A thermal power unit NO X Emission modeling method and system

Abstract

The invention discloses a thermal power unit NO X Emission modeling method and system, the method comprising: based on the influence of NO X The basic parameters of emissions are obtained by obtaining the historical operation data of the thermal power unit and establishing the training set D0; the density clustering algorithm and the sparse greedy matrix are used to simplify the training set D0 to obtain the sample representative set D2; a semi-parametric support vector regression machine model is constructed based on the sample representative set D2, and the weights of the semi-parametric support vector regression machine model are solved based on the iterative reweighted least squares method; based on the training set D0, the Apache Spark parallel computing framework is used for parallel design to obtain the final weights of the semi-parametric support vector regression machine model; based on the final weights, the semi-parametric support vector regression machine model is parallelized using Apache Spark to obtain the thermal power unit NO X Emission model. When processing large-scale thermal power plant data, the present invention can shorten the modeling time and meet industrial requirements.

Description

Technical Field

[0001] The present invention relates to a thermal power unit NO X The invention discloses an emission modeling method and system, belonging to the field of emission technology of thermal power units. Background Art

[0002] As the country pays more attention to environmental protection issues, reducing the emission of harmful gases has become an important issue facing enterprises. Among them, nitrogen oxides NO emitted by thermal power plants X The most harmful to the ecological environment pollution is difficult to deal with. Therefore, the research on NO X Emission modeling has important environmental implications.

[0003] In recent years, support vector machine (SVM) has been widely used in the combustion process modeling of thermal power units due to its advantages such as complete theory, strong adaptability, global optimization, short training time and good generalization performance. Existing literature (Zhou Jianxin, Si Fengqi, Qiu Xiaozhi, et al. Optimization and determination of boiler operating oxygen reference value based on SVR and GA [J]. Journal of Southeast University (Natural Science Edition), 2008 (06): 1061-1066.) uses full parameter support vector machine to establish a certain boiler coal consumption rate and NO X The prediction model was developed and the field data was used to verify that the model has good prediction performance; the existing literature (Gu Y, Zhao W, Wu Z. Online adaptive least squares support vector machine and its application in utility boiler combustion optimization systems [J]. Journal of Process Control, 2011, 21(7): 1040-1048.) proposed a boiler combustion process modeling method based on an adaptive least squares support vector machine based on real-time operation data of a power plant; the existing literature (Zhang Xiaoyu, Fu Lin, Shen Jiong, et al. Research on boiler dynamic modeling method based on online support vector machine [J]. Proceedings of the CSEE, 2017(S1): 123-128.) proposed a boiler dynamic modeling method based on online support vector machine, established a mathematical model of the combustion characteristics of the reaction boiler, and verified the model using the field operation data of a power plant boiler; the existing literature (Gu Yanping, Zhao Wenjie, Wu Zhansong. Power plant boiler combustion optimization based on least squares support vector machine [J]. Proceedings of the CSEE, 2010(17): 7.) proposed a method based on least squares support vector machine (LS-SVM) to achieve the optimization of power plant boiler efficiency and NO XThe existing literature (Lü You, Liu Jizhen, Yang Tingting, et al. NOx emission characteristics modeling based on PLS feature extraction and LS-SVM [J]. Chinese Journal of Scientific Instrumentation, 2013, 34(11): 2418-2424.) proposed combining the partial least squares (PLS) method with the least squares support vector machine (LS-SVM) to establish a NOx emission model based on the field operation data of a power plant boiler. X Emission model; Existing literature (Zhou Hao, Ding Fang, Huang Yan, Zhou Kang. KVM-based power plant boiler NO x Emission characteristics big data modeling [J]. Proceedings of the CSEE, 2016, 36(3): 717-722.) proposed a NO X The emission modeling method was used to select the operating parameters of the coal mill primary air volume, primary air temperature, outlet temperature, coal feeder coal volume, etc. as the model input parameters for a supercritical boiler. X The emission characteristics model was constructed and the generalization ability of the model was verified.

