CN104504292A

CN104504292A - Method for predicting optimum working temperature of circulating fluidized bed boiler based on BP neural network

Info

Publication number: CN104504292A
Application number: CN201510018649.0A
Authority: CN
Inventors: 申涛; 任万杰; 栾维磊; 刘晓璞
Original assignee: University of Jinan
Current assignee: University of Jinan
Priority date: 2015-01-14
Filing date: 2015-01-14
Publication date: 2015-04-08

Abstract

The invention discloses a method for predicting the optimal working temperature of a circulating fluidized bed boiler based on a BP neural network. The optimal working temperature of the circulating fluidized bed boiler is taken as the output of the BP neural network; the historical data of the site is recorded and stored and filtered, and these data are selected as the training set samples to determine the number of nodes in the input layer of the BP neural network, hidden The number, weight and threshold parameters of layer-containing nodes; through the BP neural network method combined with the input parameter analysis and calculation, the predicted optimal working temperature of the fluidized bed boiler in the thermal power plant is obtained; the simulation test is carried out, and the predicted results will be compared with the field The actual results of the fluidized bed are compared and analyzed; the fluidized bed boiler operator's judgment deviation of the optimal working temperature of the boiler is solved, and the optimal working temperature of the fluidized bed has large fluctuations and poor stability.

Description

Method of Predicting Optimum Working Temperature of Circulating Fluidized Bed Boiler Based on BP Neural Network

技术领域technical field

本发明涉及一种基于BP神经网络预测循环流化床锅炉最佳工作温度的方法。The invention relates to a method for predicting the optimal working temperature of a circulating fluidized bed boiler based on a BP neural network.

背景技术Background technique

循环流化床锅炉技术是近十几年来迅速发展的一项高效低污染清洁燃烧技术。国际上这项技术在电站锅炉、工业锅炉和废弃物处理利用等领域已得到广泛的商业应用，并向几十万千瓦级规模的大型循环流化床锅炉发展；国内在这方面的研究、开发和应用也逐渐兴起，已有上百台循环流化床锅炉投入运行或正在制造之中。未来几年将是循环流化床飞速发展的时期。Circulating fluidized bed boiler technology is a high-efficiency, low-pollution and clean combustion technology that has developed rapidly in the past ten years. Internationally, this technology has been widely used commercially in the fields of power station boilers, industrial boilers and waste treatment and utilization, and is developing towards large-scale circulating fluidized bed boilers with a scale of hundreds of thousands of kilowatts; domestic research and development in this area And applications are also gradually emerging, and hundreds of circulating fluidized bed boilers have been put into operation or are being manufactured. The next few years will be a period of rapid development of circulating fluidized bed.

循环流化床锅炉系统通常由流化床燃烧室(炉膛)、循环灰分离器、飞灰回送装置、尾部受热面和辅助设备等组成。循环流化床锅炉系统通常由燃烧系统和汽水系统所组成，燃料在锅炉的燃烧系统中完成燃烧过程，循环流化床的燃料及脱硫剂经多次循环、反复地进行脱硫反应，具有低NO_x排放量，脱硫效率高，而且具有燃料适应性广、负荷调节性能好、灰渣易于综合利用等优点，在国内以及国际上使用较广泛，推广较迅速。The circulating fluidized bed boiler system usually consists of a fluidized bed combustion chamber (furnace), a circulating ash separator, a fly ash return device, a rear heating surface and auxiliary equipment. The circulating fluidized bed boiler system is usually composed of a combustion system and a steam-water system. The fuel completes the combustion process in the combustion system of the boiler. The fuel and desulfurizer in the circulating fluidized bed undergo repeated desulfurization reactions through multiple cycles, with low NO _X emissions, high desulfurization efficiency, wide fuel adaptability, good load regulation performance, easy comprehensive utilization of ash and residue, etc., are widely used at home and abroad, and promoted rapidly.

床层温度是一个直接影响锅炉能否安全连续运行的重要控制参数，同时也直接影响着锅炉运行中的脱硫效率及NO_x的产生量。操作员对床层工作温度的设定带有很强的随意性，考虑的因素过少，往往导致流化床锅炉的床层温度忽高忽低，进而燃料无法充分燃烧，脱硫效率较低，造成不必要的环境污染。The bed temperature is an important control parameter that directly affects the safe and continuous operation of the boiler, and also directly affects the desulfurization efficiency and _NOx production during the boiler operation. The operator sets the working temperature of the bed with strong arbitrariness. Too few factors are considered, which often leads to high and low bed temperature of the fluidized bed boiler, and then the fuel cannot be fully combusted, and the desulfurization efficiency is low. cause unnecessary environmental pollution.

循环流化床锅炉的燃烧控制较为复杂，是一个具有强干扰、非线性、时变、多变量相关联的过程，燃料量、石灰石量、一次风量、二次风量等现场变量都对燃烧过程有影响，控制精度低，对锅炉的最佳工作温度的预测是难点，在实际的生产过程中确定流化床锅炉的最佳工作温度是提高燃烧效率，提高脱硫效率的关键，目前调节锅炉的工作温度的设定主要通过操作人员的经验，存在以下不足：The combustion control of circulating fluidized bed boilers is relatively complicated, and it is a process with strong interference, nonlinearity, time-varying, and multi-variable correlations. Field variables such as fuel volume, limestone volume, primary air volume, and secondary air volume all have an impact on the combustion process. It is difficult to predict the best working temperature of the boiler. In the actual production process, determining the best working temperature of the fluidized bed boiler is the key to improving the combustion efficiency and desulfurization efficiency. At present, the work of adjusting the boiler The setting of the temperature is mainly through the experience of the operator, and there are the following deficiencies:

1、操作员的主观臆断性太强；1. The operator is too subjective;

2、操作员的操作具有明显的滞后性；2. The operator's operation has obvious hysteresis;

3、硫化床工作温度波动较大、稳定性差；3. The working temperature of the fluidized bed fluctuates greatly and the stability is poor;

4、煤的燃烧率达不到理想的最高值，浪费能源。4. The combustion rate of coal cannot reach the ideal maximum value, wasting energy.

