Nuclear evidence prediction method for smoke concentration of first three electric fields of dry electric precipitator of coal-fired unit

Abstract

The invention discloses a nuclear evidence prediction method for smoke concentration of first three electric fields of a dry electric precipitator of a coal-fired unit, which comprises the following steps of: firstly, obtaining outlet smoke concentration of the unit under each typical load and secondary current value of each channel in each stage of electric field of the electric precipitator; according to the static characteristics of the fourth electric field and the fifth electric field, obtaining a three-electric-field outlet smoke concentration converted value; and carrying out data normalization and constructing a training data set. Then constructing a smoke concentration nuclear evidence prediction model of the first three electric fields, wherein a radial basis function is selected as a kernel function; and training model parameters by using the training data set and optimizing kernel function bandwidth by using a k-fold cross validation method. And finally, obtaining smoke concentration prediction results of the first three electric fields by utilizing the prediction model, and converting the prediction results into a dimensionalized form. The method can be used as the premise and guarantee of power supply parameter optimization, so that the energy-saving operation of the dry-type electric dust remover is realized under the condition that the smoke concentration at the outlet meets the environment-friendly requirement.

Description

Technical Field

[0001] The invention relates to the field of energy-saving and environmentally friendly operation of coal-fired units, and in particular to a method for predicting the nuclear evidence of smoke concentration in the first three electric fields of a dry electric precipitator of a coal-fired unit. Background Art

[0002] In order to prevent the large amount of smoke and dust generated by the combustion of coal powder in thermal power units from polluting the atmospheric environment, coal-fired power plants are usually equipped with dry electrostatic precipitators at the outlet of the flue at the tail of the boiler to reduce the emission of this pollutant. Among the factors affecting the efficiency of dry electrostatic precipitators, the design parameters usually do not change significantly with the change of operating conditions during operation; the flue gas parameters will change when the unit load and the type of coal used change, thereby causing disturbances to the electrostatic precipitator process; the operating parameters are the main operating quantities for adjusting the electrostatic precipitator process. During operation, the outlet smoke concentration is usually controlled by adjusting the power supply parameters of the dry electrostatic precipitator. When the flue gas parameters are relatively stable, the dust removal efficiency is closely related to the power supply parameters. Improving the dust removal effect of the electrostatic precipitator means increasing the power consumption of the electrostatic precipitator, thereby increasing the power consumption of the plant. Since the dry electrostatic precipitator only has a smoke concentration measurement point at the outlet, only the dynamic characteristics of the fifth level (or the last few levels as a whole) electric field and the static characteristics of the first few levels of electric field can be obtained. The fifth electric field of the dry electrostatic precipitator has limited ability to regulate the smoke concentration. Therefore, the fourth and fifth electric fields are regarded as a whole, and a static characteristic model of the first three electric fields can be established. However, how to establish an accurate model of the first three electric fields becomes a difficulty. Summary of the invention

[0003] In view of the problems existing in the prior art, the present invention provides a nuclear evidence prediction method for the smoke concentration in the first three electric fields of a dry electrostatic precipitator of a coal-fired unit, aiming to establish a prediction model for the power supply parameters of the dry electrostatic precipitator and the smoke concentration at the outlet of the first three electric fields in a data-driven manner.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for predicting the nuclear evidence of smoke concentration in the first three electric fields of a dry electrostatic precipitator of a coal-fired unit, comprising the following steps:

[0005] S1. Obtain the outlet dust concentration of the unit under each typical load and the secondary current value of each channel in each level of the electric field of the electrostatic precipitator from the database of the distributed control system DCS of the coal-fired thermal power unit; for each level of the electric field, use the average value of the secondary current value of each channel in the electric field as the secondary current value of the electric field, and obtain the converted value of the outlet dust concentration of the third electric field according to the static characteristics of the fourth and fifth electric fields;

[0006] S2. Construct a training data set, including the secondary current values ​​of the first three electric fields and the corresponding converted values ​​of the smoke concentration at the outlets of the three electric fields. The converted values ​​are all normalized to the range of [-1,1]. When constructing the training data set, the data with sudden current changes in the fourth and fifth electric fields are removed;

