (4) The matching algorithm is used to calculate the matching value and filter it, thereby obtaining the causal inference function of the intervention variable and the outcome variable;

[0009] (5) Evaluate the sensitivity of low-carbon operation control of thermal power plants based on the causal inference function in step (4).

[0010] Preferably, the thermal power plant data in step (1) includes power supply, heat supply, total heat input, carbon content per unit calorific value, and carbon emissions.

[0011] Preferably, step (2) includes constructing a directed acyclic graph reflecting the causal relationship between operating control quantities and carbon emissions based on prior knowledge of boiler thermal dynamics principles and power generation emission principles.

[0012] Preferably, the operation control quantities include power supply and heat supply.

[0013] Preferably, the process of identifying control variables in step (3) includes: first, considering the backdoor criterion of the causal graph method; if a variable satisfies the backdoor criterion in the causal graph, then the variable is defined as a confounding variable and selected as the control variable, thus completing the determination of the control variable; when no variable satisfies the backdoor criterion, considering the frontdoor criterion of the causal graph method; if a variable satisfies the frontdoor criterion in the causal graph, then the variable is defined as a mediator variable and selected as the control variable, thus completing the determination of the control variable; when neither the backdoor criterion nor the frontdoor criterion is satisfied, then other variables besides the intervention variable and the outcome variable are selected as control variables.

[0014] Preferably, step (4) includes:

[0015] (4.1) A matching algorithm was used to obtain the matching value for each intervention variable sample;

[0016] (4.2) Based on the matching value P of different samples i Calculate its exit probability d(P) i ),

[0017] d(P i )=0.5+0.5×P i ×log2(P i )+0.5×(1-P i )×log2(1-P i )

[0018] The lower the probability of withdrawal, the more reliable the sample is, and it will be retained; the higher the probability of withdrawal, the less reliable the sample is, and it will be removed.

[0019] (4.3) Utilize the selected intervention variable sample values ​​and their matching values ​​P i A conditional expectation function is constructed to calculate the outcome variable, and the parameters in the function expression are calculated using the least squares method, thereby obtaining the causal inference relationship expression.

[0020] Preferably, the intervention variable is electricity supply, and the outcome variable is carbon emissions; the causal inference expression is a single-valued function of carbon emissions E(q) and electricity supply q, i.e.

[0021] E[E(q)]=K0(M i β)+K1(M i β)q+K2(M i β)q 2

[0022] In the formula, the values ​​of parameters K0, K1, and K2 are shown in the following formulas:

[0023]

[0024] Preferably, the assessment of the sensitivity of low-carbon operation control of thermal power plants in step (5) includes estimating the changes in carbon emissions of thermal power plants based on the causal inference relationship function between the intervention variable power supply and the outcome variable carbon emissions, so as to take measures in advance to assist decision-making for low-carbon operation of thermal power plants.

[0025] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. By constructing a causal graph through prior knowledge, the relationship between the operation control quantity and carbon emissions is fitted based on the matching model, and the carbon emissions of the unit are predicted, so as to realize the auxiliary decision-making for the low-carbon operation of thermal power plants; 2. By screening the intervention variable samples and their matching values, unreliable samples that are prone to errors are eliminated, which can improve the fitting accuracy of the causal inference function. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method of the present invention;

[0027] Figure 2 This invention relates to a causal graph constructed based on prior knowledge.

[0028] Figure 3 This is a backdoor path diagram for the present invention;

[0029] Figure 4 This is the front door path diagram of the present invention;

[0030] Figure 5 This is a comparison chart of the fitting curve of the causal inference function of the present invention and the traditional fitting curve;

[0031] Figure 6 This is a distribution diagram of the matching values ​​of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0033] like Figure 1 As shown, the present invention provides a data-driven causal inference-based auxiliary decision-making method for low-carbon operation of power generation, comprising the following steps:

[0034] (1) Data collection and preprocessing of thermal power plants.