发明内容Contents of the invention

本发明为了解决上述问题，提出了一种基于BP神经网络预测循环流化床锅炉最佳工作温度的方法，本方法利用BP神经网络方法预测热电厂流化床锅炉的最佳工作温度，进而能够给操作人员提供流化床锅炉的最佳工作温度这一关键参数，提高燃烧效率，提高脱硫效率，减少硫化物的排放量，达到节能减排的目的。In order to solve the above-mentioned problems, the present invention proposes a method for predicting the optimal operating temperature of a circulating fluidized bed boiler based on a BP neural network. The operator provides the key parameter of the optimal working temperature of the fluidized bed boiler to improve the combustion efficiency, improve the desulfurization efficiency, reduce the emission of sulfide, and achieve the purpose of energy saving and emission reduction.

为了实现上述目的，本发明采用如下技术方案：In order to achieve the above object, the present invention adopts the following technical solutions:

一种基于BP神经网络预测循环流化床锅炉最佳工作温度的方法，包括以下步骤：A method for predicting the optimal operating temperature of a circulating fluidized bed boiler based on a BP neural network, comprising the following steps:

(1)根据热电厂循环流化床锅炉的实际运行情况，选取燃料量x₁、石灰石量x₂、一次分量x₃、二次风量x₄作为BP神经网络模型的输入，将循环流化床锅炉的最佳工作温度作为BP神经网络的输出；(1) According to the actual operation of the circulating fluidized bed boiler in the thermal power plant, the fuel quantity x ₁ , the limestone quantity x ₂ , the primary component x ₃ and the secondary air volume x ₄ are selected as the input of the BP neural network model, and the circulating fluidized bed boiler The optimal working temperature of is used as the output of the BP neural network;

(2)记录并存储现场的历史数据并作滤波处理，选取这些数据作为训练集样本，确定BP神经网络的输入层节点个数、隐含层结点个数、权值和阈值参数；(2) record and store the historical data of the scene and perform filtering process, select these data as the training set sample, determine the input layer node number, hidden layer node number, weight and threshold value parameters of BP neural network;

(3)通过BP神经网络的方法结合输入参数进行分析计算，得到预测出的热电厂流化床锅炉的最佳工作温度；(3) Analyzing and calculating through the method of BP neural network combined with the input parameters to obtain the predicted optimal working temperature of the fluidized bed boiler in the thermal power plant;

(4)进行仿真测试，将通过预测的结果与现场的实际结果进行比对分析。(4) Carry out the simulation test, and compare and analyze the predicted results with the actual results on site.

所述步骤(1)中，具体方法为：根据热电厂循环流化床锅炉的实际运行情况，分析相关的输入输出量，通过仿真实验，筛选出能够对循环流化床最佳工作温度进行分析的变量，作为BP神经网络模型的输入，将循环流化床锅炉的最佳工作温度作为BP神经网络的输出，最终选取燃料量x₁、石灰石量x₂、一次分量x₃、二次风量x₄作为输入，流化床最佳工作温度y₁作为输出。In the step (1), the specific method is: according to the actual operation of the circulating fluidized bed boiler in the thermal power plant, analyze the relevant input and output quantities, and screen out the best working temperature of the circulating fluidized bed through the simulation experiment. variable, as the input of the BP neural network model, and the optimal operating temperature of the circulating fluidized bed boiler as the output of the BP neural network, and finally select the fuel amount x ₁ , the limestone amount x ₂ , the primary component x ₃ , and the secondary air volume x ₄ As input, the fluidized bed optimal working temperature _y1 as output.

所述步骤(2)中，具体方法包括：In described step (2), specific method comprises:

(a)记录并存储现场的历史数据并作滤波处理，选取这些数据作为训练集样本，训练集样本共有4组，分别为燃料量训练集、石灰石量训练集、一次风量训练集、二次风量训练集；(a) Record and store the historical data of the site and perform filtering processing. Select these data as training set samples. There are 4 groups of training set samples, namely fuel volume training set, limestone volume training set, primary air volume training set, and secondary air volume Training set;

(b)定义BP神经网络的输入层的节点个数为n，由前面的分析得到n＝4，隐含层的节点个数为q，输入层和隐含层的权值为ν_ki(k＝1,2,…,q；i＝1,2,…,n)，阈值为θ_i(i＝1,2,…,n)，输出层的节点个数为m，可知m＝1，隐含层和输出层的权值为ω_jk(j＝1,2,…,m；k＝1,2,…,q)，阈值为f₁(·)为隐含层的传递函数，f₂(·)为输出层的传递函数。(b) Define the number of nodes in the input layer of the BP neural network as n, get n=4 from the previous analysis, the number of nodes in the hidden layer is q, and the weights of the input layer and the hidden layer are ν _ki (k =1,2,…,q; i=1,2,…,n), the threshold is θ _i (i=1,2,…,n), the number of nodes in the output layer is m, we know that m=1, The weights of the hidden layer and the output layer are ω _jk (j=1,2,…,m; k=1,2,…,q), and the threshold is f ₁ (·) is the transfer function of the hidden layer, and f ₂ (·) is the transfer function of the output layer.

