CN118039022A

CN118039022A - Sulfur dioxide concentration prediction method and device and computer equipment

Info

Publication number: CN118039022A
Application number: CN202410433019.9A
Authority: CN
Inventors: 郭锦涛
Original assignee: Guoneng Longyuan Environmental Protection Co Ltd
Current assignee: Guoneng Longyuan Environmental Protection Co Ltd
Priority date: 2024-04-11
Filing date: 2024-04-11
Publication date: 2024-05-14
Anticipated expiration: 2044-04-11
Also published as: CN118039022B

Abstract

The invention provides a sulfur dioxide concentration prediction method and device and computer equipment, and relates to the technical field of energy conservation and emission reduction. The method comprises the following steps: determining characteristic variables influencing the sulfur dioxide concentration of a desulfurization outlet; and inputting the characteristic variable into a constructed prediction model to obtain a sulfur dioxide concentration prediction result, wherein the prediction model is obtained by performing super-parameter optimization and training on an initially constructed neural network, and an improved SSA algorithm is adopted in the super-parameter optimization process. In the improved SSA algorithm, a butterfly algorithm is adopted for the position update of a finder, after the positions of the finder, a joiner and a alerter are updated in each iteration, an elimination recombination mechanism is introduced, the positions of eliminated individuals are reset and then form a new population with non-eliminated individuals, the positions of the eliminated individuals are between the positions of the eliminated individuals and the global optimal individual positions in the last iteration after the reset, and the balance and the precision of the super-parameter optimization process are improved, so that the prediction precision of a prediction model is improved.

Description

Sulfur dioxide concentration prediction method and device and computer equipment

Technical Field

The invention belongs to the technical field of energy conservation and emission reduction, and particularly relates to a sulfur dioxide concentration prediction method, a sulfur dioxide concentration prediction device, a machine-readable storage medium and computer equipment.

Background

In thermal power generation, sulfur dioxide generated by coal combustion causes direct pollution to air, and the improvement of desulfurization efficiency is beneficial to energy conservation, emission reduction and consumption reduction, so that the combination of prevention and control is realized. At present, a limestone-gypsum wet method is mostly adopted in a desulfurization system, and the system has the characteristics of large delay, large inertia, nonlinearity and the like, so that the control difficulty of the system is high. The wet desulfurization technology adopts a forced oxidation process, and an oxidation fan is used for blowing air into the slurry of the absorption tower to carry out forced oxidation. The current of the oxidation blower influences the full extent of the slurry oxidation reaction, thereby influencing the desulfurization efficiency and the system power consumption rate. Therefore, the improvement of the prediction precision of the sulfur dioxide concentration of the desulfurization outlet has important significance for timely adjusting the operation parameters of the oxidation fan, improving the desulfurization efficiency and optimizing and controlling the desulfurization system.

At present, sulfur dioxide concentration prediction at a desulfurization outlet is performed based on a pre-constructed neural network prediction model, however, the working condition of a desulfurization system is complex, influence factors of sulfur dioxide concentration at the desulfurization outlet are numerous, and the input variables of the neural network prediction model have the characteristics of complex coupling relation, strong nonlinearity and the like, so that accurate mechanism modeling is difficult. In order to improve the desulfurization efficiency and optimize the desulfurization control process, it is necessary to improve the prediction accuracy of the existing neural network prediction model.

Disclosure of Invention

The embodiment of the invention aims to provide a sulfur dioxide concentration prediction method, a sulfur dioxide concentration prediction device, a machine-readable storage medium and computer equipment, which are used for overcoming the defect of lower prediction precision of a neural network prediction model for predicting the sulfur dioxide concentration of a desulfurization outlet in the prior art.

To achieve the above object, a first aspect of an embodiment of the present invention provides a method for predicting a concentration of sulfur dioxide, the method including:

determining characteristic variables influencing the sulfur dioxide concentration of a desulfurization outlet;

Inputting the characteristic variable into a constructed prediction model to obtain a sulfur dioxide concentration prediction result, wherein the prediction model is obtained by performing super-parameter optimization and training on an initially constructed neural network;

wherein, the super parameter optimization process is as follows:

taking the randomly generated super parameters as initial population individuals, and dividing the individuals into discoverers and joiners;

updating the positions of a finder, a jointer and a alerter by adopting an iteration method, calculating the fitness value of an individual after the position update, eliminating the individual in the population according to the fitness value, resetting the positions of the eliminated individual, and forming a new population with the individual which is not eliminated until the stopping condition is met, and generating the optimal super parameter;

The position of the eliminated individual is between the position of the eliminated individual and the position of the last iterative global optimal individual, wherein the position of the individual represents the value of the super parameter, the alerter is generated from the discoverer and the enrollee through the foraging process, a butterfly algorithm is adopted when the position of the discoverer is updated, and the position of the eliminated individual is reset.