[0007] S3. Based on the nuclear evidence regression algorithm, a nuclear evidence prediction model for smoke concentration in the first three electric fields is constructed, and the radial basis function is selected as the kernel function; the secondary current values ​​of the first three electric fields are used as the model input, and the converted values ​​of the smoke concentration at the outlets of the three electric fields are used as the output, and the parameters of the nuclear evidence prediction model for smoke concentration in the first three electric fields are trained and the kernel function bandwidth is optimized;

[0008] S4. The collected secondary current values ​​of the three electric fields are normalized and input into the nuclear evidence prediction model of the smoke concentration of the first three electric fields. After the predicted values ​​of the smoke concentration of the first three electric fields are obtained, they are converted into dimensional predicted values.

[0009] Furthermore, in the aforementioned step S3, the kernel evidence regression algorithm includes the following sub-steps:

[0010] S31, expressing the output data in the form of a fuzzy quality function;

[0011] S32, initializing model parameters, namely, the dual form and variance of the nonlinear regression coefficient, and setting the number of parameter iterations q to 0; constructing the nonlinear function expressing the input-output relationship of the model into a kernel function form, selecting a radial basis function as the kernel function and setting the kernel function bandwidth;

[0012] S33, optimizing model parameters using fuzzy evidence EM algorithm: calculating the log-likelihood of the observed data under the qth round parameters, and then calculating the expectation of the fuzzy evidence EM algorithm;

[0013] S34, maximize the conditional expectation, obtain the parameters of the q+1 round, use the parameters of the new round to calculate the log-likelihood of the observed data, and then calculate the relative error of the log-likelihood of the observed data of adjacent iterations,

[0014] S35, determine whether the relative error is greater than the critical value ε, if yes, return to execute S33; otherwise, execute S36;

[0015] S36, outputting the parameters at this time as model parameters;

[0016] S37. After inputting the normalized secondary current into the model, the dimensionless three-electric field smoke concentration prediction result can be obtained, and the result can be converted into a dimensional form.

[0017] Furthermore, the aforementioned step S31 is specifically as follows:

[0018] Based on each component x of the output vector data x i are all the same deterministic function f(u i ) and random noise e, the random noise e follows the normal distribution N(0,σ), and the components x of the output data i All obey the normal distribution N(v i ,σ),vi =f(u i ), the output data vector x obeys a multidimensional Gaussian distribution:

[0019] x i =f(u i )+e,

[0020] Using an estimator Estimate the unknown function f(u), which defines a set of parameters ψ=ψ1,ψ1,…,ψ d , that is, σ and ω, where ω is the dual form of the regression coefficient b;

[0021] The output data is expressed as a fuzzy quality function:

[0022]

[0023] Among them, Ω X is the dimensionless output Range p i is the probability that the sensor is in good working condition, p i =1,i=1,2,...,n, triangular fuzzy number Indicates the output when the output sensor is in good working condition, δ i Dimensionless output for all sampling points in the continuous data segment to which the sampling point belongs The sample standard deviation of .

[0024] Furthermore, the aforementioned step S32 is specifically as follows:

[0025] Input value pair {(u i ,m i )|i=1,2,...,n}, preset the critical value ε as the iteration termination condition, initialize the variance σ and ω, that is, the initial parameter ω (0) and σ (0) ), set the parameter iteration number q to 0;

[0026] Constructing nonlinear functions of input-output relationships Through the kernel function method, the radial basis function is selected as the kernel function, and φ(·) is its related feature map, then As follows:

[0027]

[0028] Among them, φ(U) is the training sample in the feature space, and the input matrix U=(u1,...,u n ),u i =(u i,1 ,...,u i,p)'; regression coefficient b=(b0,...,b p )'; The radial basis function is as follows:

[0029] Where s is the bandwidth.