所述步骤(a)中，4组样本数量都确定为165。In the step (a), the number of samples in the four groups is determined to be 165.

所述步骤(3)中，具体的分析计算方法为：计算隐含层节点的输出和输出层节点的输出，定义神经网络的期望值输出，并将计算的输出误差展开到隐含层和输入层中，按照使BP神经网络的权值和负梯度成正比的方法调整权值，训练网络。In the described step (3), the specific analysis and calculation method is: calculate the output of the hidden layer node and the output of the output layer node, define the expected value output of the neural network, and expand the output error of the calculation to the hidden layer and the input layer Among them, adjust the weight according to the method of making the weight of the BP neural network proportional to the negative gradient, and train the network.

所述步骤(3)中，隐含层节点的输出为：In described step (3), the output of hidden layer node is:

${z z}_{k k} = = {f f}_{11} (({Σ Σ}_{i i = = 00}^{n no} {v v}_{ki the ki} {x x}_{i i} - - {θ θ}_{k k})) - - - - - - ((11))$

上式中k＝1,2,…,q；In the above formula, k=1,2,...,q;

输出层节点的输出为：The output of the output layer node is:

上式中j＝1,2,…,m；In the above formula, j=1,2,...,m;

设神经网络的期望值输出为d＝(d₁,…,d_j,…,d_m)，网络输出和期望值的输出不一致时，会存在输出误差，我们定义输出误差E如下：Suppose the expected value output of the neural network is d=(d ₁ ,…,d _j ,…,d _m ), when the output of the network is inconsistent with the output of the expected value, there will be an output error. We define the output error E as follows:

$E E. = = \frac{11}{22} {((d d - - y the y))}^{22} = = \frac{11}{22} {Σ Σ}_{j j = = 11}^{m m} {(({d d}_{j j} - - {y the y}_{j j}))}^{22} - - - - - - ((33))$

将式(3)展开到隐含层，有：Expanding formula (3) to the hidden layer, we have:

将式(4)展开到输入层，有：Expanding formula (4) to the input layer, we have:

所述步骤(3)中，根据上式(5)得到，E是ν_ki、ω_jk的函数，如果改变ν_ki、ω_jk的值，那么误差E的值也会发生改变；BP神经网络的最优结果就是使E变得尽量小以满足我们的要求，使BP神经网络的权值和负梯度成正比，在确定流化床锅炉的最佳工作温度的过程中，就是按照这样的方法调整权值，即：In the step (3), according to the above formula (5), E is a function of ν _ki and ω _jk , if the values of ν _ki and ω _jk are changed, the value of the error E will also change; the BP neural network The optimal result is to make E as small as possible to meet our requirements, so that the weight of the BP neural network is proportional to the negative gradient. In the process of determining the optimal working temperature of the fluidized bed boiler, it is adjusted according to this method Weights, namely:

${Δv Δv}_{ki the ki} = = - - η η \frac{&PartialD; &PartialD; E E.}{{&PartialD; &PartialD; v v}_{ki the ki}} - - - - - - ((66))$

上式中的k＝1,2,…,q，i＝1,2,…,n；k=1,2,...,q in the above formula, i=1,2,...,n;

${Δω Δω}_{jk jk} = = - - η η \frac{&PartialD; &PartialD; E E.}{{&PartialD; &PartialD; ω ω}_{jk jk}} - - - - - - ((77))$

上式中的j＝1,2,…,m，k＝1,2,…,q；j=1,2,...,m in the above formula, k=1,2,...,q;

式(6)和(7)中的负号表示梯度下降，η∈(0,1)表示比例系数，它反映了学习的速度。The negative sign in Equations (6) and (7) indicates gradient descent, and η∈(0,1) indicates the scaling factor, which reflects the speed of learning.

所述步骤(3)中，训练网络的具体方法为：把隐含层节点个数设置为1，训练网络，然后逐步增加节点个数，用相同的样本训练，确定误差最小的时候对应的节点数；用此方法时，借用一些经验公式；经验公式得出的节点数是粗略的估计值，作为试凑法的最初值。In the step (3), the specific method of training the network is: the number of hidden layer nodes is set to 1, the training network, and then gradually increase the number of nodes, with the same sample training, to determine the corresponding node when the error is the smallest When using this method, some empirical formulas are used; the number of nodes obtained by the empirical formula is a rough estimate, which is used as the initial value of the trial and error method.

先试最初值，如果不行，节点数加1，接着再试，如果还不行，接着加；数值越大收敛越快；一直到无法加快收敛速度时就得到了合适的节点数，常用的经验公式如下：Try the initial value first, if it doesn’t work, add 1 to the number of nodes, and then try again, if it still doesn’t work, then add it; the larger the value, the faster the convergence; until the convergence speed cannot be accelerated, the appropriate number of nodes is obtained, the commonly used empirical formula as follows:

$q q = = \sqrt{n no + + l l} + + α α - - - - - - ((88))$

q＝log₂n (9)q=log ₂ n (9)

$q q = = \sqrt{nl nl} - - - - - - ((1010))$

$q q = = \sqrt{0.43 0.43 qn qn + + {0.12 0.12 q q}^{22} + + 2.54 2.54 n no + + 0.77 0.77 q q + + 0.35 0.35} + + 0.51 0.51 - - - - - - ((1111))$

上述经验公式中，n为输入层向量维数，l为输出层向量维数，α是1与10之间的常数，计算完成后得到的m值四舍五入取整数。In the above empirical formula, n is the vector dimension of the input layer, l is the vector dimension of the output layer, α is a constant between 1 and 10, and the value of m obtained after the calculation is rounded to an integer.