In one embodiment of the present invention, the eliminating the individuals in the population according to the fitness value, and forming a new population with the individuals not eliminated after resetting the positions of the individuals eliminated, includes:

Sorting the fitness values of the individuals after the position updating;

The fitness value is from low to high, and individuals with preset proportions are eliminated in sequence;

resetting the positions of the eliminated individuals according to a preset rule, and forming a new population by the eliminated individuals after the position resetting and the non-eliminated individuals;

Wherein, the preset rule is: first, the Individual at/>Reset position of dimension by at/>Individual's/>The first random value and the first/>, based on the preset lower limit value of the super parameter represented by the dimension, are addedIndividual's/>The ratio of the difference between the preset upper limit value and the lower limit value of the super parameter represented by the dimension is a random number between 0 and 1, and the first/>Individual at/>The reset position of the dimension is between the/>, when eliminatedIndividual at/>The position of the dimension and the global optimum individual position of the last iteration.

In a specific embodiment of the present invention, after the randomly generated hyper-parameters are used as the individuals of the initial population, the chaotic mapping is adopted to map the initial population, so as to obtain the mapped initial population.

In a specific embodiment of the invention, before training an initially constructed neural network, calculating mutual information between historical characteristic variables under different time delays in a training sample and sulfur dioxide concentrations corresponding to the historical characteristic vectors by adopting a mutual information method, taking a time difference between the historical characteristic variable with the maximum mutual information and the sulfur dioxide concentration corresponding to the historical characteristic variable as delay time, and performing delay compensation on all the historical characteristic variables in the training sample by utilizing the delay time.

In a specific embodiment of the present invention, the neural network is a BP neural network, and the value of the hyper-parameter is the value of all weights and biases in the BP neural network.

In one embodiment of the present invention, the determining of the characteristic variable affecting the sulfur dioxide concentration at the desulfurization outlet comprises:

acquiring field production data of a desulfurization system;

and adopting a nuclear principal component analysis method to reduce the dimension of the multidimensional initial variable influencing the sulfur dioxide concentration of the desulfurization outlet in the on-site production data to obtain a characteristic variable.

In one embodiment of the invention, the initial variables include desulfurization inlet oxygen concentration, desulfurization inlet sulfur dioxide concentration, desulfurization inlet flue gas flow, desulfurization outlet oxygen concentration, desulfurization outlet flue gas flow, unit load, slurry concentration, slurry flow, adsorption column level, oxidation blower current, and circulating pump current.

A second aspect of the present invention provides a sulfur dioxide concentration prediction apparatus, the apparatus comprising:

The characteristic variable determining module is used for determining characteristic variables affecting the sulfur dioxide concentration of the desulfurization outlet;

The sulfur dioxide concentration prediction module is used for inputting the characteristic variable into the constructed prediction model to obtain a sulfur dioxide concentration prediction result, and the prediction model is obtained by performing super-parameter optimization and training on the initially constructed neural network;

wherein, the super parameter optimization process is as follows:

A third aspect of the embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the method for predicting sulfur dioxide concentration according to the first aspect of the embodiment of the present invention when executing the program.

A fourth aspect of the embodiments of the present invention provides a machine readable storage medium having stored thereon a computer program which when executed by a processor implements the method for predicting sulphur dioxide concentration according to the first aspect of the embodiments of the present invention.

In the above technical solution, the location update of the finder adopts a butterfly algorithm, and after each iteration updates the locations of the finder, the enrollee and the alerter, an elimination recombination mechanism is introduced, and recombination means that after the locations of the eliminated individuals are reset, the eliminated individuals and the individuals not eliminated form a new population, the new population is a recombined population, and the locations of the eliminated individuals are between the locations of the eliminated individuals and the locations of the globally optimal individuals in the previous iteration. In the prior art, when the position of the finder is updated, when the early warning value is smaller than the safety threshold, the position of each dimension of the finder is reduced and converged to 0, so that the super-parameter optimization process has the tendency of converging to 0 and approaching to the current global optimal solution at the initial stage of iteration, thereby leading to premature convergence and sinking into local optimal, updating the position of the finder by using the global searching stage position updating strategy of the butterfly algorithm, overcoming the problem of local optimal, leading certain dimensions sinking into the local optimal in the population to have the opportunity to jump out of iteration, and improving the global searching capability. Meanwhile, by introducing a elimination recombination mechanism, the method overcomes the problem of local optimum, improves the global searching capability, ensures the convergence capability, realizes the balance of the super-parameter optimization process, ensures the convergence speed of the super-parameter optimization process, improves the construction efficiency of a prediction model, improves the accuracy of the generated optimum super-parameter, and further improves the prediction precision of the sulfur dioxide concentration.

Additional features and advantages of embodiments of the invention will be set forth in the detailed description which follows.