[0004] However, the support vector regression modeling methods in the above literature are all based on small sample data sets. When the number of samples is too large, a large number of support vectors will be generated, resulting in an overly complex support vector machine model structure. In addition, the above support vector machine training process adopts a serial mechanism. When processing large-scale thermal power plant data, the modeling time is too long, and it is even difficult to converge, the accuracy is poor, and it is difficult to meet industrial requirements. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a thermal power unit NO X The emission modeling method and system can shorten the modeling time and meet industrial requirements when processing large-scale thermal power plant data.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a thermal power unit NO X Emissions modelling methods, including:

[0008] Based on the influence of NO X The basic emission parameters obtain the historical operation data of thermal power units and establish the training set D0;

[0009] The training set D0 is simplified using the density clustering algorithm and the sparse greedy matrix to obtain the sample representative set D2;

[0010] A semi-parametric support vector regression model is constructed based on the sample representative set D2, and the weights of the semi-parametric support vector regression model are solved based on the iterative reweighted least squares method;

[0011] Based on the training set D0, the Apache Spark parallel computing framework is used for parallel design to obtain the final weights of the semi-parametric support vector regression model;

[0012] Based on the final weights, the semi-parametric support vector regression model is parallelized using Apache Spark to obtain the NO X Emission model.

[0013] In combination with the first aspect, further, establishing the training set D0 includes:

[0014] According to the combustion process, NO x Generation mechanism, determine the impact of NO x Input variables for emission values ​​u1,u2,...,u q ;

[0015] Let the order of each input variable be o1, o2, ..., o q ;

[0016] Establish training set D0 = {(Φ i ,y i ),i=1,…,N0}, where N0 is the number of samples, Φ i is the input vector of the i-th sample, Φ i Consists of the previous output and input, Φ′ i ∈R o×1 .

[0017] In combination with the first aspect, further, simplifying the training set D0 by using a density clustering algorithm and a sparse greedy matrix includes:

[0018] The density clustering algorithm is used to roughly simplify the training set D0, and the cluster center set D1 = {(Φ′ i ,y′ i ),i=1,…,N1}, where N1 is the number of cluster centers and N1<N0; Φ′ i is the input vector of the i-th sample in the cluster center set, Φ′ i ∈R o×1 ;y′ i is the output of the i-th sample in the cluster center set, y′∈R;

[0019] A sparse greedy matrix algorithm is used to extract a set of sample representative sets D2 that can maximally represent the entire data set from the cluster center set D1. i ,yi ),i=1,…N2}, where N2 is the number of samples and N2<N1; {X i , i=1,…,N2} is the basic sample input vector set, X i is the input vector of the i-th learning training sample in the sample representative set, and X i ∈R o×1 ;y i is the output of the i-th sample in the sample representative set, and y i ∈R.

[0020] In combination with the first aspect, further, constructing a semi-parametric support vector regression model based on the sample representative set D2 includes:

[0021] The sample representative set D2 is used to estimate the parameter w of the support vector regression machine, and the parameter w is estimated by the following predefined model:

[0022]

[0023] In formula (1), N2 is the number of samples; X i is the input vector of the i-th learning training sample in the sample representative set, and X i ∈R o×1 ; φ(·) represents a nonlinear mapping to a high-dimensional space; Ψ is defined by:

[0024]

[0025] In formula (1), Λ is the weight, which is expressed by the following formula:

[0026]

[0027] According to the samples in the sample representative set D2, the samples are fitted with a linear function without error under the accuracy ε, and the relaxation factor ξ is introduced i ≥0 and The construction of the semi-parametric support vector regression model is transformed into minimizing the structural risk function, which is expressed by the following formula:

[0028]

[0029] In formula (4), C is the penalty factor, b is the bias, ε is the accuracy; It is the kernel matrix of the sample input vector in the sample representative set D2, which is expressed by the following formula:

[0030]

[0031] In formula (5), the kernel function k(X i ,X j ) Select the Gaussian kernel function, which is expressed by the following formula:

[0032]

[0033] In formula (5), σ is the width parameter of the Gaussian kernel function;

[0034] In formula (4), k i Defined by:

[0035]

[0036] In formula (7), Φ i is the input vector of the i-th sample;

[0037] The optimal decision function of the semi-parametric support vector regression machine is obtained as follows:

[0038]

[0039] In order to obtain the weight Λ, formula (4) is rewritten as:

[0040]

[0041] Formula (9) is the model formula of the semi-parametric support vector regression machine;

[0042] Using the Lagrangian function to process equation (9) we get the Lagrangian equation, and taking the partial derivative of Λ, b in the equation, we get:

[0043]

[0044] In formula (10), is the kernel matrix of the sample input vector in the training set D0 and the sample representative set D2; Vector 1 = [1,…,1] T ,

[0045] In combination with the first aspect, further, solving the weight of the semi-parametric support vector regression model based on iterative reweighted least squares method includes:

[0046] Give a i and The initial value of , solve equation (10) to get the initial value of weight Λ;

[0047] According to the constraints in formula (9) and the initial value of weight Λ, update a i and

[0048] According to the updated a i and The weight Λ is calculated again by formula (10);

[0049] After executing k updates, if ||Λ k -Λ k-12 ||<10 -3 , end the iterative calculation and output the weight Λ.