所述步骤(3)中，确定隐含层的节点个数时必须满足下列两个基本条件：In the step (3), the following two basic conditions must be met when determining the number of nodes in the hidden layer:

(1)训练样本数一定要多于模型连接权数，为2-10倍；否则的话，样本需要分成几个部分，采取“轮流训练”的办法，以此来得到性能良好的的网络模型；(1) The number of training samples must be more than the number of model connection weights, which is 2-10 times; otherwise, the samples need to be divided into several parts, and the "rotational training" method is adopted to obtain a network model with good performance ;

(2)隐含层的节点个数要小于N-1(为N训练样本数)；否则的话，模型的误差与样本特性无关，趋于零，也就是说建立的模型无泛化能力；相同的道理，输入层节点的个数也要小于N-1。(2) The number of nodes in the hidden layer should be less than N-1 (the number of training samples for N); otherwise, the error of the model has nothing to do with the sample characteristics and tends to zero, that is to say, the established model has no generalization ability; the same The reason, the number of input layer nodes should be less than N-1.

本发明的有益效果为：The beneficial effects of the present invention are:

(1)从仿真结果可以看出通过BP神经网络预测得到的流化床锅炉最佳工作温度与现场得到的最佳工作温度误差在±3％以内，且变化趋势一致，说明本发明通过BP神经网络方法很好的建立了流化床锅炉的模型，并能够预测出最佳的工作温度；(1) From the simulation results, it can be seen that the optimal operating temperature of the fluidized bed boiler predicted by the BP neural network and the optimal operating temperature error obtained on the spot are within ± 3%, and the trend of change is consistent. The network method establishes the model of the fluidized bed boiler very well, and can predict the best working temperature;

(2)能够给操作人员提供流化床锅炉最佳工作温度这一关键参数，提高燃烧效率，提高脱硫效率，减少硫化物的排放量，达到节能减排的目的；(2) It can provide the operator with the key parameter of the optimal working temperature of the fluidized bed boiler, improve combustion efficiency, improve desulfurization efficiency, reduce sulfide emissions, and achieve the purpose of energy saving and emission reduction;

(3)解决了流化床锅炉操作员对锅炉的最佳工作温度的判断偏差，以及流化床的最佳工作温度的波动大、稳定性差等问题。(3) The fluidized bed boiler operator's judgment deviation of the optimum working temperature of the boiler, as well as the large fluctuation and poor stability of the optimum working temperature of the fluidized bed are solved.

附图说明Description of drawings

图1为本发明的BP神经网络的结构示意图。Fig. 1 is the structural representation of the BP neural network of the present invention.

图2为本发明的程序流程图；Fig. 2 is a program flow chart of the present invention;

图3(a)为本发明选取的燃料量经均值滤波处理后图形；Fig. 3 (a) is that the fuel quantity that the present invention chooses is processed through mean value filtering figure;

图3(b)为本发明选取的石灰石量经均值滤波处理后图形；Fig. 3 (b) is that the amount of limestone that the present invention chooses is processed figure after mean filtering;

图3(c)为本发明选取的一次风量经均值滤波处理后图形；Fig. 3 (c) is that the primary air volume selected by the present invention is processed by means of filtering;

图3(d)为本发明选取的二次风量经均值滤波处理后图形；Fig. 3 (d) is that the secondary air volume that the present invention chooses is processed through mean value filtering figure;

图4为本发明的效果比对图。Fig. 4 is the comparison diagram of the effect of the present invention.

具体实施方式：Detailed ways:

下面结合附图与实施例对本发明作进一步说明。The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

如图2所示，利用BP神经网络方法预测热电厂硫化床锅炉的最佳工作温度，具体包括以下步骤：As shown in Figure 2, using the BP neural network method to predict the optimal working temperature of the fluidized bed boiler in the thermal power plant includes the following steps:

步骤1，根据热电厂循环流化床锅炉的实际运行情况，分析相关的输入输出量，通过仿真实验，筛选出能够对循环流化床最佳工作温度进行分析的变量，作为BP神经网络模型的输入，将循环流化床锅炉的最佳工作温度作为BP神经网络的输出。最终选取燃料量x₁、石灰石量x₂、一次分量x₃、二次风量x₄作为输入，流化床最佳工作温度y₁作为输出。Step 1. According to the actual operation of the circulating fluidized bed boiler in the thermal power plant, analyze the relevant input and output quantities, and through the simulation experiment, select the variables that can analyze the optimal working temperature of the circulating fluidized bed as the input of the BP neural network model , taking the optimum operating temperature of the circulating fluidized bed boiler as the output of the BP neural network. Finally, the fuel quantity x ₁ , the limestone quantity x ₂ , the primary component x ₃ , and the secondary air volume x ₄ are selected as the input, and the optimum fluidized bed temperature y ₁ is taken as the output.

步骤2，确定训练集样本、输入层节点个数、隐含层结点个数、权值、阈值等BP神经网络参数，并通过BP神经网络的方法进行分析计算，最终得到预测出的热电厂流化床锅炉的最佳工作温度。Step 2, determine the BP neural network parameters such as the training set samples, the number of input layer nodes, the number of hidden layer nodes, weights, thresholds, etc., and analyze and calculate through the BP neural network method, and finally obtain the predicted thermal power plant flow Optimum working temperature of bed boiler.

