A sulfur constraint boundary dynamic adjustment method based on an open-loop coal blending optimization model

Abstract

A dynamic adjustment method for sulfur constraint boundary based on an open-loop coal blending optimization model is proposed. The objective function is to dynamically adjust the sulfur constraint boundary based on the minimum coal blending cost to meet boiler combustion requirements. The coal blending optimization model is Equation (1): In Equation (1): minf(x) is the minimum cost of coal blending; C i x is the cost of the i-th single coal; i b is the blending ratio of the i-th single coal; i ΔS represents the virtual coal coefficient and the dynamic adjustment of the sulfur constraint limit. This invention guides future coal blending based on the current combustion status of the boiler, building upon open-loop coal blending. It focuses on online monitoring of sulfur dioxide emissions and real-time load in boiler combustion. Through fuzzy modeling and adaptive decision-making, it continuously compensates for the sulfur constraint upper bound of the open-loop optimization model. By applying negative or positive biases to the sulfur constraint upper bound, it aims to reduce boiler outlet sulfide emissions or improve the economic efficiency of coal blending, thus promoting the development of closed-loop coal blending. This invention has significant foreseeable economic, social, and environmental value.

Description

TECHNICAL FIELD

[0001] The present application relates to the design method of a coal blending optimization model for thermal power plants, in particular to a new method for dynamically adjusting the sulfur constraint boundary based on an open-loop coal blending optimization model. BACKGROUND

[0002] At present, the main power generation method in China is still "thermal power generation dominated by coal-fired power generation". According to incomplete statistics, thermal power accounts for about 73%, and strictly controlling the emission of harmful substances such as sulfides and nitrogen oxides by thermal power enterprises is of great significance to environmental protection and is an inherent requirement for building a new development pattern and promoting high-quality development.

[0003] Due to the influence of various factors such as limited coal resources, uneven distribution, rising prices, coal market supply situation, and the policy of using low-quality coal as much as possible for boiler and other combustion equipment in China, power plants cannot use single coal for power generation, and have to use mixed coal. The current coal blending of thermal power enterprises basically adopts open-loop coal blending, that is, coal blending before the furnace.

[0004] The sulfur constraint boundary of the open-loop coal blending optimization model is usually calculated based on the load and coal supply amount, but there is a lot of uncertainty in the combustion process of mixed coal in the boiler, which makes the sulfur emission at the outlet of the boiler high or low, affecting environmental protection or economy. Under the consideration of the uncertainty of boiler combustion, how to accurately design the sulfur constraint boundary based on the open-loop coal blending optimization model is a scientific problem.

[0005] Fuzzy set and fuzzy system methods are widely used because they can well embed expert knowledge to generate rules or mine rules from small sample data, mainly including one-type fuzzy and two-type fuzzy. In the early years, one-type fuzzy systems have been comprehensively developed in terms of modeling, control, data extraction rules, adaptive mechanisms, and stability. Two-type fuzzy systems have been rapidly developed in recent decades under the promotion of Professor Mendel of the University of Southern California. Its main feature is strong handling of uncertainty.

[0006] Chinese patent 202011285884.1 relates to a fuzzy comprehensive evaluation method for coal blending and blending based on an improved entropy weight method. This method is also for the evaluation and optimization scheme of power plant coal blending and blending. The method used is an improved entropy weight method, but it is still an open-loop coal blending method and does not realize dynamic adjustment of the open-loop coal blending scheme. It also does not form a closed-loop optimization combined with boiler feedback.

[0007] Coal-fired power plants and other coal-using enterprises urgently need a new method for dynamically adjusting the sulfur constraint boundary based on the open-loop coal blending optimization model with excellent technical effects. SUMMARY

[0008] The problem to be solved by this invention is to provide a dynamic adjustment method for sulfur constraint boundaries based on an open-loop coal blending optimization model, and further develop towards a closed-loop coal blending method with better technical performance.

[0009] This invention specifically relates to a method for dynamically adjusting the sulfur constraint boundary based on an open-loop coal blending optimization model.

[0010] This invention discloses a dynamic adjustment method for sulfur constraint boundary based on an open-loop coal blending optimization model. Its features include: balancing the environmental friendliness, economic efficiency, and practicality of power plants (considering the impact of boiler combustion uncertainties on environmental protection and economic efficiency); using the minimum coal blending cost to meet boiler combustion requirements as the objective function; and employing the calorific value, sulfur, moisture, and ash content of coal as constraints to dynamically adjust the sulfur constraint boundary based on the open-loop coal blending optimization model.

[0011] The established coal blending optimization model and its constraints are as follows:

[0012] (1)

[0013]

[0014] In equation (1) and its constraints: This represents the minimum cost for mixed coal; It is the first The cost, i.e., the price, of a single coal unit; It is the first The proportion of a single coal unit participating in coal blending; It refers to the number of coal types involved in the blending; For the first Coal feed rate of each coal mill; The number of coal mills or coal bunkers involved in coal blending; This is a virtual coal coefficient; when taking a virtual coal type... ,otherwise , This is the upper limit of the sulfur content of the blended coal; This is the dynamic adjustment amount for the upper limit of the sulfur constraint; , , These are the lower limit of the calorific value, and the upper limit of the moisture and ash content of the blended coal, respectively. , , , The first The sulfur content, calorific value, moisture content, and ash content of each coal type;

[0015] Specifically, the proportion of each coal type is restricted in the constraints of formula (1) in order to meet the requirements.