[0035] To analyze the relationship between the operational control variables of thermal power plants and their carbon emissions, electrical, thermal, and carbon parameters such as power supply, heat supply, total heat input, carbon content per unit calorific value, and carbon emissions are considered. Power supply and heat supply parameters are directly collected, and the total heat input Q... h The calculation is performed using fuel consumption and fuel calorific value parameters, as shown in equation (1).

[0036]

[0037] Among them, Q h FC represents the total heat input, k is the code for the type of fossil fuel. k For the consumption of the kth type of fossil fuel, NCV k Let be the daily average lower heating value of the kth type of fossil fuel.

[0038] The average unit calorific value of fuel and its carbon content CC can be calculated using equation (2) as follows.

[0039]

[0040] Where CC is the average carbon content per unit calorific value of the fuel, and Q is... h FC represents the total heat input, k is the code for the type of fossil fuel. k For the consumption of the kth type of fossil fuel, NCV k Let CC be the daily average lower heating value of the kth fossil fuel. k Let be the carbon content per unit calorific value of the k-th fossil fuel.

[0041] Carbon emissions E can be calculated using three methods: actual measurement of CO2 in flue gas, indirect measurement based on O2 concentration in flue gas, and calculation based on fuel consumption.

[0042] The measured CO2 emissions from flue gas are calculated based on flue gas concentration, flow rate and time, as shown in equation (3).

[0043]

[0044] Where E is the carbon dioxide emissions during the statistical time t, and C CO3 V represents the average volume concentration of carbon dioxide in the flue gas over time t, V represents the average flue gas velocity over time t, t represents the statistical time, and 509 indicates that the volume of 1 ton of CO2 is 509 m³. 3 .

[0045] The carbon emission is indirectly measured based on the O2 concentration in flue gas. This is based on the relationship between the oxygen concentration and the CO2 concentration in the flue gas, and the calculation formula is shown in equation (4).

[0046]

[0047] Among them, C CO2max C represents the maximum volume concentration of CO2 produced by fuel combustion. O2 V represents the average oxygen concentration over time t, V represents the average flue gas velocity over time t, t represents the statistical time, and 509 indicates that the volume of 1 ton of CO2 is 509 m³. 3 .

[0048] If the CO2 volume concentration and oxygen concentration cannot be obtained, they can be calculated based on fuel consumption, as shown in equation (5).

[0049]

[0050] Where k is the code for fossil fuel type, FC k For the consumption of the kth type of fossil fuel, OF k Let NCV be the carbon oxidation rate of the k-th fossil fuel. k Let CC be the daily average lower heating value of the kth fossil fuel. k Let be the carbon content per unit calorific value of the k-th fossil fuel. It is the ratio of the relative molecular masses of carbon dioxide to carbon.

[0051] Before further analysis, the intervention variable is normalized. In this embodiment, the intervention variable is the daily power supply. The theoretical maximum daily power supply capacity Q of the thermal power plant is used as the standard. max Based on the actual daily power supply Q a Standardize it as shown in equation (6).

[0052]

[0053] Where Q is the per-unit daily electricity supply, Q a Q represents the actual daily power supply. max C represents the maximum daily power supply capacity, C represents the installed capacity of the thermal power unit, and t represents the daily operating time of the thermal power unit.

[0054] (2) Construct a causal graph that reflects the relationship between variables.

[0055] Based on prior knowledge such as boiler thermal principles and power generation principles, a cause-and-effect diagram reflecting the relationship between operational control variables and carbon emissions is constructed, such as... Figure 2 As shown, the operational control variables include power supply and heat supply. First, the resulting cause-effect graph must be a directed acyclic graph (DAG). If closed loops exist in the graph, it cannot be analyzed using existing cause-effect graph knowledge. Second, parameters that cannot be directly measured, such as boiler efficiency, and potential unobservable factors also need to be considered when constructing the cause-effect graph to facilitate the identification of control variables in subsequent steps.