Drawings

The accompanying drawings are included to provide a further understanding of embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain, without limitation, the embodiments of the invention. In the drawings:

FIG. 1 schematically shows a flow chart of a sulfur dioxide concentration prediction method according to an embodiment of the invention;

FIG. 2 schematically illustrates a schematic diagram of a construction process of a predictive model according to an embodiment of the invention;

Fig. 3 schematically shows a schematic structural diagram of a BP neural network according to an embodiment of the present invention;

FIG. 4 schematically shows the effect comparison graph of sulfur dioxide concentration prediction based on the KPCA-MI-MSSA-BP method and sulfur dioxide concentration prediction based on a conventional neural network model;

FIG. 5 schematically shows an error comparison graph of sulfur dioxide concentration prediction based on the KPCA-MI-MSSA-BP method and sulfur dioxide concentration prediction based on a conventional neural network model;

FIG. 6 schematically shows a graph of sulfur dioxide concentration prediction effect versus a BP method, a KPCA-MI-SSA-BP method, and a KPCA-MI-MSSA-BP method;

FIG. 7 schematically shows a comparison of sulfur dioxide concentration prediction errors based on BP method, KPCA-MI-SSA-BP method, and KPCA-MI-MSSA-BP method;

Fig. 8 schematically shows a block diagram of a sulfur dioxide concentration prediction apparatus according to an embodiment of the present invention;

Fig. 9 schematically shows a block diagram of a computer device according to an embodiment of the invention.

Detailed Description

The following describes the detailed implementation of the embodiments of the present invention with reference to the drawings. It should be understood that the detailed description and specific examples, while indicating and illustrating the invention, are not intended to limit the invention.

Method embodiment

Referring to fig. 1, an embodiment of the present invention provides a method for predicting sulfur dioxide concentration, which is used for predicting sulfur dioxide concentration at a desulfurization outlet, and includes the following implementation steps:

And step S100, determining characteristic variables influencing the concentration of sulfur dioxide at a desulfurization outlet.

And step S200, inputting the characteristic variable into a constructed prediction model to obtain a sulfur dioxide concentration prediction result, wherein the prediction model is obtained by performing super-parameter optimization and training on an initially constructed neural network.

The super parameter optimization process specifically comprises the following steps:

taking the randomly generated super parameters as initial population individuals, and dividing the individuals in the population into discoverers and joiners;

the alerter randomly generates from the discoverer and the joiner through the foraging process, a butterfly algorithm is adopted when the position of the discoverer is updated, the position of the individual represents the value of the super parameter, and the position of the eliminated individual after reset is between the position of the individual when eliminated and the position of the global optimal individual in the last iteration.

It should be understood that the above-described super-parametric optimization process is an improvement based on SSA algorithm (sparrow search algorithm, english: sparrow Search Algorithm). The SSA algorithm is an intelligent search algorithm based on predation and anti-predation behaviors of sparrow populations. By observing the predation process of the sparrow population, the following rules are summarized on the basis: two classes of workers exist in the population, namely discoverers and joiners, respectively, namely, individuals in the population are divided into discoverers and joiners. The discoverers are responsible for searching for food, providing foraging directions for the population, and are typically served by higher energy individuals. The joiner, also called a follower, obtains food by following the finder. In addition, in order to ensure the safety of the population during foraging, individuals can warn whether the surrounding environment is safe or not and perform early warning during foraging, and once the early warning value exceeds a safety threshold value, the whole population leaves and flies to a safety area. In a sparrow population, the roles of discoverers and joiners are flexibly changed according to the dynamic change of the respective energy levels, but the proportion of the discoverers and joiners in the whole population is fixed. In SSA algorithm, the basis for calculating the individual energy is fitness value, which is calculated according to fitness function. In the above-mentioned super-parameter optimization process of the neural network, the fitness function may be set as a mean square error of the prediction model, etc., the variable to be optimized is a super-parameter of the neural network, and one dimension of the individual position represents one super-parameter. For example, the fitness function may be expressed asWherein B is the number of samples in the training sample set, y _b is the true value of the sulfur dioxide concentration of the desulfurization outlet corresponding to the B-th sample,/>And the sulfur dioxide concentration predicted value of the desulfurization outlet corresponding to the b sample.

In a general embodiment, the optimizing process of the conventional SSA algorithm mainly includes:

And A1, initializing sparrow population and related parameters.

Step A2, updating the position of the finder by the following updating rule:

(formula one);

in the first step, Represents the current location of the/>First/>, in the next iterationIndividual at/>The position in the dimension, namely the updated position; /(I)Represents the/>First/>, in the next iterationIndividual at/>The position in the dimension, i.e., the pre-update position; /(I)，/>Represents the number of individuals in the sparrow population, where/>The value of (2) corresponds to the number of the discoverer in the sparrow population; j=1, 2,..d, d represents the number of dimensions of the hyper-parameter; /(I)Representing a maximum number of iterations; /(I)Representing a random number, and；/>Representing random numbers and obeying normal distribution; l represents/>Is a full 1 matrix of (2); /(I)Representing the early warning value, and；/>Represents a safety threshold, and/>. When the early warning value is smaller than the safety threshold, the foraging environment is safe, the foraging environment can be searched in a wide area, otherwise, the danger is found, and the whole population needs to leave the place immediately and fly to a safe area.