[0016] (1.3.1)

[0017] In the actual application of coal blending, each coal bunker is not more than two kinds of coal, that is, two kinds of coal or single coal;

[0018] If the "virtual coal" is not introduced, the solution of the optimization model can only be the mixing of two different coals under the constraint condition of satisfying formula (1.3.1), while also requiring to satisfy formula (1.2), and the case of single coal in the coal bunker is discarded;

[0019] Therefore, the "virtual coal" is introduced to solve the contradiction between the proportion limit and the discarded single coal during coal taking. First, a "virtual coal" is introduced in each coal yard, and then formula (1.3.1) is converted into formula (1.3) and (1.4). The introduction of the "virtual coal" solves the actual problem of enterprise coal blending and effectively reduces the coal blending cost.

[0020] An innovation different from the previous open-loop coal blending optimization model is that the "virtual coal" idea is first introduced, and the constraint condition with "virtual coal" is innovatively designed, that is, the unique design of formula (1.1), (1.2), (1.3) and (1.4).

[0021] The introduction of the "virtual coal" idea in the coal blending optimization model effectively solves the contradiction between the proportion limit and the discarded single coal. The introduction of the "virtual coal" idea also focuses on solving the practical difficulties that the coal with a proportion less than 0.3 cannot be accurately taken by the stacker-reclaimer during the actual production process of the coal user such as a power plant. It is obvious that the further extension and application of the "virtual coal" idea are more in line with the actual needs of the power plant. (Further description of the "virtual coal" idea and background will be given in the example below)

[0022] The sulfur constraint boundary dynamic adjustment method based on the open-loop coal blending optimization model also introduces the dynamic adjustment of the sulfur constraint boundary.

[0023] Compared with the previous traditional open-loop coal blending optimization model, the unique design of the sulfur constraint boundary dynamic adjustment method based on the open-loop coal blending optimization model is that the dynamic adjustment amount is added to the sulfur constraint boundary, so that the coal blending optimization model has a dynamic adjustment effect.

[0024] Since the optimization coal blending model of formula (1) is still open-loop in nature, and because the boiler combustion condition is not considered in advance when blending coal;

[0025] The new method of dynamic adjustment of the sulfur constraint boundary based on the open-loop coal blending optimization model realizes the dynamic adjustment of the sulfur constraint boundary by real-time monitoring of the sulfide at the outlet of the current batch of boiler combustion, which is expressed as follows:

[0026] (2)

[0027] In formula (2): is a new constraint boundary of sulfur in the mixed coal after dynamic adjustment; is a constraint boundary of sulfur in the mixed coal given by the traditional open-loop coal blending; is a constraint boundary of sulfur in the mixed coal after dynamic adjustment compensation;

[0028] Establishing The step of dynamically adjusting the constraint boundary of the additional sulfur after compensation meets the following requirements in turn:

[0029] Preamble

[0030] Step 1: Establishing a two-type fuzzy rule base of the fuzzy model;

[0031] Step 2: Establishing the output of the fuzzy model;

[0032] Step 3: Establishing a parameter vector learning rule of the fuzzy model;

[0033] Step 4: Establishing an uncertain coverage domain self-adaptive adjustment strategy of the fuzzy model.

[0034] Further preferred technical content claimed is:

[0035] In step 1, the rule base of the two-type fuzzy model meets the following requirements:

[0036] Firstly, the input and output variables and fuzzy sets of the fuzzy model are established:

[0037] Secondly, the rule base of the two-type fuzzy model is established based on small sample experimental data by data mining:

[0038] The selected input variables include two real-time parameter variables of the boiler, wherein: the first input variable is the real-time load of the boiler, represented as , with the unit of MW; the second input variable is the detected sulfur dioxide content at the outlet of the boiler, represented as , with the unit of mg / Nm 3 ; the selected output variable of the system is the dynamic deviation compensation of the coal blending mixed coal, represented as ;

[0039] The input premise variables are each divided into five fuzzy sets, represented as , and the fuzzy sets of the premise variables adopt Gaussian functions; the conclusion variable is also divided into five fuzzy sets, represented as​​​ This indicates that the fuzzy combination of the conclusion variable adopts a trigonometric function; where: the fuzzy set corresponding to the premise variable is a type II fuzzy set, and the fuzzy set corresponding to the conclusion variable is a type I fuzzy set;

[0040] Then, support is used as the discrimination criterion for calculating the fuzzy space, and this is applied to the experimental sample data collected on-site. , First, calculate the support for each rule, where the support for the upper uncertainty bound is:

[0041] (3)

[0042] The support of the lower uncertainty bound of the uncertain coverage area is:

[0043] (4)

[0044] In equations (3) and (4): , The first Prerequisite variables for each record The corresponding type II fuzzy set The upper and lower membership functions; For the first Conclusion variables of each record The corresponding type I fuzzy set Membership function; This indicates the total number of records in the sampled data. ;

[0045] The weighted average support of this rule is calculated using the following equation (5):

[0046] (5)

[0047] The fuzzy rule subspace with the highest support is selected to construct the following interval type-2 fuzzy logic system.

[0048] (6)

[0049] In equations (5) and (6): It is a type II fuzzy set; The output variable is a type I fuzzy set; The total number of fuzzy rules ( ); For the first A fuzzy rule;

[0050] After mining fuzzy rules based on sampled data, the final type II fuzzy rule library is formed.

[0051] After mining fuzzy rules based on sampled data, power plant experts, drawing on years of boiler operation experience, verify and fine-tune these rules to form the final Type II fuzzy rule library. Adjustments can be made based on experience and knowledge; alternatively, it can be integrated with case libraries, expert systems, and other supporting automatic control systems and programs for automated processing.