[0030] Furthermore, the aforementioned step S33 is specifically: calculating the log-likelihood of the observed data under the qth round parameters, as follows:

[0031]

[0032] Among them, g(x;ψ) is the probability density function, as shown below:

[0033]

[0034] It is the contour function of the smoke concentration conversion result x of the three electric fields expressed by fuzzy evidence theory:

[0035]

[0036] In the expression of is the fuzzy focal element defined on the recognition frame Ω, for The fuzzy membership function of , F(m) is a fuzzy set;

[0037] Calculate the expectation of the fuzzy evidence EM algorithm:

[0038] in,

[0039] Output variable x i The observed value is the interval number [a i ,b i ]hour, and They are expressed as follows:

[0040]

[0041] Among them, x * =(x-ν) / σ (q) , Φ(·) is the standard normal distribution function.

[0042] Furthermore, the aforementioned step S34 is specifically as follows: Maximum, get the parameter ω of round q+1 (q+1) and σ (q+1) , the formula is:

[0043]

[0044] in The kernel matrix K is:

[0045]

[0046] Use the parameters ω obtained in the new round (q+1) and σ (q+1) Calculate the log-likelihood of the new observations logL(ψ (q+1) ; pl), and then calculate the relative error D of the log-likelihood of the observed data of adjacent iterations, the formula is as follows:

[0047]

[0048] Furthermore, in the aforementioned step S3, optimizing the optimal bandwidth of the kernel function using an optimization algorithm includes the following sub-steps:

[0049] S301, randomly divide the training set into k mutually exclusive subsets of equal size, and initialize the bandwidth s to 1;

[0050] S302, initialize j to 1, set the jth subset as the current test set, and the rest as training sets, and calculate the output prediction value of the current test set by the kernel evidence regression method;

[0051] S303, calculate the loss function CV of this time by the following formula i (s):

[0052]

[0053] in, is the predicted response of the other k-1 training samples to the test sample except the currently selected test sample, is m i The pignistic expectation is expressed as follows:

[0054]

[0055] S304, let j increase by 1, calculate the loss function when the second subset is used as the test set, further calculate the loss function when each subset is used as the test set and sum them to obtain the total loss function, that is, the loss function obtained when s=1:

[0056]

[0057] S305, s takes a value from a preset range, calculates the total loss function under each s, and obtains the optimal bandwidth

[0058] Compared with the prior art, the beneficial technical effects of the above technical solutions adopted by the present invention are as follows: the method provided by the present invention can more accurately predict the predicted value of the three-field outlet smoke concentration of the dry electrostatic precipitator under each load section through a data-driven strategy, thereby avoiding dependence on the complex mechanism process of the electrostatic precipitator. Furthermore, it can provide a calculation basis for the dry electrostatic precipitator to achieve optimal energy consumption while meeting environmental protection requirements. The model of the present invention can serve as a prerequisite for power supply parameter optimization, so that energy-saving operation of the dry electrostatic precipitator can be achieved when the outlet smoke concentration meets environmental protection requirements. Compared with mechanism modeling, data-driven modeling has better adaptability to the variable and complex actual dust removal process, and is easy to apply in intelligent control and operation optimization. The kernel evidence regression algorithm uses fuzzy belief functions to process imprecise, incomplete and uncertain data, combines the advantages of evidence theory and EM algorithm, and is suitable for linear and nonlinear regression. It has good robustness, flexibility and predictive performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is the structural diagram of the dry electrostatic precipitator.

[0060] Figure 2 Flowchart of the kernel evidence regression algorithm.

[0061] Figure 3 Flowchart for kernel function optimal bandwidth optimization.

[0062] Figure 4 Prediction curve of smoke concentration at the outlet of 650MW dry electrostatic precipitator.

[0063] Figure 5 Scatter plot of predicted dust concentration at the outlet of 650MW dry electrostatic precipitator.

[0064] Figure 6 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0065] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0066] Various aspects of the invention are described herein with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the invention are not limited to those described in the accompanying drawings. It should be understood that the invention is implemented by any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed in the invention are not limited to any implementation. In addition, some aspects disclosed in the invention may be used alone or in any appropriate combination with other aspects disclosed in the invention.