[0050] In combination with the first aspect, further, the parallel design using the Apache Spark parallel computing framework to obtain the final weight of the semi-parametric support vector regression model includes:

[0051] Input the training set D0 into the HDFS distributed data storage system;

[0052] On the master node, the training set D0 is divided into S evenly spaced sub-datasets, denoted as Distribute to S child nodes respectively;

[0053] On each of the S child nodes, the sub-datasets are simultaneously Use density clustering algorithm to determine the set of cluster centers Then use the sparse greedy matrix approximation algorithm to select the cluster center set Find the set that best represents the entire data set

[0054] The weight Λ of the semi-parametric support vector regression model is calculated by using the iterative reweighted least squares algorithm on each of the S child nodes. 1 ,…,Λ S ;

[0055] On each of the S child nodes, the model weight Λ 1 ,…,Λ S Find The support vectors in ;

[0056] The support vectors on the child nodes are sent to the main node to be merged into the support vector set SVs. On the main node, SVs is used as the input of the semi-parametric support vector regression machine, and the iterative reweighted least squares algorithm is executed again to obtain the final weight Λ of the semi-parametric support vector regression machine model.

[0057] In combination with the first aspect, further, the thermal power unit NO X Emissions models, including:

[0058] Initialize the number of density cluster centers to N1, and give the root mean square error e of the modeling;

[0059] According to the final weight Λ of the semi-parametric support vector regression model, a semi-parametric support vector regression model is obtained;

[0060] Calculate the root mean square error of the model. If the error is greater than e, increase the number of cluster centers N1, reinitialize and calculate the final weight; otherwise, the modeling process ends and the thermal power unit NO is obtained. X Emission model.

[0061] In a second aspect, the present invention provides a thermal power unit NO X Emissions modeling system, including:

[0062] Acquisition module: used to influence NO X The basic emission parameters obtain the historical operation data of thermal power units and establish the training set D0;

[0063] Simplification module: used to simplify the training set D0 using the density clustering algorithm and the sparse greedy matrix to obtain the sample representative set D2;

[0064] The first solution module is used to construct a semi-parametric support vector regression model based on the sample representative set D2, and solve the weight of the semi-parametric support vector regression model based on the iterative reweighted least squares method;

[0065] The second solution module is used to obtain the final weight of the semi-parametric support vector regression model based on the training set D0 by using the Apache Spark parallel computing framework for parallel design;

[0066] Output module: It is used to obtain the NO of thermal power units based on the final weights by parallelizing the semi-parametric support vector regression model using Apache Spark. X Emission model.

[0067] In a third aspect, the present invention provides a thermal power unit NO X An emission modeling device, including a processor and a storage medium;

[0068] The storage medium is used to store instructions;

[0069] The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0070] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0071] Compared with the prior art, the present invention provides a thermal power unit NO X The beneficial effects achieved by the emission modeling method and system include:

[0072] The present invention is based on the influence of NO XThe basic parameters of emission are used to obtain the historical operation data of the thermal power unit, and a training set D0 is established; the training set D0 is simplified by using a density clustering algorithm and a sparse greedy matrix to obtain a sample representative set D2; the present invention performs a rough simplification on the data set first and then a fine simplification, which can effectively avoid the matrix inversion dimension disaster problem that occurs when the sparse greedy matrix processes large sample data;

[0073] The present invention constructs a semi-parametric support vector regression model based on the sample representative set D2, and solves the weight of the semi-parametric support vector regression model based on the iterative reweighted least squares method; it can effectively control the complexity of the regression model, which is conducive to improving the training efficiency;

[0074] Based on the training set D0, the present invention uses the Apache Spark parallel computing framework for parallel design to obtain the final weight of the semi-parametric support vector regression model; it can improve the ability of the algorithm to process large-scale data sets;