步骤3，进行仿真测试，得出该发明专利的效果。Step 3, conduct a simulation test to obtain the effect of the invention patent.

所述步骤2中确定训练集样本、隐含层节点个数、权值等BP神经网络参数，并通过BP神经网络的方法进行分析计算，其具体步骤为：In described step 2, determine the BP neural network parameters such as training set sample, hidden layer node number, weight, and analyze and calculate by the method of BP neural network, its specific steps are:

(a)通过软件记录并存储现场的历史数据并作滤波处理，选取这些数据作为训练集样本，训练集样本共有4组，分别为燃料量训练集、石灰石量训练集、一次风量训练集、二次风量训练集，仿真分析结果表明，4组样本数量都确定为165。(a) Use the software to record and store the historical data of the site and perform filtering processing. These data are selected as training set samples. For the secondary air volume training set, the simulation analysis results show that the number of samples in the four groups is determined to be 165.

(b)定义BP神经网络的输入层的节点个数为n，由前面的分析得到n＝4，隐含层的节点个数为q，输入层和隐含层的权值为ν_ki(k＝1,2,…,q；i＝1,2,…,n)，阈值为θ_i(i＝1,2,…,n)，输出层的节点个数为m，可知m＝1，隐含层和输出层的权值为ω_jk(j＝1,2,…,m；k＝1,2,…,q)，阈值为f₁(·)为隐含层的传递函数，f₂(·)为输出层的传递函数，如图1所示。(b) Define the number of nodes in the input layer of the BP neural network as n, get n=4 from the previous analysis, the number of nodes in the hidden layer is q, and the weights of the input layer and the hidden layer are ν _ki (k =1,2,…,q; i=1,2,…,n), the threshold is θ _i (i=1,2,…,n), the number of nodes in the output layer is m, we know that m=1, The weights of the hidden layer and the output layer are ω _jk (j=1,2,…,m; k=1,2,…,q), and the threshold is f ₁ (·) is the transfer function of the hidden layer, and f ₂ (·) is the transfer function of the output layer, as shown in Figure 1.

下面是具体的分析计算过程：The following is the specific analysis and calculation process:

1)隐含层节点的输出为：1) The output of the hidden layer node is:

上式中k＝1,2,…,q。In the above formula, k=1,2,...,q.

输出层节点的输出为：The output of the output layer node is:

上式中j＝1,2,…,m。In the above formula, j=1,2,...,m.

2)根据上式(5)我们可以得到，E是ν_ki、ω_jk的函数，如果改变ν_ki、ω_jk的值，那么误差E的值也会发生改变。BP神经网络的最优结果就是使E变得尽量小以满足我们的要求，使BP神经网络的权值和负梯度成正比，在确定流化床锅炉的最佳工作温度的过程中，就是按照这样的方法调整权值，即：2) According to the above formula (5), we can get that E is a function of ν _ki and ω _jk . If the values of ν _ki and ω _jk are changed, the value of error E will also change. The optimal result of the BP neural network is to make E as small as possible to meet our requirements, so that the weight of the BP neural network is proportional to the negative gradient. In the process of determining the optimal working temperature of the fluidized bed boiler, it is according to Such a method adjusts the weights, namely:

上式中的k＝1,2,…,q，i＝1,2,…,n。In the above formula, k=1,2,...,q, i=1,2,...,n.

上式中的j＝1,2,…,m，k＝1,2,…,q。In the above formula, j=1, 2,..., m, k=1, 2,..., q.

3)把隐含层节点个数设置为很小，训练网络，看结果，然后逐步增加节点个数，用相同的样本训练，确定误差最小的时候对应的节点数。用此方法时，我们可以借用一些经验公式。经验公式得出的节点数是粗略的估计值，这就可以作为试凑法的最初值。先试最初值，如果不行，节点数加1，接着再试，如果还不行，接着加。数值越大收敛越快。一直到无法加快收敛速度时就得到了合适的节点数，常用的经验公式如下：3) Set the number of hidden layer nodes to be very small, train the network, see the results, then gradually increase the number of nodes, train with the same sample, and determine the corresponding number of nodes when the error is the smallest. When using this method, we can borrow some empirical formulas. The number of nodes obtained by the empirical formula is a rough estimate, which can be used as the initial value of the trial and error method. Try the initial value first, if it doesn't work, add 1 to the number of nodes, and then try again, if it still doesn't work, then add it. The larger the value, the faster the convergence. Until the convergence speed cannot be accelerated, the appropriate number of nodes is obtained. The commonly used empirical formula is as follows:

$q q = = \sqrt{n no + + l l} + + α α - - - - - - ((88))$

q＝log₂n (9)q=log ₂ n (9)

$q q = = \sqrt{nl nl} - - - - - - ((1010))$

根据上面的分析，可以得到确定隐含层的节点个数时必须满足下列两个基本条件。According to the above analysis, it can be obtained that the following two basic conditions must be met when determining the number of nodes in the hidden layer.

(1)训练样本数一定要多于模型连接权数，一般情况是2-10倍。否则的话，样本需要分成几个部分，采取“轮流训练”的办法，以此来得到性能良好的的网络模型。(1) The number of training samples must be more than the number of model connection weights, generally 2-10 times. Otherwise, the sample needs to be divided into several parts, and the "rotational training" method is adopted to obtain a network model with good performance.