[0056] (3) Identification of control variables in causal inference.

[0057] On the causal graph constructed in step (2), intervention variables and outcome variables are defined, and control variables are identified based on the backdoor criterion and frontdoor criterion in the causal graph method.

[0058] The power supply Q from the thermal power plant on day i. iDefined as an intervention variable, the carbon emissions E on day i. i Defined as an outcome variable. In a cause-and-effect graph, from Q... i To E i Of all the paths, there is an arrow pointing to Q. i A path where not all arrows point in the same direction is called a backdoor path, such as... Figure 3 As shown. From Q i To E i Of all the paths, from Q i Emitted, not directly pointing to E i A path in which all arrows point in the same direction is called a front door path, such as... Figure 4 As shown.

[0059] Therefore, the process of identifying control variables includes:

[0060] First, according to the backdoor criterion of the cause-effect graph method, if there exists a variable W1 in the cause-effect graph located at Q... i To E i In the backdoor path, variable W1 is defined as a promiscuous variable, and this promiscuous variable is selected as the control variable to complete the determination of the control variable.

[0061] If there are no confounding variables, i.e. from Q i To E i If no backdoor path exists, then the frontdoor criterion of the cause-effect graph method is considered. If there exists a variable W2 in the cause-effect graph located at the junction of Q and Q, then the frontdoor criterion is considered. i To E i In the front door path, and Q i To W2, W2 to E i When there is no backdoor path between them, variable W2 is defined as an intermediary variable, and this intermediary variable is selected as the control variable to complete the determination of the control variable.

[0062] If there are neither confounding variables nor mediating variables, that is, if no variables in the causal diagram satisfy the backdoor criterion and the frontdoor criterion, then other variables besides the intervention variable and the outcome variable need to be considered comprehensively.

[0063] Depend on Figure 2 As can be seen from the above, there is no confounding variable W1 in this embodiment. Therefore, the intermediate variable W2, the total heat input, is determined by the front door path. Thus, the total heat input is selected as the control variable, and the determination of the control variable is completed.

[0064] (4) Using a matching algorithm to identify causal effects, specifically including:

[0065] (4.1) A matching algorithm was used to obtain the matching value for each intervention variable sample;

[0066] Define the vector of control variables as the matching variable and name it M.i For matching variable M i Calculate the intervention variable sample, i.e., each power supply sample Q. i Matching value P i Since the only control variable in this example is the total heat input, M in this example... i This is the value input for total heat.

[0067]

[0068]

[0069] 0 < Q i <1 (9)

[0070] In the formula, P i For intervention variable Q i and matching variable M i The matching value, where β is the matching parameter.

[0071] To facilitate the calculation of β, equation (8) is transformed to obtain equation (10):

[0072]

[0073] The least squares method is used to solve equation (10) to obtain β, and then β is substituted into equation (7) to obtain the matching value P. i .

[0074] (4.2) Screening of intervention variable samples and their matching values

[0075] To mitigate the impact of selection bias, further filtering of the matching values ​​is necessary. The sample set is regularized using the matching values ​​of each sample; that is, for all samples, different matching values ​​P are used to regularize the sample set. i Calculate its exit probability d(P) i The exit probability, i.e., the probability of discarding the corresponding sample, is shown in equation (11):

[0076] d(P i )=0.5+0.5×P i ×log2(P i )+0.5×(1-P i )×log2(1-P i (11)

[0077] Specifically, when the matching value is 0 or 1, the exit probability is 0.5; while when the matching value is close to 0.5, the exit probability is close to 0. That is, the lower the exit probability of a sample's matching value, the more reliable the sample is, and it is retained; the higher the exit probability of a sample's matching value, the less reliable the sample is, and it is discarded. This method applies a higher exit probability to samples with poor matching values, automatically removing them from the sample set and preventing normal samples from participating in the function fitting of subsequent steps, which would increase the error in the result.