Step A3, updating the position of the subscriber through the following updating rules:

(formula II);

In the second step, the first step is performed, For/>The position of the optimal individual in the secondary iteration, namely the super-parameter optimal value which enables the prediction effect of the prediction model to be best so far; /(I)Represents the/>The position of the globally worst individual in the secondary iteration is the super-parameter value which makes the prediction effect of the prediction model worst; /(I)Representation/>The elements of which are randomly selected in 1 or-1,，/>Representation/>Is a transposed matrix of (a); when/>When (1)The food is obtained after the individual joiners approach the optimal location, thus continuing to execute the follow-up strategy,/>The representative participants follow the discoverers but do not rob food, and at this time, the policy needs to be changed to fly to other areas for foraging.

Step A4, the proportion of the alerter in the sparrow population is generally 10% -20%, the alerter and the joiner are randomly generated in the foraging process, the discoverer and the joiner can be alerters, and the positions of the alerters are updated through the following updating rules:

(formula III);

In the third step, the first step is performed, Represents the/>Global optimal position in the secondary iteration; /(I)Representing the current fitness of the individual; /(I)Representing the global optimum fitness so far; /(I)Representing a global worst fitness; /(I)And/>All represent step control parameters; /(I)Is a random number and obeys standard normal distribution; /(I)Is a random number, and/>；/>Taking the minimum constant to prevent the denominator from being 0; when the individual is currently fitness/>I.e., the individual is not at the optimal location, indicating that it is at the edge of the population, more susceptible to predation by natural enemies, and executing a strategy of moving toward the vicinity of the optimal location; when the individual is currently fitness/>When an individual currently in an optimal location is aware that there is a hazard to the surrounding environment, it will be close to other individuals around to avoid the hazard.

And A5, calculating the fitness value of the current sparrow population, judging whether the stopping condition is met, if so, ending iteration, outputting an optimal result, and otherwise, returning to the step A2.

As can be seen, the stop conditions are typically: the iteration times reach the maximum iteration times or individuals with fitness values meeting preset values exist in the current sparrow population.

In this embodiment, the above-described super-parameter optimization procedure is an improvement over the SSA algorithm procedure in the general embodiment. Wherein, when updating the position of the finder, a butterfly algorithm is adopted, namely the step A2 is improved; after updating the finder position, the joiner position and the alerter position in each iteration, an elimination recombination mechanism is introduced, namely, before the step A5 and after the step A4.

In a specific embodiment, the finder position is updated by adopting a butterfly algorithm at each iteration, and the updating rule is as follows:

When the early warning value is smaller than the safety threshold value, the current first First/>, in the next iterationIndividual discoverers at/>Post-update location of dimensions by the method at position/>First/>, in the next iterationIndividual discoverers at/>The second random value is obtained after the second random value is added on the basis of the position where the dimension is located, and the second random value is determined according to the first difference value and the fragrance perception intensity of the finder;

when the early warning value is greater than or equal to the safety threshold value, the current first First/>, in the next iterationIndividual discoverers at/>The updated position of the dimension is the position of the finder at the/>Randomly moving the positions according to normal distribution on the basis of the positions in the secondary iteration;

Wherein the first difference is the first The product of the optimal individual position and the first random number in the iteration is the same as the first/>First/>, in the next iterationIndividual discoverers at/>The locations where the dimensions are found are poor, and the perceived intensity of the fragrance is determined based on the perceived factor of the fragrance, the intensity of the stimulus, and a control constant for compressing the intensity of the stimulus.

For example, the update rules based on the butterfly algorithm may be expressed as follows:

(formula IV);

In the fourth step, the first step is performed, Represents the/>Optimal individual position in the secondary iteration; /(I)Representing a random number from 0 to 1,/>Representing the first random number; /(I)Represents the/>The individual discoverers' perceived intensity of fragrance, and/>，/>Representing fragrance perception factors,/>Representing the stimulus intensity,/>Indicating the control constant for compressing the stimulus intensity.

In one embodiment, the elimination recombination mechanism comprises the following implementation steps:

Sorting the fitness values of the individuals after the position updating;

Wherein, the preset rule refers to: first, the Individual at/>Reset position of dimension by at/>Individual's/>The first random value and the first/>, which are obtained by adding the first random value on the basis of the preset lower limit value of the super parameter represented by the dimensionIndividual's/>The ratio of the difference between the preset upper limit value and the lower limit value of the super parameter represented by the dimension is a random number between 0 and 1, and the first/>Individual at/>The reset position of the dimension is between the/>, when eliminatedIndividual at/>The position of the dimension and the global optimum individual position of the last iteration.

For example: assuming that the number of individuals eliminated due to low fitness values is denoted as E, E can be calculated by:

(formula five);

in the fifth step, the first step is performed, Representing an integer function for returning a minimum integer not smaller than the expression in brackets; /(I)The ratio of the eliminated individuals can be set to 0.1; /(I)Representing the total number of individuals in the current population.

For example: assume the firstIndividual's/>The preset upper limit value of the super parameter represented by the dimension is expressed as/>First/>Individual's/>The preset lower limit value of the super parameter represented by the dimension is expressed as/>The preset rule may be expressed by the following equation:

(formula six);

In the sixth step, the first step, Expressed at present/>First/>, in the next iterationIndividual at/>A reset position of the dimension; /(I)Expressed at present/>First/>, when eliminated in a second iterationIndividual at/>The location of the dimension, i.e., the historical location; /(I)Expressed in/>Global optimal individual position in the secondary iteration; /(I)The random numbers in the [0, 1] interval are represented.