[0052] Establish in step 2 The output of the fuzzy model must meet the following requirements:

[0053] For a certain rule in equation (6), first calculate the input vector. Activation range:

[0054] (7)

[0055] For the Fuzzy rules, output variables The value corresponding to the maximum membership degree is:

[0056] (8)

[0057] In the formula: It is the first Rules in fuzzy membership functions The corresponding value when the maximum value is obtained Point value, that is Time corresponding Point value;

[0058] In the preceding activation region and the successor After the output calculation, the reduced form based on the KM algorithm can be represented as follows:

[0059] (9)

[0060] In the formula: The minimum and maximum output intervals obtained using the KM algorithm; , The first one calculated using the KM algorithm The minimum and maximum activation weights corresponding to the fuzzy rules;

[0061] After weighted average defuzzification, the above interval type II fuzzy system can be written in the form of the following basis functions:

[0062] (10)

[0063] In the formula: The adaptive parameter vector of the type-II fuzzy system; the basis function vectors formed by the activation weights are respectively , .

[0064] Step 3 meets the following requirements:

[0065] which is based on small sample data using gradient descent algorithm to establish the parameter vector learning of the bivalent fuzzy model;

[0066] Let The number of input-output sample data pairs is , the current input variable is The predicted output of the bivalent fuzzy model is , and the target real output is

[0067] (11)

[0068] The gradient descent algorithm is used to minimize the above target error function, thereby dynamically optimizing and adjusting the parameters of the fuzzy model, and improving the modeling accuracy of the fuzzy model;

[0069] Therefore, according to the gradient descent method, the update rule of the parameter is as follows:

[0070] (12)

[0071] In the formula: is the number of iterations; is a positive learning rate.

[0072] Step 4 meets the following requirements:

[0073] which is based on setting ideal compensation using the gradient descent algorithm to adaptively adjust the uncertain coverage domain LMF of the bivalent fuzzy model:

[0074] The uncertain coverage domain UMF of the input variable is represented by the following Gaussian function

[0075] (13)

[0076] In the formula: the parameters of UMF are , , which are specified and no longer adjusted, and the LMF of the uncertain coverage domain is represented by the following Gaussian function:

[0077] (14)

[0078] In the formula: the parameters of LMF are , As the UMF, the LMF is adjusted by The size of the uncertain coverage area is changed;

[0079] Due to the uncertainty of the boiler combustion, the LMF is established The adaptive adjustment strategy of the parameters is as follows: under the condition that the actually detected sulfur dioxide and the expected sulfur dioxide meet certain conditions, the parameters of the LMF are dynamically adjusted

[0080] (15)

[0081] (16)

[0082] In formulas (15) and (16): is the model output; is a positive learning rate; is the expected value of the experimental sulfur dioxide corresponding to the compensation amount.

[0083] The schematic diagram of the above process is shown in Figure 1 and Figure 2 .

[0084] Other related contents of the present application are described as follows:

[0085] Compared with the prior art, the main advantages of the sulfur constraint boundary dynamic adjustment method based on the open-loop coal blending optimization model are as follows:

[0086] 1. The real-time boiler combustion problem is considered, and the upper bound of the sulfur constraint of the next round of open-loop optimization coal blending is compensated by online parameter monitoring, modeling and decision making of the boiler combustion state. Compared with the traditional open-loop coal blending method, the emission of sulfides can be reduced through negative bias compensation; under the premise that the sulfides are far below the standard, the economic efficiency of the coal blending scheme can be improved through appropriate positive bias .

[0087] 2. The two-type fuzzy method is used to establish the system model, which can better handle the uncertainty generated by the boiler combustion system compared with the traditional one-type fuzzy method, so that the bias calculation is more accurate.

[0088] 3. In the design of the two-type fuzzy system, an adaptive mechanism is used to change the uncertain coverage area of the fuzzy set, which further improves the accuracy of the dynamic modeling of the two-type fuzzy system compared with the traditional fixed uncertain coverage area method.

[0089] The main innovations of the present application are four:

[0090] 1. The concept of "virtual coal type" is introduced for the first time in the constraint condition section of the coal blending optimization model. Its main advantage is that it can better solve the constraint conflict between the proportional tonnage limit of the stacker-reclaimer in the coal yard and the single coal type in the coal bunker, thus saving the enterprise coal blending cost.

[0091] 2. In order to achieve dynamic adjustment of open-loop coal blending, a method for dynamically adjusting the sulfur constraint of the open-loop optimization model is proposed for the first time. Its main purpose is to promote the development of open-loop coal blending towards closed-loop coal blending, taking into account both the economic and environmental benefits of coal blending and combustion.

[0092] 3. To address the drawback of open-loop coal blending not considering the uncertainties of boiler combustion, this paper proposes for the first time to use a type-two fuzzy system to dynamically adjust the coal blending optimization model. A fuzzy rule base is constructed by combining data mining and expert domain knowledge, making the dynamically adjusted model more accurate. Its main purpose is to improve the accuracy of the open-loop coal blending optimization model and to take into account the impact of boiler combustion uncertainties.

[0093] 4. In the process of using the Type II fuzzy system for dynamic coal blending adjustment, an adaptive adjustment of the uncertain coverage domain of the Type II fuzzy system was proposed for the first time, which solved the problem that the traditional fixed coverage domain Type II fuzzy system does not have adaptive capability, and further improved the accuracy of dynamic modeling of the Type II fuzzy system.

[0094] In existing technologies, open-loop coal blending in power plants involves inputting information such as coal quality parameters, planned load, and coal mill parameters of each individual coal without considering boiler combustion conditions. Based on an optimization model, a coal blending output plan is formulated for boiler co-firing in future periods. This is pre-furnace coal blending.

[0095] The closed-loop coal blending proposed in this invention, applicable to power plants, builds upon open-loop coal blending by incorporating the current boiler combustion status to guide future coal blending. Addressing the drawback of currently widely used open-loop coal blending, which does not consider real-time boiler combustion conditions, this invention employs online monitoring of sulfur dioxide and real-time load parameters related to boiler combustion, fuzzy modeling, and adaptive decision-making with rolling compensation to establish a sulfur constraint upper bound in the open-loop optimization model. This overcomes the shortcomings of neglecting boiler combustion conditions by applying a negative bias to the sulfur constraint upper bound to reduce boiler outlet sulfide emissions and a positive bias to improve coal blending economics, thus driving the development of currently used open-loop coal blending towards closed-loop blending. This invention has significant foreseeable economic, social, and environmental value.