[0078] (4.3) Utilize the selected intervention variable sample values ​​and their matching values ​​P i A conditional expectation function is constructed, and the parameters in the function expression are calculated using the least squares method, thereby obtaining the causal inference relationship expression.

[0079] First, the carbon emissions E are known from prior knowledge. i With power supply Q i There is a quadratic function relationship, therefore the power supply Q is used. i And the matching value P selected in step 4 i Construct a system for calculating carbon emissions E i The conditional expectation function is given by equation (12):

[0080]

[0081] In the formula, α0, α1, α2, α3, α4, and α5 are the coefficients of the functions relating carbon emissions and electricity supply that are to be solved.

[0082] The values ​​of coefficients α0, α1, α2, α3, α4, and α5 in equation (12) are obtained using the least squares method. Substituting the values ​​of coefficients α0, α1, α2, α3, α4, and α5 into the model of equation (13), the causal inference function of expected carbon emissions and electricity generation can be obtained:

[0083]

[0084] In the formula p i Let the power supply be q and the matching variable be M. i The matching value at that time.

[0085] Since the calculated relationship function between electricity supply and carbon emissions, E[E(q)], includes not only electricity supply q but also a matching value p as an independent variable. i Therefore, in order to obtain the single-valued functional relationship between carbon emissions E(q) and electricity supply q, we perform a Taylor series expansion on equation (8) to obtain equation (14) as follows:

[0086] E[E(q)]=K0(M i β)+K1(M i β)q+K2(Mi β)q 2 (14)

[0087] In the formula, the values ​​of K0, K1, and K2 are shown in equation (15):

[0088]

[0089] At this point, a causal inference function containing only the intervention variable (electricity supply) and the outcome variable (carbon emissions) is obtained. Building upon traditional matching methods, this further develops a single-valued functional expression between the intervention and outcome variables. Using the total boiler heat input as the matching variable, the fitted curve between electricity supply and carbon emissions is shown below. Figure 5 As shown, the corresponding distribution of matching values ​​is as follows: Figure 6 As shown. By Figure 5 It can be seen that the fitting curve obtained by this invention shows a quadratic function relationship between power supply and carbon emissions, which is more consistent with the actual relationship between the two in the operation of thermal power plants and is superior to traditional fitting methods.

[0090] (5) Evaluation of the sensitivity of low-carbon operation control in thermal power plants.

[0091] Based on the single-valued function relationship between carbon emissions and power supply in step (4), the carbon emissions of a thermal power plant under a certain total heat input can be estimated, providing support for low-carbon operation decision-making of thermal power plants.

[0092] If the power supply intensity is temporarily changed while keeping the total heat input of the boiler constant, the possible change in carbon emissions can be estimated in advance according to formula (14), so that corresponding measures can be taken in advance to control carbon emissions, reduce the operating cost of the power plant, and achieve the goal of low-carbon operation auxiliary decision-making for thermal power plants.

[0093] Consider matching variable M i With the power supply remaining constant, what happens if the power supply increases or decreases by Δq?

[0094] E[E(q±Δq)]=K0(M i β)+K1(M i β)(q±Δq)+K2(M i β)(q±Δq) 2 (16)

[0095] Taking the difference between equation (14) and equation (16), we get:

[0096] E[E(q±Δq)]-E[E(q)]=±[K1(M i β)+2qK2(M i β)]Δq+K2(M i β)Δq 2 (17)

[0097] When the power supply increases by Δq, the carbon emissions are expected to increase by [K1(M]]. i β)+2qK2(M i β)]Δq+K2(M i β)Δq 2 When the power supply decreases by Δq, carbon emissions are expected to decrease by [K1(M]]. i β)+2qk2(M I β)]Δq-K2(M i β)Δq 2 .