In the prior art, the neural network applied to sulfur dioxide concentration prediction modeling of the desulfurization system mainly comprises an RBF network (radial basis function neural network, english is called Radial Basis Function), an LSTM network (Long Short-Term Memory network, english is called Long Short-Term Memory), an SVM network (support vector machine, english is called Support Vector Machine) and the like. However, the BP neural network (Back Propagation neural network, english is called Back Propagation) comprises an input layer, an hidden layer and an output layer, and has the characteristic of simple network structure. Therefore, in a specific embodiment, the network structure of the prediction model adopts a BP neural network, and accordingly, the value of the hyper-parameter is the value of all weights and biases in the BP neural network.

It is known that the influence factors of the sulfur dioxide concentration of the desulfurization outlet are numerous, and each influence factor has coupling characteristics, so that each influence factor with coupling relation can be subjected to correlation analysis to obtain a plurality of influence factors which are not related to each other, the influence factors which are not related to each other are used as the input of a prediction model, the prediction effect can be improved, and a principal component analysis method and the like can be adopted for the correlation analysis.

For example, in one particular embodiment, step S100, determining a characteristic variable that affects sulfur dioxide concentration at the desulfurization outlet includes:

step S110, acquiring field production data of a desulfurization system;

Step S120, performing dimension reduction on a multidimensional initial variable affecting the concentration of sulfur dioxide at a desulfurization outlet in the acquired field production data by adopting a nuclear principal component analysis method to obtain a characteristic variable; the kernel principal component analysis method is also called as KPCA (English full name: KERNEL PRINCIPAL Component Analysis) method.

In combination with the specific operating conditions of the desulfurization system, in one specific embodiment, the multidimensional initial variables affecting the outlet sulfur dioxide concentration include desulfurization inlet oxygen concentration, desulfurization inlet sulfur dioxide concentration, desulfurization inlet flue gas flow, desulfurization outlet oxygen concentration, desulfurization outlet flue gas flow, unit load, slurry concentration, slurry flow, adsorption column liquid level, oxidation blower current, and circulating pump current. It should be appreciated that the initial variables are not limited to the eleven variables described above, and that other initial variables may exist that affect the sulfur dioxide concentration at the desulfurization outlet for a particular operating condition of the desulfurization system.

Illustratively, in one embodiment, after taking the randomly generated hyper-parameters as the initial population of individuals, the method further comprises the following implementation steps:

and mapping the initial population by adopting chaotic mapping to obtain the mapped initial population.

The initial population of the SSA algorithm is randomly generated, the individual positions in the initial population are possibly smaller in distribution area at the initial time, so that the later stage is easy to fall into local optimum, the optimizing precision is influenced, and aiming at the problem of poor quality of the initial population, chaotic mapping is introduced in the technical scheme to map the initial population, the initial population is enabled to obtain more uniform and dispersed initial positions by utilizing the characteristic of strong randomness of a chaotic sequence, the global searching capacity of the super-parameter optimizing process is improved, and the predicting precision of a predicting model is further improved.

For example, in one particular embodiment, the chaotic map is a piecewise linear chaotic map, which is a form of chaotic map, also known as PWLCM (all English: PIECEWISE LINEAR Chaotic Map) map. The piecewise linear chaotic mapping process may be expressed as follows:

(formula seven);

In the seventh aspect, the method comprises the steps of, Representing a random number between (0, 1)/>For the first/>, in the initial populationLocation of individual,/>For the/>, in the mapped initial populationThe location of the individual.

In a specific embodiment, before training an initially constructed neural network, calculating mutual information between historical characteristic variables under different time lags and sulfur dioxide concentrations corresponding to the historical characteristic vectors in a training sample by adopting a mutual information method, taking a time difference between the historical characteristic variable with the maximum mutual information and the sulfur dioxide concentration corresponding to the historical characteristic variable as delay time, and performing delay compensation on all the historical characteristic variables in the training sample by utilizing the delay time.

It is known that the mutual information method, also called MI (english: mutual Information) algorithm, is a measure of the degree of interdependence of two random variables, and that the larger the mutual information, the larger the correlation between the two random variables, and vice versa.

For example, for two discrete sets of random variables X and Y, mutual information between the twoTwo randomly varying edge entropies/>, can be used、/>And joint entropy/>The expression is as follows:

(formula eight);

(formula nine);

(formula ten);

(formula eleven);

In the nine of which, Edge probability distribution function representing random variable X,/>Representing the dimension of the random variable X; in ten,/>Edge probability distribution function representing random variable Y,/>Representing the dimension of the random variable Y; in the eleventh aspect, the present invention provides a method,Representing the joint probability distribution function of the random variable X and the random variable Y.

Based on this, mutual information between the twoIt can also be expressed as follows:

(equation twelve).

