A prediction method for pollutant emission concentration of a biomass cogeneration system

Abstract

The present invention relates to a method for predicting the pollutant emission concentration of a biomass cogeneration system, establishing a prediction model and a correlation function system for the pollutant emission concentration of the biomass cogeneration system; initializing the parameters of the prediction model and the correlation function system, and offline training the prediction model to obtain the parameter variables in the prediction model, so as to obtain the predicted concentration of the pollutant emission of the biomass cogeneration system with the trained prediction model; the method can be applied to the biomass cogeneration system, and by inputting a number of variable data, the predicted pollutant emission concentration of the system is output by the prediction model and the correlation function system. The present invention is simple in design, easy to understand, strong in practicability, wide in application range, has extremely strong interpretability, and can greatly shorten the operation time on the premise of ensuring the prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of computing, reckoning or counting, and is applied to the field of biomass cogeneration. In particular, it relates to a method for predicting the pollutant emission concentration of a biomass cogeneration system based on a fast relevance vector machine. Background Art

[0002] With the development of technology and the progress of society, the optimization of the ecological environment will be a long-term cause pursued by mankind. In this context, clean, stable and efficient biomass cogeneration systems are increasingly favored.

[0003] However, as a renewable energy source, biomass is often limited by the diversity and intermittency of resources, which makes the cogeneration system operate under frequent, rapid variable loads or long-term low loads, thus causing unstable combustion conditions and increasing pollutant emissions.

[0004] The concentration of pollutants emitted by a biomass cogeneration system can directly reflect the stability of the system's combustion conditions. In order to timely grasp the stability of the combustion conditions of a biomass cogeneration system, a key technology is proposed to predict the pollutant emission concentration of a biomass cogeneration system.

[0005] In the prior art, the methods for predicting the pollutant emission concentration of a biomass cogeneration system mainly include the instrument method, the simulation method and the data-driven method. The instrument method uses, for example, a continuous emission prediction system to monitor pollutant emissions in real time. Due to the high cost of instrument maintenance and frequent offline operation, it is difficult to ensure continuous and stable continuous monitoring. The simulation method can clearly explain the process of pollutant formation, but its simulation analysis is complex and the calculation cost is high, making it difficult to achieve online monitoring. The data-driven method uses artificial intelligence algorithms, such as various artificial neural networks, to monitor pollutant emissions. However, the main obstacle faced by this method is the selection of input variables. It is difficult to determine which variables are optimal. If blindly selected, it is easy to bring the risk of data redundancy to the prediction model, and the artificial neural network is similar to a black box model, with poor interpretability. Therefore, it is difficult to directly predict the pollutant concentration of a biomass cogeneration system. Summary of the Invention

[0006] The present invention provides a method for predicting the pollutant emission concentration of a biomass cogeneration system, which overcomes the problems of lag, complex calculation and high economic cost existing in the existing pollutant emission prediction methods for biomass cogeneration systems, and has the advantages of simple design, easy operation and strong interpretability.

[0007] The technical solution adopted by the present invention is a method for predicting the pollutant emission concentration of a biomass cogeneration system, and the method establishes a prediction model and an associated function system for the pollutant emission concentration of a biomass cogeneration system;

[0008] Initialize the parameters of the prediction model and the associated function system, and train the prediction model offline to obtain the parameter variables in the prediction model, so as to obtain the predicted concentration of pollutants in the biomass cogeneration system with the trained prediction model.

[0009] Preferably, the historical data set of the given pollutant emission concentration is {(x n , t n ), n = 1, 2, …, N}, x n ∈ R N×N , t n ∈ R N , where x n represents the input matrix composed of the variable data of the biomass cogeneration system, t n represents the pollutant emission concentration of the biomass cogeneration system, n is the sample index, and there are a total of N samples;

[0010] A total of 23 variables of the biomass cogeneration system are actually collected, including the active power of the generator, the instantaneous value of the main steam flow, the outlet temperature of the boiler main steam header, the total flow of the primary air duct, the outlet temperature of the primary hot air, the total flow of the secondary air duct, the outlet temperature of the secondary hot air, the flue gas temperature (left), the flue gas temperature (right), the temperatures at different positions in the combustion chamber, the furnace outlet temperature (left), the furnace outlet temperature (right), the oxygen content at the outlet of the low-temperature superheater (left), the oxygen content at the outlet of the low-temperature superheater (right), the cyclone outlet temperature (left), the cyclone outlet temperature (right), the oxygen content of the flue gas at the outlet of the air preheater (left), the oxygen content of the flue gas at the outlet of the air preheater (right); the temperatures at different positions in the combustion chamber include the temperature below the boiling point of the combustion chamber (front 6), the temperature below the boiling point of the combustion chamber (rear 3), the temperature in the middle of the boiling point of the combustion chamber (left 1), the temperature in the middle of the combustion chamber (right 1), the temperature in the middle of the combustion chamber (left 3), the temperature in the middle of the combustion chamber (right 3).