[0098] Since various operational control parameters of thermal power plants, such as power supply and heating supply, are closely related to carbon emissions, constructing a causal graph and fitting the relationship between operational control parameters and carbon emissions based on data-driven causal inference, conducting sensitivity analysis on carbon emission levels, and thus achieving reasonable control of carbon emissions, is a practical and effective auxiliary decision for low-carbon operation of thermal power plants.

Claims

1. A data-driven causal inference-based auxiliary decision-making method for low-carbon operation of power generation, characterized in that, Includes the following steps: (1) Collect data from thermal power plants and perform preprocessing; (2) Construct a causal graph reflecting the relationships between variables; (3) Identify the control variables in causal inference; (4) The matching algorithm is used to calculate the matching value and filter it to obtain the causal inference function of the intervention variable and the outcome variable; Step (4) includes: (4.1) A matching algorithm was used to obtain the matching value for each intervention variable sample; Step (4.1) includes: Define the vector of control variables as the matching variable and name it... For matching variables Calculate the intervention variable sample Matching value : In the formula, Intervention variables and matching variables The matching value, For matching parameters; (4.2) Based on the matching values ​​of different samples Calculate an exit probability That is, the probability of discarding the corresponding sample: The lower the probability of withdrawal, the more reliable the sample is, and it will be retained; the higher the probability of withdrawal, the less reliable the sample is, and it will be removed. (4.3) Utilize the selected intervention variable sample values ​​and their matching values A conditional expectation function is constructed to calculate the outcome variable, and the parameters in the function expression are calculated using the least squares method, thereby obtaining the causal inference relationship expression; (5) Evaluate the sensitivity of low-carbon operation control of thermal power plants based on the causal inference function in step (4).

2. The power generation low-carbon operation auxiliary decision-making method according to claim 1, characterized in that, The data for thermal power plants in step (1) include power supply, heat supply, total heat input, carbon content per unit calorific value, and carbon emissions.

3. The power generation low-carbon operation auxiliary decision-making method according to claim 2, characterized in that, Step (2) includes constructing a directed acyclic graph that reflects the causal relationship between operating control quantities and carbon emissions based on prior knowledge of boiler thermal dynamics principles and power generation emission principles.

4. The power generation low-carbon operation auxiliary decision-making method according to claim 3, characterized in that, The operational control quantities include power supply and heat supply.

5. The power generation low-carbon operation auxiliary decision-making method according to claim 1, characterized in that, The process of identifying control variables in step (3) includes: first, considering the backdoor criterion of the causal graph method; if a variable satisfies the backdoor criterion in the causal graph, then the variable is defined as a confounding variable and selected as the control variable, thus completing the determination of the control variable; when no variable satisfies the backdoor criterion, considering the frontdoor criterion of the causal graph method; if a variable satisfies the frontdoor criterion in the causal graph, then the variable is defined as a mediator variable and selected as the control variable, thus completing the determination of the control variable; when no variable satisfies either the backdoor criterion or the frontdoor criterion, then other variables besides the intervention variable and the outcome variable are selected as control variables.

6. The power generation low-carbon operation auxiliary decision-making method according to claim 1, characterized in that, The intervention variable is electricity supply, and the outcome variable is carbon emissions; a system is constructed to calculate carbon emissions. The conditional expectation function is: In the formula , , , , , The coefficients of each term in the function relating carbon emissions and electricity consumption are given; thus, the causal relationship expression is obtained as carbon emissions. With power supply A single-valued function, i.e. In the formula, the parameter , , The values ​​are shown in the following formulas: 。 7. The power generation low-carbon operation auxiliary decision-making method according to claim 1, characterized in that, The assessment of the sensitivity of low-carbon operation control of thermal power plants in step (5) includes estimating the changes in carbon emissions of thermal power plants based on the causal inference relationship function between the intervention variable power supply and the outcome variable carbon emissions, so as to take measures in advance and assist decision-making for low-carbon operation of thermal power plants.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1-7.

Images