(2)隐含层的节点个数要小于N-1(为N训练样本数)。否则的话，模型的误差与样本特性无关，趋于零，也就是说建立的模型无泛化能力。相同的道理，输入层节点的个数也要小于N-1。(2) The number of nodes in the hidden layer should be less than N-1 (N is the number of training samples). Otherwise, the error of the model has nothing to do with the sample characteristics and tends to zero, which means that the established model has no generalization ability. For the same reason, the number of input layer nodes should be less than N-1.

至此，完成了BP神经网络的参数确定以及分析计算的过程。So far, the process of parameter determination and analysis and calculation of BP neural network is completed.

步骤1，根据热电厂循环流化床锅炉的实际运行情况，分析相关的输入输出量，通过仿真实验，筛选出能够对循环流化床最佳工作温度进行分析的变量，作为BP神经网络模型的输入，最终选取燃料量x₁、石灰石量x₂、一次分量x₃、二次风量x₄作为输入，数据个数都分别为165个，现场采集的历史数据曲线经均值滤波处理后如图3(a)-图3(d)所示。Step 1. According to the actual operation of the circulating fluidized bed boiler in the thermal power plant, analyze the relevant input and output quantities, and through the simulation experiment, select the variables that can analyze the optimal working temperature of the circulating fluidized bed as the input of the BP neural network model , and finally select fuel volume x ₁ , limestone volume x ₂ , primary component x ₃ , and secondary air volume x ₄ as inputs, and the number of data is 165 respectively. a)-shown in Figure 3(d).

步骤2，确定训练集样本、输入层节点个数、隐含层结点个数、权值、阈值等BP神经网络参数，并通过BP神经网络的方法进行分析计算，最终得到预测出的热电厂流化床锅炉的最佳工作温度。据公式(8)得出隐含层节点个数的范围是4到14，据式(9)得出隐含层节点个数为6，据式(10)得出隐含层节点个数为6，据式(11)得出隐含层节点个数为9。因此以下章节我们在研究时，首先选用隐含层节点个数为6测试，然后逐渐增加进而达到最优的性能。Step 2, determine the BP neural network parameters such as the training set samples, the number of input layer nodes, the number of hidden layer nodes, weights, thresholds, etc., and analyze and calculate through the BP neural network method, and finally obtain the predicted thermal power plant flow Optimum working temperature of bed boiler. According to formula (8), the number of hidden layer nodes ranges from 4 to 14; according to formula (9), the number of hidden layer nodes is 6; according to formula (10), the number of hidden layer nodes is 6. According to formula (11), the number of hidden layer nodes is 9. Therefore, in the following chapters, when we study, we first select the number of hidden layer nodes as 6 for testing, and then gradually increase to achieve the optimal performance.

根据以上章节我们得知训练集样本个数为165，测试集样本个数为50，选择双曲正切S型函数作为隐含层传递函数，根据经验初始连接权值一般在区间(-1，1)之间，这里我们选择(-0.3，0.3)之间的随机数作为初始连接权值，我们限制最大的网络训练次数为10000。According to the above chapters, we know that the number of samples in the training set is 165, and the number of samples in the test set is 50. The hyperbolic tangent S-type function is selected as the hidden layer transfer function. According to experience, the initial connection weight is generally in the interval (-1, 1 ), here we choose a random number between (-0.3, 0.3) as the initial connection weight, and we limit the maximum number of network training to 10,000.

为了更好地验证仿真效果，我们定义以下性能指标。In order to better verify the simulation effect, we define the following performance indicators.

(1)定义误差均方根RMSE的值为：(1) Define the value of root mean square error RMSE:

$RMSE RMSE = = \sqrt{{Σ Σ}_{i i = = 11}^{N N} \frac{{(({y the y}_{i i} - - {\overset{^^}{y the y}}_{i i}))}^{22}}{N N}} - - - - - - ((1212))$

(2)定义相对误差均值meanA_E的计算公式为：(2) Define the calculation formula of relative error mean A _E as:

${meanA meanA}_{E E.} = = \frac{11}{N N} {Σ Σ}_{i i = = 11}^{n no} | | {y the y}_{i i} - - {\overset{^^}{y the y}}_{i i} | | - - - - - - ((1313))$

上述式(12)和(13)中N为样本数据个数，y_i为流化床锅炉实际最佳设定温度，为使用BP神经网络仿真分析得到的预测最佳设定温度。In the above formulas (12) and (13), N is the number of sample data, y _i is the actual best set temperature of the fluidized bed boiler, The best set temperature is predicted by using BP neural network simulation analysis.

当隐含层节点个数取不同值时，BP神经网络的各项性能指标如表1所示。When the number of hidden layer nodes takes different values, the performance indicators of BP neural network are shown in Table 1.

表1 隐含层节点数对模型性能的影响Table 1 The influence of the number of hidden layer nodes on the performance of the model

从表1可以看出，随着BP神经网络的隐含层节点个数的增加，模型的训练误差均方根逐渐减少，训练相对误差均值也逐渐减少，训练时间呈现出逐渐增加的趋势。但是在隐含层的节点个数为9时，该模型的测试误差均方根与测试相对误差均值达到最小，此时训练误差均方根为0.8575，训练相对误差均值为0.0013，测试误差均方根为1.9624，测试相对误差均值为0.0028。但是从这以后，随着隐含层节点个数的继续增加，训练误差均方根随之减少，训练相对误差均值也随着减少，但是测试误差均方根却随着增大了，测试相对误差均值也随着增大了。这说明泛化能力开始下降，即出现“过拟合”现象。It can be seen from Table 1 that with the increase in the number of hidden layer nodes of the BP neural network, the root mean square of the training error of the model gradually decreases, the mean value of the relative training error also gradually decreases, and the training time shows a trend of gradually increasing. However, when the number of nodes in the hidden layer is 9, the root mean square error of the test and the mean value of the relative test error of the model reach the minimum. At this time, the mean square value of the training error is 0.8575, the mean value of the training relative error is 0.0013, and the mean square value of the test error is 0.8575. The root is 1.9624, and the mean relative error of the test is 0.0028. But since then, as the number of hidden layer nodes continues to increase, the root mean square of the training error decreases, and the mean value of the relative error of the training also decreases, but the root mean square of the test error increases. The mean error also increases. This shows that the generalization ability begins to decline, that is, the phenomenon of "overfitting" occurs.