[0067] The present invention provides a method for predicting the smoke concentration in the three electric fields before the dry electrostatic precipitator of a coal-fired unit. The method uses a nuclear evidence regression method in a data-driven manner to establish a prediction model for the effect of the power supply parameters of the electrostatic precipitator on the smoke concentration in the three electric fields. This method can provide a calculation basis for the dry electrostatic precipitator to achieve optimal energy consumption while meeting environmental protection requirements. Generally, a dry electrostatic precipitator has multiple electric fields. This embodiment selects a six-channel, five-electric field structure for illustration. The structure of the electrostatic precipitator in this example is as follows: Figure 1 As shown, taking a 1000MW coal-fired unit as an example, typical operating conditions are generally 850MW, 750MW, and 650MW, and the flue gas flows through the #1, #2, #3, #4, and #5 electric fields in sequence.

[0068] refer to Figure 6 , the method of the present invention specifically comprises the following steps:

[0069] S1. Obtain the outlet dust concentration of the unit under each typical load and the secondary current value of each channel in each level of the electric field of the electrostatic precipitator from the database of the distributed control system DCS of the coal-fired thermal power unit; for each level of the electric field, use the average value of the secondary current value of each channel in the electric field as the secondary current value of the electric field, and obtain the converted value of the outlet dust concentration of the third electric field according to the static characteristics of the fourth and fifth electric fields;

[0070] S2. Construct a training data set, including the secondary current values ​​of the first three electric fields and the corresponding converted values ​​of the smoke concentration at the outlets of the three electric fields. The converted values ​​are all normalized to the range of [-1,1]. When constructing the training data set, the data with sudden current changes in the fourth and fifth electric fields are removed;

[0071] S3. Based on the nuclear evidence regression algorithm, a nuclear evidence prediction model for smoke concentration in the first three electric fields is constructed, and the radial basis function is selected as the kernel function; the secondary current values ​​of the first three electric fields are used as the model input, and the converted values ​​of the smoke concentration at the outlets of the three electric fields are used as the output, and the parameters of the nuclear evidence prediction model for smoke concentration in the first three electric fields are trained and the kernel function bandwidth is optimized;

[0072] S4. The collected secondary current values ​​of the three electric fields are normalized and input into the nuclear evidence prediction model of the smoke concentration of the first three electric fields. After obtaining the predicted values ​​of the smoke concentration of the first three electric fields (dimensionless), they are converted into dimensional predicted values.

[0073] In step S2, since the sudden current of the four-five electric field will cause the converted smoke concentration value of the three electric fields to have pulses, in order to avoid errors, the data with sudden current in the four-five electric field is not selected as training data. Several representative continuous data segments are selected to form the training data set of the load, and the training data set is dimensionlessly preprocessed so that the value range of the dimensionless input and output of the training set is [-1,1].

[0074] like Figure 2As shown, as a preferred embodiment of the method for predicting the smoke concentration of the three electric fields before the dry electrostatic precipitator of a coal-fired unit of the present invention, in step S3, the nuclear evidence regression algorithm includes the following sub-steps:

[0075] S31, expressing the output data in the form of a fuzzy quality function;

[0076] S32, initializing model parameters (dual form and variance of nonlinear regression coefficients), setting the number of parameter iterations q to 0; constructing the nonlinear function expressing the input-output relationship of the model into a kernel function form, selecting a radial basis function as the kernel function and setting the kernel function bandwidth;

[0077] S33, optimizing model parameters using fuzzy evidence EM algorithm: calculating the log-likelihood of the observed data under the qth round parameters, and then calculating the expectation of the fuzzy evidence EM algorithm;

[0078] S34, maximize the conditional expectation, obtain the parameters of the q+1 round, use the parameters of the new round to calculate the log-likelihood of the observed data, and then calculate the relative error of the log-likelihood of the observed data of adjacent iterations,

[0079] S35, determine whether the relative error is greater than the critical value ε, if yes, return to execute S33; otherwise, execute S36;

[0080] S36, outputting the parameters at this time as model parameters;

[0081] S37. After inputting the normalized secondary current into the model, the dimensionless three-electric field smoke concentration prediction result can be obtained, and the result can be converted into a dimensional form.