In the prior art, the problem of time delay exists between a characteristic variable (an input variable of a prediction model) influencing the sulfur dioxide concentration of a desulfurization outlet and the sulfur dioxide concentration (an output variable of the prediction model), the delay time corresponding to the maximum mutual information is determined through a mutual information method, and the delay compensation is carried out on the input variable through the delay time, so that the correlation between the input variable and the output variable of the prediction model is improved, the accuracy of a training sample is improved, and the prediction precision of the prediction model is further improved.

In one application, a prediction model is constructed by adopting a BP neural network, a kernel principal component analysis method is adopted to reduce the dimension of a multidimensional initial variable affecting sulfur dioxide concentration in field data of a desulfurization system, delay compensation is carried out through a mutual information method after the dimension reduction, the BP neural network which is initially constructed is trained by utilizing a training sample after the delay compensation, an improved SSA algorithm is adopted when optimization of each weight and bias in the BP neural network is carried out, the model is defined as an MSSA algorithm in an application example, namely, after an initial population is constructed, piecewise linear chaotic mapping is adopted, a butterfly algorithm is adopted to update the position of a finder, and a elimination recombination mechanism is introduced after the position of the finder, the position of a subscriber and the position of an alerter are updated in each iteration. Based on this, the prediction model in the present application example is referred to as a KPCA-MI-MSSA-BP network model, and the construction process of the model is shown in fig. 2.

Specifically, the construction process of the KPCA-MI-MSSA-BP network model comprises the following implementation steps:

step SS1, historical production data of a desulfurization system is obtained.

And step SS2, performing decentralization treatment on the historical production data to obtain a standardized data set.

And step SS3, performing dimension reduction on the standardized data set by adopting a kernel principal component analysis method to obtain a training sample.

Specifically, 2000 groups of typical working condition points are selected, abnormal data points are removed according to the 3 sigma principle, a training set and a verification set are divided according to 8:2, different kernel function checking effects are selected, and under the same data dimension, a linear kernel function with the highest contribution rate is taken for KPCA dimension reduction, and data normalization is carried out.

And (3) adaptively determining the optimal hidden layer node number of the BP neural network according to the mean square error of the training sample so as to determine the basic structure of the BP neural network as shown in figure 3. In fig. 3, the input layer receives feature variables after dimension reduction, and the feature variables after dimension reduction include eight, so each feature variable corresponds to nodes m ₁ to m ₈ of the input layer one by one, a node of the hidden layer is denoted by sigma, an output layer is denoted by y, the hidden layer is responsible for weighting and biasing the input feature variable and sending a result to the output layer after calculation, and the output layer outputs a dependent variable, namely a predicted value of sulfur dioxide concentration of a desulfurization outlet.

And step SS4, performing time delay analysis and time delay compensation on characteristic variables in the training samples by adopting a mutual information method to obtain target training samples.

And step SS5, optimizing all weights and biases in the BP neural network by adopting an MSSA algorithm according to the target training sample, and assigning values to the BP neural network by utilizing the obtained optimal weights and the optimal biases to obtain a prediction model.

Wherein, the MSSA algorithm is adopted to optimize the ownership weight and bias in the BP neural network, comprising the following implementation steps:

at step SS51, an initial population is randomly generated.

And step SS52, mapping the initial population by adopting chaotic mapping to obtain the mapped initial population.

And step SS53, iteratively updating the positions of the discoverer, the joiner and the alerter according to the training sample and the mapped initial population, calculating the fitness value after updating each time, eliminating the individuals in the population according to the fitness value, resetting the positions of the eliminated individuals to form a new population with the individuals which are not eliminated until the stopping condition is met, and obtaining the optimal weight and the optimal bias of the BP neural network.

Referring to fig. 4 to 7, the validity of the prediction model constructed by the above application is verified in conjunction with the specific example.

Historical production data of a desulfurization system of a certain power plant is collected, 2000 groups of representative data with a sampling period of 5s are used for prediction, a training set and a verification set are divided, 1500 groups are taken as the training set, and 500 groups are taken as the test set. The evaluation indexes of the prediction model are selected from RMSE (totally: root Mean SquareError, root mean square error), MAPE (totally: mean Absolute Percentage Error, mean absolute percentage error) and MAE (Mean Absolute Error ).

Firstly, the improvement effect of the KPCA method on the prediction precision of the BP neural network is verified. The KPCA method is applied to reduce the dimension of the original data, extract low-dimension components which can fully represent the characteristics of the original data, and reduce the calculated amount. The initial variable of ten dimensions is reduced in dimension using the KPCA method. The contribution rates after the dimension reduction by adopting different kernel functions are shown in the table I. Comprehensively considering that the KPCA method adopting the linear kernel has the best effect of dimension reduction on the data of the experiment, and can fully represent the characteristic information of the original data, therefore, the linear kernel is selected as a kernel function of sulfur dioxide concentration prediction at the time, the first eight main components are taken, and the cumulative contribution rate is 97.17 percent, as shown in a table II.

List one

Watch II

In order to explore the influence of the KPCA method on the accuracy of the prediction model, the model effects before and after the KPCA dimension reduction are compared, the result is shown in a third table, and under the same actual working condition, compared with a traditional BP neural network, the accuracy of the prediction model is improved while the calculated amount of the KPCA dimension reduction is reduced.