[0075] Based on the final weights, the semi-parametric support vector regression model is parallelized using Apache Spark to obtain the NO X Emission model; can accurately establish NO X Emission model to meet industrial production needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A thermal power unit NO provided in the first embodiment of the present invention X Flowchart of the emissions modelling approach;

[0077] Figure 2 A thermal power unit NO provided in the first embodiment of the present invention X The influence of NOx on thermal power units in emission modeling method X Schematic diagram of the basic parameters of emissions;

[0078] Figure 3 A thermal power unit NO provided in the first embodiment of the present invention X Flowchart of parallel design in emission modeling approach;

[0079] Figure 4 A thermal power unit NO provided in the first embodiment of the present invention X The impact of different numbers of cluster centers on the root mean square error in emission modeling methods;

[0080] Figure 5 A thermal power unit NO provided in the first embodiment of the present invention X The impact of different numbers of cluster centers on modeling time in emission modeling methods. DETAILED DESCRIPTION

[0081] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0082] Embodiment 1:

[0083] like Figure 1 As shown, the embodiment of the present invention provides a thermal power unit NO X Emissions modelling methods, including:

[0084] Based on the influence of NO X The basic emission parameters obtain the historical operation data of thermal power units and establish the training set D0;

[0085] The training set D0 is simplified using the density clustering algorithm and the sparse greedy matrix to obtain the sample representative set D2;

[0086] A semi-parametric support vector regression model is constructed based on the sample representative set D2, and the weights of the semi-parametric support vector regression model are solved based on the iterative reweighted least squares method;

[0087] Based on the training set D0, the Apache Spark parallel computing framework is used for parallel design to obtain the final weights of the semi-parametric support vector regression model;

[0088] Based on the final weights, the semi-parametric support vector regression model is parallelized using Apache Spark to obtain the NO X Emission model.

[0089] The specific steps are as follows:

[0090] Step 1: Based on the impact of thermal power unit NO X The basic emission parameters are obtained from the historical operation data of thermal power units to establish the training set D0.

[0091] Step 1.1: Figure 2 As shown, according to the NO X Generation mechanism. In this example, the total amount of coal entering the furnace, the total amount of air entering the furnace, the differential pressure between the secondary wind box and the furnace, the ventilation volume of the coal mill, the characteristics of the coal type, the swing angle of the burner, the opening of each secondary air door, and the opening of each burnout air are used as the input variables u1, u2, ..., u q , NO X The emission concentration is used as the output variable of the training set D0.

[0092] Step 1.2: Let the order of each input variable be o1, o2, ..., o q .

[0093] Step 1.3: Create a training set D0 = {(Φi ,y i ),i=1,…,N0}, where N0 is the number of samples, Φ i is the input vector of the i-th sample, and is also the recursive vector that affects the system input; Φ i Output variable y from previous z dimension i and input variables u1,u2,...,u q composition,

[0094] Step 2: Use the density clustering algorithm and sparse greedy matrix to simplify the training set D0 to obtain the sample representative set D2.

[0095] Step 2.1: Use the density clustering algorithm to roughly simplify the training set D0 and obtain the cluster center set D1 = {(Φ′ i ,y′ i ),i=1,…,N1}, where N1 is the number of cluster centers and N1<N0; Φ′ i is the input vector of the i-th sample in the cluster center set, y′ i is the output of the i-th sample in the cluster center set, y′∈R.

[0096] Step 2.2: Use the sparse greedy matrix algorithm to extract a set of sample representative sets D2 that can best represent the entire data set from the cluster center set D1. i ,y i ),i=1,…N2}, where N2 is the number of samples and N2<N1; {X i ,=1,…,N2} is the basic sample input vector set, X i is the input vector of the i-th learning training sample in the sample representative set, and is called X i,i =1,...,N2 and Also called X i , i=1,...,N2 is the basic sample input vector set; y i is the output of the i-th sample in the sample representative set, and y i ∈R.

[0097] In this embodiment, the number of cluster centers N1 ranges from 1000 to 15000 with a step length of 1000, and 15 groups of experiments are performed.

[0098] Step 3: Construct a semi-parametric support vector regression model based on the sample representative set D2, and solve the weights of the semi-parametric support vector regression model based on the iterative reweighted least squares method.