[0096] In summary, this invention has considerable foreseeable economic and social value.

[0097] Figure and Table Description

[0098] Figure 1 A schematic diagram illustrating the principle of the dynamic adjustment method for sulfur constraint boundaries in type II fuzzy modeling;

[0099] Figure 2 A schematic diagram illustrating the principle of adaptively and dynamically adjusting the lower bound of the uncertain coverage area;

[0100] Figure 3 This is a simplified schematic diagram illustrating the overall principle of the dynamic adjustment method for sulfur constraint boundaries based on an open-loop coal blending optimization model.

[0101] Figure 4 This is a simplified schematic diagram illustrating the overall principle of the dynamic adjustment method for sulfur constraint boundaries based on an open-loop coal blending optimization model.

[0102] Figure 5 The universe of discourse for the real-time load, the input variable in Example 1, is: With the corresponding membership function;

[0103] Figure 6 The domain of discourse for the input variable sulfur dioxide in Example 1 is: With the corresponding membership function;

[0104] Figure 7 For the output variables in Example 1 The domain of discourse is set as With the corresponding membership function;

[0105] Figure 8 One of the simplified flowcharts for solving a dynamic coal blending optimization model using the two-stage simplex method;

[0106] Figure 9 The second simplified flowchart for solving the dynamic coal blending optimization model using the two-stage simplex method. Detailed Implementation

[0107] Example 1: A method for dynamic adjustment of sulfur constraint boundary based on an open-loop coal blending optimization model

[0108] Reference Figure 3 The process, specifically the steps in this embodiment, is as follows:

[0109] I. Calculation of open-loop coal blending content value

[0110] in accordance with Figure 4 The software automatically calculates the planned load for power generation;

[0111] The open-loop coal blending was calculated based on the coal mill feed rate and the desulfurization capacity of each boiler, combined with Table 1. The value. For example, in this case, the desulfurization capacity of a certain boiler is 1T / H, and the total coal consumption is 125 T / H. According to Table 1, the value can be calculated. .

[0112] Table 1 Summary of Sulfur Constraints Calculated Theoretically by Open-Loop Coal Blending

[0113]

[0114] II. Calculation of dynamic adjustment of the value

[0115] This part is the core part of the invention, the specific process is as follows:

[0116] (1) to establish the input and output variables of fuzzy model and fuzzy set

[0117] Select the input variable including two boiler real-time parameter variables, wherein: the first input variable is the real-time load of the boiler, represented as , unit: MW; the second input variable is the detected sulfur dioxide content at the outlet of the boiler, represented as , unit: mg / Nm3; the output variable of the system is selected as the dynamic deviation compensation of coal blending and mixing, represented as .

[0118] The domain of the input variables of this example, real-time load and sulfur dioxide, is respectively set as and , the domain of the output variable is set as , and the corresponding membership functions are shown in Figure 5 , Figure 6 , Figure 7 .

[0119] (2) to establish the two-type fuzzy rule base of fuzzy model

[0120] Take the support degree as the criterion for calculating the fuzzy space, and according to the test sample data collected on site, , , first calculate the support degree of each rule, wherein the support degree of the lower uncertainty boundary is:

[0121] (3)

[0122] The support degree of the lower uncertainty boundary of the uncertain coverage domain is:

[0123] (4)

[0124] In formula (1) and (2): , are the upper and lower membership functions of the two-type fuzzy set corresponding to the premise variable of the first record; is a one-type fuzzy set corresponding to the conclusion variable of the first record Membership function; This indicates the total number of records for the sampled data. ;

[0125] The weighted average support of this rule is calculated using the following formula (3):

[0126] (5)

[0127] The fuzzy rule subspace with the highest support is selected to construct the following interval type-2 fuzzy logic system.

[0128] (6)

[0129] In equations (3) and (4): It is a type II fuzzy set; The output variable is a type I fuzzy set; The total number of fuzzy rules ( ); For the first A fuzzy rule;

[0130] Based on the calculations above, the fuzzy rule table in Table 3 can be obtained using the sample data in Table 2.

[0131] Table 2 Sample Data Table of Type II Fuzzy Systems

[0132]

[0133] Table 3. Example rules for constraint boundaries of rolling compensation in type-II fuzzy systems.

[0134]

[0135] In Table 3 (after normalization): NB is the negative large fuzzy set; NS is the negative small fuzzy set; ZE is the intermediate fuzzy set; PB is the positive large fuzzy set; PS is the positive small fuzzy set;

[0136] (3) Establish Output of fuzzy model

[0137] For a certain rule in equation (4), calculate the input vector. Activation range:

[0138] (7)

[0139] For the Fuzzy rules, output variables The value corresponding to the maximum membership degree is:

[0140] (8)

[0141] where is the th rule in the fuzzy membership function reaches the maximum value, that is, the point value when

[0142] After the antecedent activation interval and the consequent output calculation, the defuzzification based on the KM algorithm can be expressed as follows:

[0143] (9)

[0144] where are the minimum and maximum output intervals obtained by the KM algorithm; , are the minimum and maximum activation weights corresponding to the th fuzzy rule calculated by the KM algorithm;

[0145] After weighted average defuzzification, the above interval type-2 fuzzy system can be written in the form of the following basis functions:

[0146] (10)

[0147] where is the adaptive parameter vector of the type-2 fuzzy system; the basis function vectors composed of activation weights are , respectively.

[0148] The initial value of the example parameter is .

[0149] (4) Learning of the parameter vector of the fuzzy model Based on small sample data, the gradient descent algorithm is used to learn the parameter vector of the type-2 fuzzy model.