Watch III

Besides the introduction of the KPCA method, the BP network further improves the prediction accuracy of the sulfur dioxide concentration of the desulfurization outlet by utilizing the mutual information method to perform time delay compensation. In order to explore the influence of the mutual information method on the accuracy of a prediction model based on the KPCA-BP method, the comparison result of the prediction effect of the model after time delay compensation (the prediction model based on the KPCA-MI-BP method) is shown in a table IV, the RMSE after time delay compensation is reduced by 12.46% and the overall accuracy is increased by 0.4792% under the same actual working condition.

Table four

Finally, compared with a prediction model based on a KPCA-MI-MSSA-BP method, a prediction model (a prediction model based on a KPCA-MI-MSSA-BP method) obtained by performing super-parameter optimization on a BP neural network by using an MSSA algorithm has higher stability and prediction precision compared with a prediction model based on a KPCA-MI-SSA-BP method, a traditional BP neural network, a traditional RBF neural network and a traditional SVM network, the prediction model has higher stability and prediction precision, and the comparison result is shown in a table five.

TABLE five

Device embodiment

Referring to fig. 8, an embodiment of the present invention provides a sulfur dioxide concentration prediction apparatus 400, including a feature variable determining module 410 and a sulfur dioxide concentration prediction module 420 connected in sequence, wherein:

A characteristic variable determination module 410 for determining a characteristic variable affecting the sulfur dioxide concentration at the desulfurization outlet;

The sulfur dioxide concentration prediction module 420 is configured to input the feature variable into a constructed prediction model to obtain a sulfur dioxide concentration prediction result, where the prediction model is obtained by performing super-parameter optimization and training on an initially constructed neural network;

Wherein, the super parameter optimization process is:

Taking the randomly generated super parameters as initial population individuals, and calculating individual fitness values, wherein the individuals are divided into discoverers and joiners;

the alerter randomly generates from the discoverer and the joiner through the foraging process, a butterfly algorithm is adopted when the position of the discoverer is updated, the position of the individual represents the value of the super parameter, and the position of the eliminated individual is between the position of the eliminated individual and the position of the last iterative global optimal individual.

In one embodiment, the method includes eliminating individuals in a population according to fitness values, resetting positions of the eliminated individuals to form a new population with non-eliminated individuals, and the method includes:

Sorting the fitness values of the individuals after the position updating;

The preset rules are as follows: first, the Individual at/>Reset position of dimension by at/>Individual's/>The first random value and the first/>, which are obtained by adding the first random value on the basis of the preset lower limit value of the super parameter represented by the dimensionIndividual firstThe ratio of the difference between the preset upper limit value and the lower limit value of the super parameter represented by the dimension is a random number between 0 and 1, and the first/>Individual at/>The reset position of the dimension is between the/>, when eliminatedIndividual at/>The position of the dimension and the global optimum individual position of the last iteration.

In a specific embodiment, after taking the randomly generated hyper-parameters as the individuals of the initial population, mapping the initial population by adopting chaotic mapping to obtain the mapped initial population.

In a specific embodiment, before training an initially constructed neural network, calculating mutual information between a historical characteristic variable under different time lags and sulfur dioxide concentration corresponding to the historical characteristic vector in a training sample by adopting a mutual information method, taking a time difference between the historical characteristic variable with the maximum mutual information and the sulfur dioxide concentration corresponding to the historical characteristic variable as delay time, and performing delay compensation on all the historical characteristic variables in the training sample by utilizing the delay time.

In a specific embodiment, the neural network is a BP neural network, and the value of the hyper-parameter is the value of all weights and biases in the BP neural network.

In one embodiment, determining a characteristic variable that affects sulfur dioxide concentration at a desulfurization outlet includes:

acquiring field production data of a desulfurization system;

and (3) reducing the dimension of a multidimensional initial variable influencing the sulfur dioxide concentration of the desulfurization outlet in the on-site production data by adopting a nuclear principal component analysis method to obtain a characteristic variable.

In one embodiment, the initial variables include desulfurization inlet oxygen concentration, desulfurization inlet sulfur dioxide concentration, desulfurization inlet flue gas flow, desulfurization outlet oxygen concentration, desulfurization outlet flue gas flow, unit load, slurry concentration, slurry flow, adsorption column liquid level, oxidation blower current, and circulating pump current.

The sulfur dioxide concentration prediction apparatus 400 includes a processor and a memory, and both the characteristic variable determination module 410 and the sulfur dioxide concentration prediction module 420 are stored as program elements in the memory, and the program elements stored in the memory are executed by the processor to implement the corresponding functions. The processor includes a kernel, and the kernel fetches the corresponding program unit from the memory. The kernel can be provided with one or more than one, and the sulfur dioxide concentration prediction method is realized by adjusting kernel parameters. The memory may include volatile memory, random Access Memory (RAM), and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM), among other forms in computer readable media, the memory including at least one memory chip.

In another aspect, embodiments of the present invention also provide a machine-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the respective sulfur dioxide concentration prediction methods described above.