[0082] As a preferred method for predicting the smoke concentration nuclear evidence of the first three electric fields of a dry electrostatic precipitator of a coal-fired unit of the present invention, step S31 is specifically as follows:

[0083] First, based on each component x of the output vector data x i are all the same deterministic function f(u i ) and random noise e, the random noise e follows the normal distribution N(0,σ), and the components x of the output data i All obey the normal distribution N(v i ,σ),v i =f(u i ), the output data vector x obeys a multidimensional Gaussian distribution:

[0084] x i =f(u i )+e,

[0085] Using an estimator Estimate the unknown function f(u), which defines a set of parameters ψ=ψ1,ψ1,…,ψd , that is, σ and ω, where ω is the dual form of the regression coefficient b;

[0086] The output data is expressed as a fuzzy quality function:

[0087]

[0088] Among them, Ω X is the dimensionless output Range p i is the probability that the sensor is in good working condition, p i =1,i=1,2,...,n, triangular fuzzy number Indicates the output when the output sensor is in good working condition, δ i Dimensionless output for all sampling points in the continuous data segment to which the sampling point belongs The sample standard deviation of .

[0089] As a preferred method for predicting the smoke concentration nuclear evidence of the first three electric fields of a dry electrostatic precipitator of a coal-fired unit of the present invention, step S32 is specifically as follows:

[0090] Input value pair {(u i ,m i )|i=1,2,...,n}, preset the critical value ε as the iteration termination condition, initialize the variance σ and ω, that is, the initial parameter ω (0) and σ (0) ), set the parameter iteration number q to 0;

[0091] Constructing nonlinear functions of input-output relationships Through the kernel function method, the radial basis function is selected as the kernel function, and φ(·) is its related feature map, then As follows:

[0092]

[0093] Among them, φ(U) is the training sample in the feature space, and the input matrix U=(u1,...,u n ),u i =(u i,1 ,...,u i,p )'; regression coefficient b=(b0,...,b p )'; The radial basis function is as follows:

[0094] Where s is the bandwidth.

[0095] As a preferred embodiment of the method for predicting the smoke concentration nuclear evidence of the first three electric fields of a dry electrostatic precipitator of a coal-fired unit of the present invention, step S33 is specifically: calculating the logarithmic likelihood of the observed data under the qth round of parameters, as follows:

[0096]

[0097] Among them, g(x;ψ) is the probability density function, as shown below:

[0098]

[0099] It is the contour function of the smoke concentration conversion result x of the three electric fields expressed by fuzzy evidence theory:

[0100]

[0101] In the expression of is the fuzzy focal element defined on the recognition frame Ω, for The fuzzy membership function of , F(m) is a fuzzy set;

[0102] Calculate the expectation of the fuzzy evidence EM algorithm:

[0103] in,

[0104] Output variable x i The observed value is the interval number [a i ,b i ]hour, and They are expressed as the following formulas:

[0105]

[0106]

[0107] Among them, x * =(x-ν) / σ (q) , Φ(·) is the standard normal distribution function.

[0108] As a preferred method for predicting the dust concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit of the present invention, step S34 is specifically: making the conditional expectation Maximum, get the parameter ω of the q+1 round (q+1) and σ (q+1) , the formula is:

[0109]

[0110] in The kernel matrix K is:

[0111]

[0112] Use the parameters ω obtained in the new round (q+1) and σ (q+1) Calculate the log-likelihood of the new observations logL(ψ (q+1) ; pl), and then calculate the relative error D of the log-likelihood of the observed data of adjacent iterations, the formula is as follows:

[0113]

[0114] refer to Figure 3 As a preferred embodiment of the method for predicting the smoke concentration kernel evidence of the first three electric fields of a dry electrostatic precipitator of a coal-fired unit of the present invention, in step S3, optimizing the optimal bandwidth of the kernel function by using an optimization algorithm includes the following sub-steps:

[0115] S301, randomly divide the training set into k mutually exclusive subsets of equal size, and initialize the bandwidth s to 1;