[0150] Let the number of input-output sample data pairs be

[0151] , the current input variable be , and the predicted output of the type-2 fuzzy model be , the target real output be , then the training error function can be defined as:

[0152] (11) ​​

[0153] The gradient descent algorithm is used to minimize the objective error function in the above equation, thereby dynamically optimizing and adjusting the parameters of the fuzzy model and improving the modeling accuracy of the fuzzy model.

[0154] Therefore, according to the gradient descent method, the parameters The update rule is derived as follows:

[0155] (12)

[0156] In the formula: This represents the number of iterations. It is a positive learning rate.

[0157] Parameters in this example After learning using the gradient descent method above, it is changed to .

[0158] (5) Establish Adaptive adjustment of the uncertainty coverage of the fuzzy model

[0159] Input variables The UMF of an uncertain coverage area is represented by the following Gaussian function.

[0160] (13)

[0161] Where: UMF parameters , , Once specified, no further adjustments are made, and the LMF (Local Mean Square) of the coverage region is not determined. It is represented by the following Gaussian function:

[0162] (14)

[0163] Where: parameters of LMF , Similar to UMF, by adjusting Change the size of the uncertain coverage area;

[0164] Due to the uncertainty of boiler combustion, the establishment of LMF The adaptive adjustment strategy for the parameters is as follows: Under certain conditions, the actual detected sulfur dioxide and the desired sulfur dioxide levels are satisfied, and the LMF is dynamically adjusted. parameter

[0165] (15)

[0166] (16)

[0167] In equations (13) and (14): Output for the model; is a positive learning rate; is the expected value of sulfur dioxide for the second experiment is the compensation amount.

[0168] The membership function used for the input variable in this example is

[0169] (13) (14)

[0170] The parameters of the membership function are: .

[0171] The parameters of the membership function are: .

[0172] The results of the three dynamic compensation of sulfur constraint boundaries in this example are shown in Table 4. After determining the dynamic coal blending optimization model, the two-stage simplex method is used to solve the coal blending output scheme according to Figure 8 and Figure 9 process.

[0173] Table 4 Summary of the example of rolling compensation of the constraint boundary of sulfur by the two fuzzy systems

[0174]

[0175] Example 2

[0176] A dynamic adjustment method of sulfur constraint boundary based on open-loop coal blending optimization model, which takes into account the environmental protection, economy and practicability of power plant (considering the influence of boiler combustion uncertainty on environmental protection and economy), takes the minimum cost of mixed coal to meet the requirements of boiler combustion as the objective function, and uses the calorific value, sulfur, moisture and ash of coal as constraint conditions, to perform dynamic adjustment of sulfur constraint boundary based on open-loop coal blending optimization model.

[0177] The established coal blending optimization model and its constraint conditions are as follows:

[0178] (1)

[0179]

[0180]

[0181] In formula (1) and its constraint conditions: is the minimum cost of mixed coal; is the cost of the th single coal, i.e. the price; is the mixed coal ratio of the th single coal; is the number of coal types participating in coal blending; is the The proportion of the first type of coal in a coal mill or coal bunker; For the first to participate in coal blending The proportion of the second type of coal in the coal bunker or coal mill; For the first to participate in coal blending Coal feed rate of the coal mill; The number of coal mills in operation participating in coal blending; For the first to participate in coal blending The first coal type corresponding to the Coal feed rate of the coal mill To round down, representing the first... The coal bunker or coal mill number for each type of coal; This is a virtual coal coefficient; when taking a virtual coal type... ,otherwise The virtual coal type is added to each coal yard to resolve the conflict between the requirement that the proportion of coal in the coal yard should not be too small and the constraint of the proportion of coal types. This is the upper limit of the sulfur content of the blended coal; This is the dynamic adjustment amount for the upper limit of the sulfur constraint; , , These are the lower limit of the calorific value, and the upper limit of the moisture and ash content of the blended coal, respectively. , , , The first The sulfur content, calorific value, moisture content, and ash content of each coal type;

[0182] Specifically, the proportion of each coal type is restricted in the constraints of formula (1) in order to meet the requirements.

[0183] ( 1.3 .1)

[0184] In actual coal blending applications, each coal bunker shall not exceed two types of coal, that is, two types of coal or a single type of coal;

[0185] If "virtual coal type" is not introduced, under the constraints of the aforementioned equation (1.3.1) and also the constraints of equation (1.2), the solution of the optimization model can only be a mixture of two different coal types, and the case where the coal bunker is a single coal type is discarded.

[0186] To address this issue, a "virtual coal type" was introduced to resolve the contradiction between the proportion restrictions during coal extraction and the discarding of single coal types. First, a "virtual coal type" was introduced in each coal yard. Then, equation (1.3.1) was transformed into equations (1.3) and (1.4). The introduction of the "virtual coal type" solved the practical problem of coal blending for enterprises and effectively reduced coal blending costs.

[0187] One innovation different from the previous open-loop coal blending optimization model is: for the first time, the idea of "virtual coal" is introduced, and the constraint condition with "virtual coal" is innovatively designed, that is, the unique design of formulas (1.1), (1.2), (1.3) and (1.4).

[0188] The introduction of the innovative idea of "virtual coal" in the coal blending optimization model effectively solves the contradiction between the proportion limit when taking coal and the single coal being discarded. The introduction of "virtual coal" also focuses on solving the realistic dilemma that the coal less than 0.3 proportion taken by the stacker-reclaimer in the actual production process of the coal user such as power plant cannot be accurately taken. It is obvious that the introduction of the idea of "virtual coal" and its further extension application are more in line with the actual needs of the power plant.

[0189] The sulfur constraint boundary dynamic adjustment method based on the open-loop coal blending optimization model also introduces the dynamic adjustment of the sulfur constraint boundary.