根据性能指标可得单隐含层就能满足要求，因此，基于BP神经网络的隐含层节点数确定为9，网络结构为：According to the performance index, a single hidden layer can meet the requirements. Therefore, the number of hidden layer nodes based on BP neural network is determined to be 9, and the network structure is:

(1)输入到隐层权值：(1) Input to hidden layer weights:

$v v = = [\begin{matrix} - - 6.0404 6.0404 & 20.5921 20.5921 & - - 12.8676 12.8676 & 5.6208 5.6208 & 1.6882 1.6882 & 20.9227 20.9227 & 21.2285 21.2285 & - - 15.4897 15.4897 \\ 11.2562 11.2562 & 10.1158 10.1158 & - - 4.3327 4.3327 & - - 9.2332 9.2332 & 17.5343 17.5343 & 3.0811 3.0811 & - - 9.2746 9.2746 & 5.7843 5.7843 \\ - - 12.7889 12.7889 & - - 13.3140 13.3140 & - - 23.8600 23.8600 & 11.0918 11.0918 & - - 22.2564 22.2564 & 6.6677 6.6677 & 14.0212 14.0212 & 3.1021 3.1021 \\ 0.0587 0.0587 & 1.5509 1.5509 & - - 10.1367 10.1367 & - - 16.4317 16.4317 & - - 13.0337 13.0337 & - - 26.1119 26.1119 & - - 16.1807 16.1807 & 1.2229 1.2229 \\ 6.9535 6.9535 & - - 16.3618 16.3618 & 12.0347 12.0347 & 12.6808 12.6808 & 5.6723 5.6723 & 24.6739 24.6739 & 21.8484 21.8484 & - - 13.0471 13.0471 \\ - - 0.6129 0.6129 & - - 16.1455 16.1455 & 5.6687 5.6687 & - - 4.7238 4.7238 & 33.4284 33.4284 & - - 12.6362 12.6362 & - - 21.8464 21.8464 & 10.7048 10.7048 \\ - - 11.6364 11.6364 & - - 38.4894 38.4894 & - - 23.4389 23.4389 & - - 14.9062 14.9062 & 38.8148 38.8148 & - - 0.2016 0.2016 & 21.2143 21.2143 & 12.7493 12.7493 \\ - - 21.6638 21.6638 & - - 7.0040 7.0040 & - - 6.7117 6.7117 & 4.5349 4.5349 & - - 18.4027 18.4027 & - - 14.4674 14.4674 & - - 1.9573 1.9573 & - - 6.4926 6.4926 \\ 28.4760 28.4760 & - - 4.1843 4.1843 & - - 8.3949 8.3949 & 5.3214 5.3214 & - - 9.6978 9.6978 & 10.8687 10.8687 & - - 7.6755 7.6755 & - - 2.9484 2.9484 \end{matrix}]$

(2)隐含层阈值：(2) Hidden layer threshold:

θ＝[7.4546 11.6187 -14.4297 -8.1823 -5.4403 -6.0824 -31.6805 0.2332 9.3391]^T θ＝[7.4546 11.6187 -14.4297 -8.1823 -5.4403 -6.0824 -31.6805 0.2332 9.3391] ^T

(3)隐含层到输出层权值：(3) Hidden layer to output layer weight:

ω＝[0.6122 -0.5591 -0.6537 1.3151 0.8652 0.6033 -0.6286 -0.7779 -0.5384]ω＝[0.6122 -0.5591 -0.6537 1.3151 0.8652 0.6033 -0.6286 -0.7779 -0.5384]

(4)输出层阈值： (4) Output layer threshold:

仿真结果如图4所示，从图4中可以看出通过BP神经网络预测得到的流化床锅炉最佳工作温度与现场得到的最佳工作温度误差在±3％以内，且变化趋势一致，说明本发明通过BP神经网络方法很好的建立了流化床锅炉的模型，并能够预测出最佳的工作温度，进而能够给操作人员提供流化床锅炉最佳工作温度这一关键参数，提高燃烧效率，提高脱硫效率，减少硫化物的排放量，达到节能减排的目的。The simulation results are shown in Fig. 4. It can be seen from Fig. 4 that the error between the optimal operating temperature of the fluidized bed boiler predicted by the BP neural network and the optimal operating temperature obtained in the field is within ± 3%, and the trend of change is consistent. Illustrate that the present invention has set up the model of fluidized bed boiler very well by BP neural network method, and can predict the best working temperature, and then can provide operator this key parameter of best working temperature of fluidized bed boiler, improve Combustion efficiency, improve desulfurization efficiency, reduce sulfide emissions, and achieve the purpose of energy saving and emission reduction.