[0116] S302, initialize j to 1, set the jth subset as the current test set, and the rest as training sets, and calculate the output prediction value of the current test set by the kernel evidence regression method;

[0117] S303, calculate the loss function CV of this time by the following formula i (s):

[0118]

[0119] in, is the predicted response of the other k-1 training samples to the test sample except the currently selected test sample, is m i The pignistic expectation is expressed as follows:

[0120]

[0121] S304, let j increase by 1, calculate the loss function when the second subset is used as the test set, further calculate the loss function when each subset is used as the test set and sum them to obtain the total loss function, that is, the loss function obtained when s=1:

[0122]

[0123] S305, s takes a value from a preset range, calculates the total loss function under each s, and obtains the optimal bandwidth

[0124] Taking the 650MW working condition as an example, the optimal bandwidth s = 10, after executing the nuclear evidence regression algorithm and converting the outlet smoke concentration, we can get Figure 4 The prediction curve of smoke concentration at the outlet of 650MW dry electrostatic precipitator shown in the figure has a mean square error (MSE) of 1.1554 between the predicted value and the actual value, and the prediction effect is good. Figure 5 This is a scatter plot of the predicted dust concentration at the outlet of a 650MW dry electrostatic precipitator. It can be seen that the number of scattered points near the diagonal accounts for a large proportion.

[0125] Although the present invention has been described above with preferred embodiments, it is not intended to limit the present invention. A person skilled in the art of the present invention may make various modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the definition of the claims.

Claims

1. A method for predicting the dust concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit, characterized in that: The following steps are involved: S1. Obtain the outlet dust concentration of the unit under each typical load and the secondary current value of each channel in each level of the electric field of the electrostatic precipitator from the database of the distributed control system DCS of the coal-fired thermal power unit; for each level of the electric field, use the average value of the secondary current value of each channel in the electric field as the secondary current value of the electric field, and obtain the converted value of the outlet dust concentration of the third electric field according to the static characteristics of the fourth and fifth electric fields; S2. Construct a training data set, including the secondary current values ​​of the first three electric fields and the corresponding converted values ​​of the smoke concentration at the outlets of the three electric fields. The converted values ​​are all normalized to the range of [-1,1]. When constructing the training data set, the data with sudden current changes in the fourth and fifth electric fields are removed; S3. Based on the nuclear evidence regression algorithm, a nuclear evidence prediction model for smoke concentration in the first three electric fields is constructed, and the radial basis function is selected as the kernel function; the secondary current values ​​of the first three electric fields are used as the model input, and the converted values ​​of the smoke concentration at the outlets of the three electric fields are used as the output, and the parameters of the nuclear evidence prediction model for smoke concentration in the first three electric fields are trained and the kernel function bandwidth is optimized; S4. The collected secondary current values ​​of the three electric fields are normalized and input into the nuclear evidence prediction model of the smoke concentration of the first three electric fields. After the predicted values ​​of the smoke concentration of the first three electric fields are obtained, they are converted into dimensional predicted values.

2. According to claim 1, a method for predicting the dust concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit is characterized in that: In step S3, the kernel evidence regression algorithm includes the following sub-steps: S31, expressing the output data in the form of a fuzzy quality function; S32, initializing model parameters, namely, the dual form and variance of the nonlinear regression coefficient, and setting the number of parameter iterations q to 0; constructing the nonlinear function expressing the input-output relationship of the model into a kernel function form, selecting a radial basis function as the kernel function and setting the kernel function bandwidth; S33, optimizing model parameters using fuzzy evidence EM algorithm: calculating the log-likelihood of the observed data under the qth round parameters, and then calculating the expectation of the fuzzy evidence EM algorithm; S34, maximize the conditional expectation, obtain the parameters of the q+1 round, use the parameters of the new round to calculate the log-likelihood of the observed data, and then calculate the relative error of the log-likelihood of the observed data of adjacent iterations, S35, determine whether the relative error is greater than the critical value ε, if yes, return to execute S33; otherwise, execute S36; S36, outputting the parameters at this time as model parameters; S37. After inputting the normalized secondary current into the model, the dimensionless three-electric field smoke concentration prediction result can be obtained, and the result can be converted into a dimensional form.