[0190] Compared with the previous traditional open-loop coal blending optimization model, the unique design of the sulfur constraint boundary dynamic adjustment method based on the open-loop coal blending optimization model is to increase the dynamic adjustment amount of the sulfur constraint boundary, so that the coal blending optimization model has a dynamic adjustment effect.

[0191] Since the optimization coal blending model of formula (1) is still open-loop in nature, and because the boiler combustion condition is not considered when blending coal in advance;

[0192] The new method of dynamic adjustment of the sulfur constraint boundary based on the open-loop coal blending optimization model realizes the dynamic adjustment of the sulfur constraint boundary by real-time monitoring of the sulfide at the outlet of the boiler combustion of the current batch, which is represented as follows:

[0193] (2)

[0194] In formula (2): is the new constraint boundary of sulfur in the blended coal after dynamic adjustment; is the constraint boundary of sulfur in the blended coal given by the traditional open-loop coal blending; is the sulfur constraint boundary of the dynamic adjustment compensation addition;

[0195] The establishment The steps of dynamically adjusting the sulfur constraint boundary of the compensation addition meet the following requirements in turn:

[0196] Preamble

[0197] In step 1, the establishment The rule base of the type-2 fuzzy model meets the following requirements:

[0198] First, the input and output variables and fuzzy sets of the fuzzy model are established:

[0199] ​It is based on small sample experimental data and uses data mining to establish Rule base for type II fuzzy models;

[0200] Selected input variables It includes two real-time boiler parameter variables, where the first input variable is the real-time boiler load, expressed as: The unit is MW; the second input variable is the detected sulfur dioxide content at the boiler outlet, expressed as... The detection limit is expressed in mg / Nm³. 3 The system output variable is selected as the dynamic deviation compensation for sulfur in blended coal, expressed as: ;

[0201] For input prerequisite variables Each is divided into 5 fuzzy sets, using This indicates that the fuzzy set of the premise variables uses a Gaussian function; the conclusion variable... It is also divided into 5 fuzzy sets, using This indicates that the fuzzy combination of the conclusion variable adopts a trigonometric function; where: the fuzzy set corresponding to the premise variable is a type II fuzzy set, and the fuzzy set corresponding to the conclusion variable is a type I fuzzy set;

[0202] Then, support is used as the discrimination criterion for calculating the fuzzy space, and this is applied to the experimental sample data collected on-site. , First, calculate the support for each rule, where the support for the upper uncertainty bound is:

[0203] (3)

[0204] The support of the lower uncertainty bound of the uncertain coverage area is:

[0205] (4)

[0206] In equations (3) and (4): , The first Prerequisite variables for each record The corresponding type II fuzzy set The upper and lower membership functions; For the first Conclusion variables of each record The corresponding type I fuzzy set Membership function; This indicates the total number of records for the sampled data. ;

[0207] The weighted average support of this rule is calculated using the following formula (6):

[0208] (5)

[0209] The fuzzy rule subspace with the highest support is selected to construct the following interval type-2 fuzzy logic system.

[0210] (6)

[0211] In equations (5) and (6): It is a type II fuzzy set; The output variable is a type I fuzzy set; The total number of fuzzy rules ( ); For the first A fuzzy rule;

[0212] After mining fuzzy rules based on sampled data, the final type II fuzzy rule library is formed.

[0213] After extracting fuzzy rules based on sampled data, power plant experts, drawing on years of boiler operation experience, verify and fine-tune these rules to form the final Type II fuzzy rule library. Adjustments can be made based on experience and knowledge; alternatively, it can be integrated with case libraries, expert systems, and other supporting automatic control systems and programs for automated processing.

[0214] Establish in step 2 The output of the fuzzy model must meet the following requirements:

[0215] For a certain rule in equation (6), first calculate the input vector. Activation range:

[0216] (7)

[0217] For the τth fuzzy rule, the output variable is... The value corresponding to the maximum membership degree is:

[0218] (8)

[0219] In the formula: It is the first Rules in fuzzy membership functions The corresponding value when the maximum value is obtained Point value, that is Time corresponding Point value;

[0220] In the preceding activation region and the successor After the output calculation, the reduced form based on the KM algorithm can be represented as follows:

[0221] (9)

[0222] wherein: are the minimum and maximum output intervals obtained by the KM algorithm; , are the minimum and maximum activation weights corresponding to the i-th fuzzy rule calculated by the KM algorithm; After weighted average defuzzification, the above interval type-2 fuzzy system can be written in the form of the following basis functions:

[0223]

[0224] (10)

[0225] wherein: is the adaptive parameter vector of the type-2 fuzzy system; the basis function vectors composed of activation weights are , .

[0226] Step 3 meets the following requirements:

[0227] It establishes the parameter vector learning of the type-2 fuzzy model based on small sample data using the gradient descent algorithm.

[0228] Let the number of input-output sample data pairs be , the current input variable be , the predicted output of the type-2 fuzzy model be , and the target real output be , then the training error function can be defined as:

[0229] (11)

[0230] The gradient descent algorithm is used to minimize the above target error function, thereby dynamically optimizing and adjusting the parameters of the fuzzy model and improving the modeling accuracy of the fuzzy model.

[0231] Therefore, according to the gradient descent method, the update rule of the parameter is derived as follows:

[0232] (12)

[0233] wherein: is the iteration number; is a positive learning rate.

[0234] Step 4 meets the following requirements:

[0235] It is based on setting an ideal compensation ​Adaptive adjustment using gradient descent algorithm Uncertainty Coverage Domain (LMF) of Type II Fuzzy Model:

[0236] Input variables The UMF of an uncertain coverage area is represented by the following Gaussian function.