In yet another aspect, an embodiment of the present invention further provides a processor, where the processor is configured to run a program, and when the program runs, execute the above-mentioned respective sulfur dioxide concentration prediction methods.

In yet another aspect, an embodiment of the present invention further provides a computer device, where the computer device may be a terminal, and an internal structure diagram of the computer device may be as shown in fig. 9. The computer apparatus includes a processor a01, a network interface a02, a display screen a04, an input device a05, and a memory (not shown in the figure) which are connected through a system bus. Wherein the processor a01 of the computer device is adapted to provide computing and control capabilities. The memory of the computer device includes an internal memory a03 and a nonvolatile storage medium a06. The nonvolatile storage medium a06 stores an operating system B01 and a computer program B02. The internal memory a03 provides an environment for the operation of the operating system B01 and the computer program B02 in the nonvolatile storage medium a06. The network interface a02 of the computer device is used for communication with an external terminal through a network connection. The computer program, when executed by the processor a01, implements a sulfur dioxide concentration prediction method. The display screen a04 of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device a05 of the computer device may be a touch layer covered on the display screen, or may be a key, a track ball or a touch pad arranged on a casing of the computer device, or may be an external keyboard, a touch pad or a mouse.

It will be appreciated by persons skilled in the art that the architecture shown in fig. 9 is merely a block diagram of some of the architecture relevant to the present inventive arrangements and is not limiting as to the computer device to which the present inventive arrangements are applicable, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In one embodiment, the sulfur dioxide concentration prediction apparatus provided by the present invention may be implemented in the form of a computer program that is executable on a computer device as shown in fig. 9. The memory of the computer device may store various program modules constituting the sulfur dioxide concentration prediction apparatus 400, for example, the characteristic variable determination module 410 and the sulfur dioxide concentration prediction module 420 shown in fig. 8. The computer program of each program module causes the processor to carry out the steps in the method for predicting the concentration of sulphur dioxide of the method embodiments described in the present specification.

The computer device shown in fig. 9 may perform step S100 through the characteristic variable determining module 410 in the sulfur dioxide concentration predicting apparatus 400 shown in fig. 8, and the computer device may perform step S200 through the sulfur dioxide concentration predicting module 420.

The embodiments of the present invention also provide a computer program product adapted to perform, when executed on a data processing apparatus, a program initialized with the method steps of:

Wherein, the super parameter optimization process is:

Sorting the fitness values of the individuals after the position updating;

acquiring field production data of a desulfurization system;

It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, etc., such as Read Only Memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

Computer-readable media include both permanent and non-permanent, removable and non-removable media, and information storage may be implemented by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.

The foregoing is merely exemplary of the present invention and is not intended to limit the present invention. Various modifications and variations of the present invention will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are to be included in the scope of the claims of the present invention.

Claims

1. A method for predicting sulfur dioxide concentration, the method comprising:

Wherein, the super parameter optimization process is:

2. The method for predicting sulfur dioxide concentration according to claim 1, wherein the step of eliminating individuals in the population according to the fitness value, and forming a new population with individuals not eliminated after resetting the positions of the eliminated individuals, comprises:

Sorting the fitness values of the individuals after the position updating;

Sequentially eliminating individuals with preset proportions according to the fitness value from low to high;

3. The method for predicting sulfur dioxide concentration according to claim 1, wherein after taking the randomly generated hyper-parameters as the individuals of the initial population, mapping the initial population by adopting chaotic mapping to obtain the mapped initial population.

4. The sulfur dioxide concentration prediction method according to claim 1, wherein before training an initially constructed neural network, calculating mutual information between a historical characteristic variable under different time lags and sulfur dioxide concentrations corresponding to the historical characteristic vector in a training sample by adopting a mutual information method, taking a time difference between the historical characteristic variable with the maximum mutual information and the sulfur dioxide concentration corresponding to the historical characteristic variable as delay time, and performing delay compensation on all the historical characteristic variables in the training sample by utilizing the delay time.

5. The sulfur dioxide concentration prediction method according to claim 1, wherein the neural network is a BP neural network, and the super-parameter is the value of all weights and offsets in the BP neural network.

6. The sulfur dioxide concentration prediction method according to claim 1, wherein the determining of the characteristic variable affecting the sulfur dioxide concentration at the desulfurization outlet comprises:

acquiring field production data of a desulfurization system;

7. The sulfur dioxide concentration prediction method of claim 6, wherein the initial variables comprise desulfurization inlet oxygen concentration, desulfurization inlet sulfur dioxide concentration, desulfurization inlet flue gas flow, desulfurization outlet oxygen concentration, desulfurization outlet flue gas flow, unit load, slurry concentration, slurry flow, adsorption column liquid level, oxidation blower current, and circulating pump current.

8. A sulfur dioxide concentration prediction apparatus, the apparatus comprising:

Wherein, the super parameter optimization process is:

9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the sulphur dioxide concentration prediction method according to any one of claims 1 to 7 when executing the program.

10. A machine readable storage medium having stored thereon a computer program, which when executed by a processor implements the sulphur dioxide concentration prediction method according to any one of claims 1 to 7.