[0099] Step 3.1: Find the weight Λ.

[0100] Step 3.1.1: Estimate the parameters w of the support vector regression machine using the sample representative set D2. Estimate the parameters w using the following predefined model:

[0101]

[0102] In formula (1), N2 is the number of samples; X i is the input vector of the i-th learning training sample in the sample representative set, and X i ∈R o×1 ; φ(·) represents a nonlinear mapping to a high-dimensional space; Ψ is defined by:

[0103]

[0104] In formula (1), Λ is the weight, which is expressed by the following formula:

[0105]

[0106] Step 3.1.2: According to the samples in the sample representative set D2, the linear function is fitted without error under the accuracy ε, and the relaxation factor ξ is introduced i ≥0 and The construction of the semi-parametric support vector regression model is transformed into minimizing the structural risk function, which is expressed by the following formula:

[0107]

[0108] In formula (4), C is the penalty factor, b is the bias, ε is the accuracy; It is the kernel matrix of the sample input vector in the sample representative set D2, which is expressed by the following formula:

[0109]

[0110] In formula (5), the kernel function K(X i ,X j ) Select the Gaussian kernel function, which is expressed by the following formula:

[0111]

[0112] In formula (5), σ is the width parameter of the Gaussian kernel function;

[0113] In formula (4), k i Defined by:

[0114]

[0115] In formula (7), Φ i is the input vector of the i-th sample;

[0116] Step 3.1.3: The optimal decision function of the semi-parametric support vector regression machine is obtained as follows:

[0117]

[0118] In order to obtain the weight Λ, formula (4) is rewritten as:

[0119]

[0120] Formula (9) is the model formula of the semi-parametric support vector regression machine;

[0121] Step 3.1.4: Use the Lagrangian function to process equation (9) to obtain the Lagrangian equation, and take the partial derivative of Λ, b in the equation to obtain:

[0122]

[0123] In formula (10), is the kernel matrix of the sample input vector in the training set D0 and the sample representative set D2; Vector 1 = [1,…,1] T ,

[0124] Step 3.1.5: Solve equation (10) to obtain the weight Λ.

[0125] Step 3.2: Solve the weights of the semi-parametric support vector regression model based on iterative reweighted least squares method.

[0126] Step 3.2.1: Assign a i and The initial value of , solving equation (10) to obtain the initial value of weight Λ.

[0127] Step 3.2.2: Update a according to the constraints in equation (9) and the initial value of the weight Λ i and

[0128] Step 3.2.3: According to the updated a i and The weight Λ is calculated again using formula (10).

[0129] Step 3.2.4: After performing k updates, if ||Λ k -Λ k-1 ||<10 -3 , end the iterative calculation and output the weight Λ.

[0130] Step 4: Based on the training set D0, Figure 3As shown, the Apache Spark parallel computing framework is used for parallel design to obtain the final weights of the semi-parametric support vector regression machine model.

[0131] Step 4.1: Input the training set D0 into the HDFS distributed data storage system.

[0132] Step 4.2: On the master node, divide the training set D0 into S evenly spaced sub-datasets, denoted as Distribute to S child nodes respectively.

[0133] Step 4.3: Perform simultaneous operations on the sub-datasets on each of the S sub-nodes. Use density clustering algorithm to determine the set of cluster centers Then use the sparse greedy matrix approximation algorithm to select the cluster center set Find the set that best represents the entire data set

[0134] Step 4.4: Use the iterative reweighted least squares algorithm to find the weight Λ of the semi-parametric support vector regression model on each of the S child nodes. 1 ,…,Λ S .

[0135] Step 4.5: On each of the S child nodes, simultaneously calculate the model weight Λ 1 ,…,Λ S Find The support vectors in .

[0136] Step 4.6: Send the support vectors on the child nodes to the master node to merge into the support vector set SVs. On the master node, use SVs as the input of the semi-parametric support vector regression machine, and execute the iterative reweighted least squares algorithm again to obtain the final weight Λ of the semi-parametric support vector regression machine model.

[0137] Step 5: Based on the final weights, use Apache Spark to parallelize the semi-parametric support vector regression model to obtain the NO X Emission model.

[0138] Step 5.1: Initialize the number of density clustering centers to N1 and give the root mean square error e of the modeling.

[0139] Step 5.2: According to the final weight Λ of the semi-parametric support vector regression model, a semi-parametric support vector regression model is obtained.