[0237] (13)

[0238] Where: UMF parameters , , Once specified, no further adjustments are made, and the LMF (Local Mean Square) with an uncertain coverage area is represented by the following Gaussian function:

[0239] (14)

[0240] Where: parameters of LMF , Similar to UMF, by adjusting Change the size of the uncertain coverage area;

[0241] Due to the uncertainty of boiler combustion, the establishment of LMF The adaptive adjustment strategy for the parameters is as follows: Under certain conditions, the actual detected sulfur dioxide and the desired sulfur dioxide levels are satisfied, and the LMF is dynamically adjusted. parameter

[0242] (15)

[0243] (16)

[0244] In equations (15) and (16): Output for the model; This is a positive learning rate; The expected value of sulfur dioxide in the experiment corresponds to Compensation amount.

[0245] The schematic diagram of the above process is shown below. Figure 1 and Figure 2 .

[0246] Other relevant details in this embodiment are explained below:

[0247] Compared with existing technologies, the main advantages of the sulfur constraint boundary dynamic adjustment method based on the open-loop coal blending optimization model described in this embodiment are as follows:

[0248] 1. This method considers the real-time combustion problem of the boiler. It uses online parameter monitoring, modeling, and decision-making regarding the boiler combustion status to continuously compensate for the upper limit of the sulfur constraint in the next round of open-loop optimized coal blending. Compared with traditional open-loop coal blending methods, this method utilizes negative bias... Compensation can reduce sulfide emissions; under the premise of ensuring that sulfide emissions are well below the limit, appropriate positive bias can be used. It can improve the economic efficiency of coal blending schemes.

[0249] 2. A type-two fuzzy method was adopted to establish the system model. Compared with the traditional type-one fuzzy method, it can better handle the uncertainties generated by the boiler combustion system, thus improving the bias... The calculations are more accurate.

[0250] 3. In the design of type II fuzzy systems, an adaptive mechanism is adopted to change the uncertainty coverage of the fuzzy set, which further improves the accuracy of dynamic modeling of type II fuzzy systems compared with the traditional fixed uncertainty coverage method.

[0251] The main innovations of this embodiment are four:

[0252] 1. The concept of "virtual coal type" is introduced for the first time in the constraint condition section of the coal blending optimization model. Its main advantage is that it can better solve the constraint conflict between the proportional tonnage limit of the stacker-reclaimer in the coal yard and the single coal type in the coal bunker, thus saving the enterprise coal blending cost.

[0253] 2. In order to achieve dynamic adjustment of open-loop coal blending, a method for dynamically adjusting the sulfur constraint of the open-loop optimization model is proposed for the first time. Its main purpose is to promote the development of open-loop coal blending towards closed-loop coal blending, taking into account both the economic and environmental benefits of coal blending and combustion.

[0254] 3. To address the drawback of open-loop coal blending not considering the uncertainties of boiler combustion, this paper proposes for the first time to use a type-two fuzzy system to dynamically adjust the coal blending optimization model. A fuzzy rule base is constructed by combining data mining and expert domain knowledge, making the dynamically adjusted model more accurate. Its main purpose is to improve the accuracy of the open-loop coal blending optimization model and to take into account the impact of boiler combustion uncertainties.

[0255] 4. In the process of using the Type II fuzzy system for dynamic coal blending adjustment, an adaptive adjustment of the uncertain coverage domain of the Type II fuzzy system was proposed for the first time, which solved the problem that the traditional fixed coverage domain Type II fuzzy system does not have adaptive capability, and further improved the accuracy of dynamic modeling of the Type II fuzzy system.

[0256] In existing technologies, open-loop coal blending in power plants involves inputting information such as coal quality parameters, planned load, and coal mill parameters of each individual coal without considering boiler combustion conditions. Based on an optimization model, a coal blending output plan is formulated for boiler co-firing in future periods. This is pre-furnace coal blending.

[0257] The closed-loop coal blending applicable to the power plant proposed in the embodiment is based on the open-loop coal blending, and combines the current combustion condition of the boiler to guide the coal blending in the future period. In view of the disadvantage that the currently widely used open-loop coal blending does not consider the real-time combustion condition of the boiler, the embodiment changes the disadvantage that the boiler combustion condition is not considered by monitoring the sulfur dioxide and real-time load online parameters of the boiler combustion condition, fuzzy modeling and self-adaptive decision rolling compensation upper limit of the sulfur constraint of the open-loop optimization model, reduces the sulfur oxide emission at the outlet of the boiler by applying a negative bias to the upper limit of the sulfur constraint, applies a positive bias to improve the coal blending economy, and makes the currently widely used open-loop coal blending develop towards the closed-loop coal blending. The embodiment has predictable greater economic and social environmental protection values.

[0258] In summary, the embodiment has predictable greater economic and social values.