上述虽然结合附图对本发明的具体实施方式进行了描述，但并非对本发明保护范围的限制，所属领域技术人员应该明白，在本发明的技术方案的基础上，本领域技术人员不需要付出创造性劳动即可做出的各种修改或变形仍在本发明的保护范围以内。Although the specific implementation of the present invention has been described above in conjunction with the accompanying drawings, it does not limit the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present invention.

Claims

1., based on a method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: comprise the following steps:

(1) according to the practical operation situation of thermal power plant circulating fluidized bed boiler, fuel quantity x is chosen ₁, lime stone amount x ₂, a component of degree n n x ₃, secondary air flow x ₄as the input of BP neural network model, using the output of the optimum working temperature of Circulating Fluidized Bed Boiler as BP neural network;

(2) record and store on-the-spot historical data and do filtering process, choosing these data as training set sample, determine the input layer number of BP neural network, hidden layer node number, weights and threshold parameter;

(3) carry out analytical calculation by the methods combining input parameter of BP neural network, obtain the optimum working temperature of the cogeneration plant's fluidized-bed combustion boiler doped;

(4) carry out emulation testing, the result by prediction is compared with on-the-spot actual result.

2. as claimed in claim 1 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (1), concrete grammar is: according to the practical operation situation of thermal power plant circulating fluidized bed boiler, analyze relevant input and output amount, pass through emulation experiment, filter out the variable can analyzed recirculating fluidized bed optimum working temperature, as the input of BP neural network model, using the output of the optimum working temperature of Circulating Fluidized Bed Boiler as BP neural network, finally choose fuel quantity x ₁, lime stone amount x ₂, a component of degree n n x ₃, secondary air flow x ₄as input, fluidized bed optimum working temperature y ₁as output.

3., as claimed in claim 1 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (2), concrete grammar comprises:

A () is recorded and is stored on-the-spot historical data and do filtering process, choose these data as training set sample, training set sample has 4 groups, is respectively fuel quantity training set, lime stone amount training set, primary air flow training set, secondary air flow training set;

B the node number of the input layer of () definition BP neural network is n, obtain n=4 by analysis above, the node number of hidden layer is q, and the weights of input layer and hidden layer are ν _ki(k=1,2 ..., q; I=1,2 ..., n), threshold value is θ _i(i=1,2 ..., n), the node number of output layer is m, known m=1, and the weights of hidden layer and output layer are ω _jk(j=1,2 ..., m; K=1,2 ..., q), threshold value is for the transport function of hidden layer, f ₂the transport function that () is output layer.

4., as claimed in claim 3 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (a), 4 groups of sample sizes are all defined as 165.

5. as claimed in claim 1 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (3), concrete analysis calculation method is: calculate the output of hidden layer node and the output of output layer node, the expectation value of definition neural network exports, and the output error of calculating is deployed in hidden layer and input layer, the method adjustment weights be directly proportional according to the weights and negative gradient that make BP neural network, training network.

6., as claimed in claim 5 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (3), the output of hidden layer node is:

z_{k} = f_{1} (Σ_{i = 0}^{n} v_{ki} x_{i} - θ_{k}) - - - (1)

K=1 in above formula, 2 ..., q;

The output of output layer node is:

J=1 in above formula, 2 ..., m;

If the expectation value of neural network exports as d=(d ₁..., d _j, d _m), when the output of network output and expectation value is inconsistent, can there is output error, it is as follows that we define output error E:

E = \frac{1}{2} {(d - y)}^{2} = \frac{1}{2} Σ_{j = 1}^{m} {(d_{j} - y_{j})}^{2} - - - (3)

Formula (3) is deployed into hidden layer, has:

Formula (4) is deployed into input layer, has:

7., as claimed in claim 6 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (3), obtain according to above formula (5), E is ν _ki, ω _jkfunction, if change ν _ki, ω _jkvalue, so the value of error E also can change; The optimal result of BP neural network is exactly make E become as far as possible little of to meet our requirement, and the weights of BP neural network and negative gradient are directly proportional, and in the process of optimum working temperature determining fluidized-bed combustion boiler, is exactly according to such method adjustment weights, that is:

{Δv}_{ki} = - η \frac{&PartialD; E}{&PartialD; v_{ki}} - - - (6)

K=1 in above formula, 2 ..., q, i=1,2 ..., n;

{Δω}_{jk} = - η \frac{&PartialD; E}{{&PartialD; ω}_{jk}} - - - (7)

J=1 in above formula, 2 ..., m, k=1,2 ..., q;

Negative sign in formula (6) and (7) represents Gradient Descent, and η ∈ (0,1) represents scale-up factor, it reflects the speed of study.

8. as claimed in claim 6 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (3), the concrete grammar of training network is: hidden layer node number is set to 1, training network, then progressively node number is increased, with identical sample training, nodes corresponding when determining that error is minimum; When using the method, use some experimental formulas; The nodes that experimental formula draws is "ball-park" estimate value, as the initial value of method of trial and error.

9., as claimed in claim 1 based on the method for BP neural network prediction Circulating Fluidized Bed Boiler optimum working temperature, it is characterized in that: in described step (3), following two pacing itemss when determining the node number of hidden layer, must be met:

(1) number of training must connect flexible strategy more than model, is 2-10 times; Otherwise, sample needs to be divided into several part, takes the way of " trained in turn ", obtains network model of good performance with this;

(2) the node number of hidden layer is less than N-1, is N number of training; Otherwise, the error of model and sample properties have nothing to do, and go to zero, the model that is set up is without generalization ability; Identical reason, the number of input layer is also less than N-1.