[0140] Step 5.3: Calculate the root mean square error of the model. If the error is greater than e, increase the number of cluster centers N1 and return to step 5.1; otherwise, the modeling process ends and the thermal power unit NO is obtained. X Emission model.

[0141] Specifically, this example uses four hosts as nodes in the cluster environment, including a master node and three slave nodes. The experimental object is a 600MW subcritical coal-fired power generation unit in a power plant, and the total amount of coal entering the furnace, the total amount of air entering the furnace, the differential pressure between the secondary wind box and the furnace, the pulverizer ventilation volume, the coal type characteristics (7 inputs), the burner swing angle, 6 secondary air door openings, and 2 burnout air openings are used as the NO X Input to emission model, NO X Emission concentration as NO X The output of the emission model is analyzed using 61,660 sets of operating data collected over a period of time. The cluster center N1 ranges from 1000 to 15000, with a step size of 1000, and 15 sets of experiments are performed. The experimental results are shown in Figure 4 , Figure 5 shown.

[0142] Figure 4 The vertical axis is the root mean square error (RMSE) of the model built by this method under different numbers of cluster centers, and the horizontal axis is the number of cluster centers set by the user. Figure 4 It can be observed that with the increase of the number of cluster centers, the root mean square error generally shows a downward trend, and the root mean square error of the model built by this method is always less than 0.3, which has a good fitting ability.

[0143] Figure 5 The vertical axis is the modeling time of the model built by this method under different numbers of cluster centers, and the horizontal axis is the number of cluster centers set by the user. Figure 5 It can be observed that with the increase of the number of cluster centers, the modeling time generally increases. Therefore, within the allowable range of model error, a smaller number of cluster centers can be selected to improve modeling efficiency.

[0144] Embodiment 2:

[0145] The embodiment of the present invention provides a thermal power unit NO X Emissions modeling system, including:

[0146] Acquisition module: used to influence NO X The basic emission parameters obtain the historical operation data of thermal power units and establish the training set D0;

[0147] Simplification module: used to simplify the training set D0 using the density clustering algorithm and the sparse greedy matrix to obtain the sample representative set D2;

[0148] The first solution module is used to construct a semi-parametric support vector regression model based on the sample representative set D2, and solve the weight of the semi-parametric support vector regression model based on the iterative reweighted least squares method;

[0149] The second solution module is used to obtain the final weight of the semi-parametric support vector regression model based on the training set D0 by using the Apache Spark parallel computing framework for parallel design;

[0150] Output module: It is used to obtain the NO of thermal power units based on the final weights by parallelizing the semi-parametric support vector regression model using Apache Spark. X Emission model.

[0151] Embodiment three:

[0152] The embodiment of the present invention provides a thermal power unit NO X An emission modeling device, including a processor and a storage medium;

[0153] The storage medium is used to store instructions;

[0154] The processor is used to operate according to the instructions to execute the steps of the method described in embodiment 1.

[0155] Embodiment 4:

[0156] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0157] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0158] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0159] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0160] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0161] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A thermal power unit NO X An emission modeling method, characterized in that The method comprises the following steps: Based on the influence of NO X The basic emission parameters obtain the historical operation data of thermal power units and establish the training set D0; The training set D0 is simplified using the density clustering algorithm and the sparse greedy matrix to obtain the sample representative set D2; A semi-parametric support vector regression model is constructed based on the sample representative set D2, and the weights of the semi-parametric support vector regression model are solved based on the iterative reweighted least squares method; The construction of a semi-parametric support vector regression model based on the sample representative set D2 includes: The sample representative set D2 is used to estimate the parameter w of the support vector regression machine, and the parameter w is estimated by the following predefined model: In formula (1), N2 is the number of samples; X i is the input vector of the i-th learning training sample in the sample representative set, and X i ∈R o ×1 ; φ(·) represents a nonlinear mapping to a high-dimensional space; Ψ is defined by: In formula (1), Λ is the weight, which is expressed by the following formula: According to the samples in the sample representative set D2, the samples are fitted with a linear function without error under the accuracy ε, and the relaxation factor ξ is introduced i ≥0 and The construction of the semi-parametric support vector regression model is transformed into minimizing the structural risk function, which is expressed by the following formula: In formula (4), C is the penalty factor, b is the bias, ε is the accuracy; It is the kernel matrix of the sample input vector in the sample representative set D2, which is expressed by the following formula: In formula (5), the kernel function K(X i ,X j ) Select the Gaussian kernel function, which is expressed by the following formula: In formula (5), σ is the width parameter of the Gaussian kernel function; In formula (4), k i Defined by: In formula (7), Φ i is the input vector of the i-th sample; The optimal decision function of the semi-parametric support vector regression machine is obtained as follows: In order to obtain the weight Λ, formula (4) is rewritten as: Formula (9) is the model formula of the semi-parametric support vector regression machine; Using the Lagrangian function to process equation (9) we get the Lagrangian equation, and taking partial derivatives of Λ, b in the equation, we get: In formula (10), is the kernel matrix of the sample input vector in the training set D0 and the sample representative set D2; Vector1 = [1,…,1] T , Wherein, the method of solving the weight of the semi-parametric support vector regression model based on iterative reweighted least squares method includes: Give a i and The initial value of , solve equation (10) to get the initial value of weight Λ; According to the constraints in formula (9) and the initial value of weight Λ, update a i and According to the updated a i and The weight Λ is calculated again by formula (10); After executing k updates, if ||Λ k -Λ k-1 ||<10 -3 , end the iterative calculation and output the weight Λ; Based on the training set D0, the Apache Spark parallel computing framework is used for parallel design to obtain the final weights of the semi-parametric support vector regression model; The method of using the Apache Spark parallel computing framework for parallel design to obtain the final weight of the semi-parametric support vector regression model includes: Input the training set D0 into the HDFS distributed data storage system; On the master node, the training set D0 is divided into S evenly spaced sub-datasets, denoted as Distribute to S child nodes respectively; On each of the S child nodes, the sub-datasets are processed simultaneously. Use density clustering algorithm to determine the set of cluster centers Then use the sparse greedy matrix approximation algorithm to select the cluster center set Find the set that best represents the entire data set The weight Λ of the semi-parametric support vector regression model is calculated by using the iterative reweighted least squares algorithm on each of the S child nodes. 1 ,…,Λ S ; On each of the S child nodes, the model weight Λ 1 ,…,Λ S Find The support vectors in ; The support vectors on the child nodes are sent to the master node to be merged into a support vector set SVs. On the master node, SVs is used as the input of the semi-parametric support vector regression machine, and the iterative reweighted least squares algorithm is executed again to obtain the final weight Λ of the semi-parametric support vector regression machine model; Based on the final weights, the semi-parametric support vector regression model is parallelized using Apache Spark to obtain the NO X Emission model.

2. The thermal power unit NO according to claim 1 X An emission modeling method, characterized in that The step of establishing the training set D0 comprises: According to the combustion process, NO X Generation mechanism, determine the impact of NO X Input variables for emission values ​​u1,u2,...,u q ; Let the order of each input variable be o1, o2, ..., o q ; Establish training set D0 = {(Φ i ,y i ),i=1,…,N0}, where N0 is the number of samples, Φ i is the input vector of the i-th sample, Φ i Composed of previous output and input, Φ i ∈R o×1 .

3. The thermal power unit NO according to claim 2 X An emission modeling method, characterized in that The method of reducing the training set D0 by using a density clustering algorithm and a sparse greedy matrix includes: The density clustering algorithm is used to roughly simplify the training set D0, and the cluster center set D1 = {(Φ′ i ,y′ i ),i=1,…,N1}, where N1 is the number of cluster centers and N1 <N0;Φ′ i is the input vector of the i-th sample in the cluster center set, Φ′ i ∈R o×1 ;y′ i is the output of the i-th sample in the cluster center set, y′∈R; A sparse greedy matrix algorithm is used to extract a set of sample representative sets D2 that can maximally represent the entire data set from the cluster center set D1. i ,y i ),i=1,…N2}, where N2 is the number of samples, and N2 <N1;{X i , i=1,…,N2} is the basic sample input vector set, X i is the input vector of the i-th learning training sample in the sample representative set, and X i ∈R o×1 ;y i is the output of the i-th sample in the sample representative set, and y i ∈R.