Claims

1. A method for dynamic adjustment of sulfur constraint boundary based on an open-loop coal blending optimization model, characterized in that: It takes the minimum mixed coal cost meeting the boiler combustion requirement as the objective function, and performs dynamic adjustment of the sulfur constraint boundary based on the open-loop coal blending optimization model; The coal blending optimization model and its constraint conditions are as follows: （1） In equation (1) and its constraints: This represents the minimum cost for mixed coal; It is the first The cost, i.e., the price, of a single coal unit; It is the first The proportion of a single coal unit participating in coal blending; It refers to the number of coal types involved in the blending; For the first Coal feed rate of each coal mill; The number of coal mills or coal bunkers involved in coal blending; This is a virtual coal coefficient; when taking a virtual coal type... ,otherwise , This is the upper limit of the sulfur content of the blended coal; This is the dynamic adjustment amount for the upper limit of the sulfur constraint; , , These are the lower limit of the calorific value, and the upper limit of the moisture and ash content of the blended coal, respectively. , , , The first The sulfur content, calorific value, moisture content, and ash content of each coal type; The new method of dynamic adjustment of the sulfur constraint boundary based on the open-loop coal blending optimization model realizes dynamic adjustment of the sulfur constraint boundary by real-time monitoring of the sulfide at the outlet of the boiler combustion of the current batch, and is expressed as follows: （2） In formula (2): is a new constraint upper bound of sulfur in the dynamically adjusted blended coal; is a constraint upper bound of sulfur in the blended coal for traditional open-loop coal blending; is a sulfur constraint bound for dynamic adjustment compensation addition; establishing The step of dynamically adjusting the compensation added sulfur constraint bound meets the following requirements in turn: Step 1: Establishing a library of two-type fuzzy rules of the fuzzy model; Step 2: Establishing the output of the fuzzy model the output of the fuzzy model established in step 2 the output of the fuzzy model established in step 2 satisfies the following requirements: Based on the reduction of KM algorithm and the anti-fuzzification, the paper establishes The output of the bivalent fuzzy model For some rule of formula (6), firstly calculate the activation interval of input vector : （7） For the τth fuzzy rule, the value corresponding to the maximum membership degree of the output variable is: （8） In the formula: It is the first Rules in fuzzy membership functions The corresponding value when the maximum value is obtained Point value, that is Time corresponding Point value; The antecedent activation interval and the consequent After the output calculation, the reduction form based on the KM algorithm can be expressed as follows: （9） wherein: is the output interval obtained with the KM algorithm; After weighted average defuzzification, the above interval type-2 fuzzy system can be written in the form of the following basis functions: （10） In the formula: is the adaptive parameter vector of the bivalent fuzzy system; the basis function vector composed of the activation weights is respectively: ;​ Step 3: Establish Parameter vector learning of fuzzy model; Step 4: Establishing an adaptive adjustment strategy for the uncertain coverage domain of the fuzzy model; Step 4 meets the following requirements: It is based on setting ideal compensation Adaptive adjustment is adopted by gradient descent algorithm Uncertain coverage domain of bivalent fuzzy model LMF: Input variables The UMF of the uncertainty coverage region of is represented by the following Gaussian function （13） where the parameters of UMF are , , which is not adjusted after designation, and the LMF without the determined coverage is expressed by the following Gaussian function: （14） where: parameters of the LMF , As for the UMF, by adjusting the size of the uncertain coverage area; Due to the uncertainty of the boiler combustion, the parameters of the LMF are adjusted adaptively as follows: the LMF parameters are adjusted dynamically when the actual detected sulfur dioxide and the expected sulfur dioxide satisfy certain conditions The adaptive adjustment strategy of the parameters is as follows: the LMF parameters are adjusted dynamically when the actual detected sulfur dioxide and the expected sulfur dioxide satisfy certain conditions parameters （15） （16） In formulas (15) and (16): is the model output; is a positive learning rate; is the experimental sulfur dioxide expected value corresponding to the compensation amount; E(n) is a training error function.

2. The method of claim 1, wherein the method further comprises: determining a sulfur constraint boundary based on the open-loop coal blending optimization model; and determining a new sulfur constraint boundary based on the open-loop coal blending optimization model and the determined sulfur constraint boundary. In step 1, establish The rule base of the bivariate fuzzy model satisfies the following requirements: It is based on small sample experimental data using data mining to establish The rule base of the type-2 fuzzy model; Firstly, the input and output variables of the fuzzy model and the fuzzy sets are established input and output variables of the fuzzy model and the fuzzy sets Selected input variables The selected input variables include two real-time parameter variables of the boiler, wherein the first input variable is a real-time load of the boiler, indicated by , and the unit is MW; the second input variable is a detected sulfur dioxide content at an outlet of the boiler, indicated by , and the unit is mg / Nm 3 ; the selected output variable of the system is a dynamic deviation compensation of the coal blending and mixing sulfur, indicated by ; input premise variables are denoted by , the fuzzy sets of premise variables are Gaussian functions; the fuzzy sets of conclusion variables are denoted by , the fuzzy sets of conclusion variables are triangular functions; wherein: the fuzzy sets of premise variables are bivariate fuzzy sets, the fuzzy sets of conclusion variables are univariate fuzzy sets; Then, the support is taken as the criterion to calculate the fuzzy space, and the support of each rule is calculated, where the support of the upper uncertainty boundary is: , , the support of each rule is calculated, where the support of the upper uncertainty boundary is: （3） The support of the lower uncertainty boundary of the uncertain coverage domain is: （4） In formula (3) and (4): , are the upper and lower membership functions of the bivalent fuzzy set corresponding to the premise variable of the m-th record is the membership function of the univalent fuzzy set corresponding to the conclusion variable of the m-th record denotes the total number of records to which the sampling data belongs, ;​​​​​​ The weighted average support of the rule is calculated by the following formula (5): （5） The fuzzy rule subspace with the maximum support is selected to constitute the following interval type-2 fuzzy logic system （6） In formulas (5) and (6): is a two-type fuzzy set; is an output variable one-type fuzzy set; R is the total number of fuzzy rules τ = 1, …, R ); is the th fuzzy rule; After mining the fuzzy rules based on the sampling data, the final type-2 fuzzy rule base is formed.

3. The method of claim 1 or 2, wherein the method further comprises: determining the sulfur constraint boundary based on the open-loop coal blending optimization model; and determining the dynamic adjustment of the sulfur constraint boundary based on the open-loop coal blending optimization model. Step 3 meets the following requirements: The gradient descent algorithm is adopted to establish the model based on small sample data Parameter vector learning of the second type fuzzy model Let The number of input-output sample data pairs is , the current input variable is is the predicted output of the bivariate fuzzy model, is the target real output, then the training error function can be defined as: （11） The gradient descent algorithm is used to minimize the above objective error function, thereby dynamically optimizing and adjusting the parameters of the fuzzy model, and improving the modeling accuracy of the fuzzy model; Thus, according to the gradient descent method, the update rule for the parameters is derived as follows: （12） In the formula: is the number of iterations; is a positive learning rate.

Images