3. According to claim 2, a method for predicting the smoke concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit is characterized in that: Step S31 is specifically as follows: Based on each component x of the output vector data x i are all the same deterministic function f(u i ) and random noise e, the random noise e follows the normal distribution N(0,σ), and the components x of the output data i All obey the normal distribution N(v i ,σ),v i =f(u i ), the output data vector x obeys a multidimensional Gaussian distribution: x i =f(u i )+e, Using an estimator Estimate the unknown function f(u), which defines a set of parameters ψ=ψ1,ψ1,…,ψ d , that is, σ and ω, where ω is the dual form of the regression coefficient b; The output data is expressed as a fuzzy quality function: Among them, Ω X is the dimensionless output Range p i is the probability that the sensor is in good working condition, p i =1,i=1,2,...,n, triangular fuzzy number Indicates the output when the output sensor is in good working condition, δ i Dimensionless output for all sampling points in the continuous data segment to which the sampling point belongs The sample standard deviation of .

4. According to claim 2, a method for predicting the dust concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit is characterized in that: Step S32 is specifically as follows: Input value pair {(u i ,m i )|i=1,2,...,n}, preset the critical value ε as the iteration termination condition, initialize the variance σ and ω, that is, the initial parameter ω (0) and σ (0) ), set the parameter iteration number q to 0; Constructing nonlinear functions of input-output relationships Through the kernel function method, the radial basis function is selected as the kernel function, and φ(·) is its related feature map, then As follows: Among them, φ(U) is the training sample in the feature space, and the input matrix U=(u1,...,u n ),u i =(u i,1 ,...,u i,p )'; regression coefficient b=(b0,...,b p )'; The radial basis function is as follows: Where s is the bandwidth.

5. According to claim 2, a method for predicting the smoke concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit is characterized in that: Step S33 is specifically: calculating the log-likelihood of the observed data under the qth round of parameters, as follows: Among them, g(x;ψ) is the probability density function, as shown below: It is the contour function of the smoke concentration conversion result x of the three electric fields expressed by fuzzy evidence theory: In the expression of is the fuzzy focal element defined on the recognition frame Ω, for The fuzzy membership function of , F(m) is a fuzzy set; Calculate the expectation of the fuzzy evidence EM algorithm: in, Output variable x i The observed value is the interval number [a i ,b i ]hour, and They are expressed as follows: Among them, x * =(x-ν) / σ (q) , Φ(·) is the standard normal distribution function.

6. A method for predicting the dust concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit according to claim 2, characterized in that: Step S34 is specifically: making the conditional expectation Maximum, get the parameter ω of round q+1 (q+1) and σ (q+1) , the formula is: in The kernel matrix K is: Use the parameters ω obtained in the new round (q+1) and σ (q+1) Calculate the log-likelihood of the new observations logL(ψ (q+1) ; pl), and then calculate the relative error D of the log-likelihood of the observed data of adjacent iterations, the formula is as follows:

7. A method for predicting the dust concentration nuclear evidence of the three electric fields before the dry electrostatic precipitator of a coal-fired unit according to claim 6, characterized in that: In step S3, optimizing the optimal bandwidth of the kernel function using an optimization algorithm includes the following sub-steps: S301, randomly divide the training set into k mutually exclusive subsets of equal size, and initialize the bandwidth s to 1; S302, initialize j to 1, set the jth subset as the current test set, and the rest as training sets, and calculate the output prediction value of the current test set by the kernel evidence regression method; S303, calculate the loss function CV of this time by the following formula i (s): in, is the predicted response of the other k-1 training samples to the test sample except the currently selected test sample, is m i The pignistic expectation is expressed as follows: S304, let j increase by 1, calculate the loss function when the second subset is used as the test set, further calculate the loss function when each subset is used as the test set and sum them to obtain the total loss function, that is, the loss function obtained when s=1: S305, s takes a value from a preset range, calculates the total loss function under each s, and obtains the optimal bandwidth

Images