4. The thermal power unit NO according to claim 1 X An emission modeling method, characterized in that The thermal power unit NO X Emissions models, including: Initialize the number of density cluster centers to N1, and give the root mean square error e of the modeling; According to the final weight Λ of the semi-parametric support vector regression model, a semi-parametric support vector regression model is obtained; Calculate the root mean square error of the model. If the error is greater than e, increase the number of cluster centers N1, reinitialize and calculate the final weight; otherwise, the modeling process ends and the thermal power unit NO is obtained. X Emission model.

5. A thermal power unit NO X An emission modeling system, characterized in that include: Acquisition module: used to influence the NO of thermal power units X The basic emission parameters obtain the historical operation data of thermal power units and establish the training set D0; Simplification module: used to simplify the training set D0 using the density clustering algorithm and the sparse greedy matrix to obtain the sample representative set D2; The first solution module is used to construct a semi-parametric support vector regression model based on the sample representative set D2, and solve the weight of the semi-parametric support vector regression model based on the iterative reweighted least squares method; The construction of a semi-parametric support vector regression model based on the sample representative set D2 includes: The sample representative set D2 is used to estimate the parameter w of the support vector regression machine, and the parameter w is estimated by the following predefined model: In formula (1), N2 is the number of samples; X i is the input vector of the i-th learning training sample in the sample representative set, and X i ∈R o ×1 ; φ(·) represents a nonlinear mapping to a high-dimensional space; Ψ is defined by: In formula (1), Λ is the weight, which is expressed by the following formula: According to the samples in the sample representative set D2, the samples are fitted with a linear function without error under the accuracy ε, and the relaxation factor ξ is introduced i ≥0 and The construction of the semi-parametric support vector regression model is transformed into minimizing the structural risk function, which is expressed by the following formula: In formula (4), C is the penalty factor, b is the bias, ε is the accuracy; It is the kernel matrix of the sample input vector in the sample representative set D2, which is expressed by the following formula: In formula (5), the kernel function K(X i ,X j ) Select the Gaussian kernel function, which is expressed by the following formula: In formula (5), σ is the width parameter of the Gaussian kernel function; In formula (4), k i Defined by: In formula (7), Φ i is the input vector of the i-th sample; The optimal decision function of the semi-parametric support vector regression machine is obtained as follows: In order to obtain the weight Λ, formula (4) is rewritten as: Formula (9) is the model formula of the semi-parametric support vector regression machine; Using the Lagrangian function to process equation (9) we get the Lagrangian equation, and taking the partial derivative of Λ, b in the equation, we get: In formula (10), is the kernel matrix of the sample input vector in the training set D0 and the sample representative set D2; Vector1 = [1,…,1] T , Wherein, the method of solving the weight of the semi-parametric support vector regression model based on iterative reweighted least squares method includes: Give a i and The initial value of , solve equation (10) to get the initial value of weight Λ; According to the constraints in formula (9) and the initial value of weight Λ, update a i and According to the updated a i and The weight Λ is calculated again by formula (10); After executing k updates, if ||Λ k -Λ k-1 ||<10 -3 , end the iterative calculation and output the weight Λ; The second solution module is used to obtain the final weight of the semi-parametric support vector regression model based on the training set D0 by using the Apache Spark parallel computing framework for parallel design; The method of using the Apache Spark parallel computing framework for parallel design to obtain the final weight of the semi-parametric support vector regression model includes: Input the training set D0 into the HDFS distributed data storage system; On the master node, the training set D0 is divided into S evenly spaced sub-datasets, denoted as Distribute to S child nodes respectively; On each of the S child nodes, the sub-datasets are processed simultaneously. Use density clustering algorithm to determine the set of cluster centers Then use the sparse greedy matrix approximation algorithm to select the cluster center set Find the set that best represents the entire data set The weight Λ of the semi-parametric support vector regression model is calculated by using the iterative reweighted least squares algorithm on each of the S child nodes. 1 ,…,Λ S ; On each of the S child nodes, the model weight Λ 1 ,…,Λ S Find The support vectors in ; The support vectors on the child nodes are sent to the master node to be merged into a support vector set SVs. On the master node, SVs is used as the input of the semi-parametric support vector regression machine, and the iterative reweighted least squares algorithm is executed again to obtain the final weight Λ of the semi-parametric support vector regression machine model; Output module: It is used to obtain the NO of thermal power units based on the final weights by parallelizing the semi-parametric support vector regression model using Apache Spark. X Emission model.

6. A thermal power unit NO X An emission modeling device, characterized in that including processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images