[0011] According to the relevant knowledge of probability theory, the model of the pollutant emission concentration of the biomass cogeneration system is Equation (1),

[0012] t n = y(x n ; w) + ε n (1)

[0013] where ε n ∈ (0, 1), ε n represents the random noise of the biomass cogeneration system, which follows a Gaussian distribution with a mean of 0 and an inverse variance (precision) of β, w represents the weight vector, w = [w 0 , w 1 , …, w N T , w i ∈ (0, 1), and i is an integer from 1 to N.​

[0014] Under the observed values of the input variables and pollutant concentrations of the known biomass cogeneration system, the mean value y(x n ) of the pollutant emission concentration of the biomass cogeneration system is expressed as Equation (2),

[0015]

[0016] where M = N + 1, Φ(x n ) is the basis vector, Φ(x n ) = [Φ(x 1 ), Φ(x 2 ), …, Φ(x N )] T , Φ(x n ) ∈ R N ; Φ(x i ) = [1 K(x n , x 1 ) K(x n , x 2 ) … K(x n , x N )] T , K(x n , x i ) is a kernel function, not restricted by positive definite kernels.

[0017] Preferably, the pollutant emission concentration correlation function system includes a prior probability density function of the emission concentration-related weight, a posterior probability density function of the emission concentration-related weight, a log-likelihood function of the emission concentration-related weight precision and the random noise precision, and a fast marginal likelihood function of the emission concentration-related weight precision and the random noise precision.

[0018] Preferably, a prior probability density function of the emission-related weight is established. The pollutant emission concentration t of the biomass cogeneration system n obeys an independent distribution. According to the relevant knowledge of probability theory, the likelihood function of t n is Equation (3),

[0019]

[0020] where Φ is an N×(N + 1) matrix, which is the aforementioned basis vector and is abbreviated here; if the derivative is directly taken for Equation (3) according to the traditional parameter solving method, overfitting is very likely to occur. Therefore, the weight w is constrained to obey a prior Gaussian distribution with a mean of 0, and its prior conditional probability density function is Equation (4),

[0021]

[0022] where M = N + 1, α i represents the precision corresponding to the weight w i , i = 1, 2, …, M, and its value directly affects the strength of the prior distribution acting on each weight parameter, and is also the key to ensuring the sparsity of the model.

[0023] Preferably, based on the prior probability density function of the emission-related weight, and given the prior likelihood function, according to Bayes' theorem (the posterior distribution is proportional to the product of the prior distribution and the likelihood function), the posterior probability density function of the emission concentration-related weight is established as Equation (5),

[0024]

[0025] where Σ is the covariance of the posterior distribution, μ is the mean, Σ = (βΦ T Φ + A) -1 , μ = βΣΦ T t, A = diag(α 1 , α 2 , …, α M ), and diag represents a diagonal matrix.

[0026] Preferably, based on the given pollutant emission concentration data x n , the marginal likelihood function of the emission concentration weight precision and the random noise precision can be obtained by integrating over the weight w; by integrating over the weight w, the log-likelihood function of the emission concentration-related weight precision and the random noise precision is established. First, Equation (6) is established,

[0027]

[0028] where the matrix C is an N×N matrix, C = β -1 I + ΦA -1 Φ T , and introducing this matrix can accelerate the operation speed of subsequent solving the weight precision and the random error precision;

[0029] Taking the logarithm of Equation (6) gives Equation (7),

[0030]

[0031] is the log-likelihood function of the emission concentration with respect to the weight precision and the random noise precision.

[0032] Preferably, a fast marginal likelihood function of the emission concentration with respect to the weight precision and the random noise precision is established. This algorithm quickly estimates the hyperparameters of the given data samples, removes the irrelevant vectors of the training samples, ensures the sparsity of the model, thereby reducing the training time, enabling the hyperparameters to quickly reach stable values and thus obtaining the weights. Transforming Equation (7) gives Equation (8),

[0033]

[0034] Among them, s i is the sparsity factor, which is used to measure the overlap degree between the basis function Φ i and the basis function Φ i already existing in the model. The basis function Φ i is an element in the basis vector Φ. The expression of s i is q i is the quality factor, which is used to measure the error generated by the model after removing the basis function Φ i . The expression is

[0035] Derive L(α i ) with respect to the weight precision α i and set it to zero to obtain Equation (9).

[0036]

[0037] Similarly, solve the expression of the random noise precision. Maximize Equation (6) and set its derivative to zero to obtain Equation (10).

[0038]

[0039] Among them, Σ mm is the m-th diagonal element of the posterior variance Σ.

[0040] Preferably, for the model after training, after learning a large amount of pollutant emission data, most hyperparameters will approach infinity, and the corresponding weights will be 0, so that the entire model has high sparsity. Given the data value x * of the biomass cogeneration system variables, the predicted value t * of the corresponding pollutant concentration satisfies the probability density function of Equation (11).

[0041]

[0042] Among them, α * , β * are the values of the emission concentration-related weight precision and the random noise precision obtained through the fast marginal likelihood function, respectively.

[0043] The present invention provides a method for predicting the pollutant emission concentration of a biomass cogeneration system, establishing a prediction model and a correlation function system for the pollutant emission concentration of the biomass cogeneration system; initializing the parameters of the prediction model and the correlation function system, and offline training the prediction model to obtain the parameter variables in the prediction model, so as to obtain the predicted concentration of the pollutant emission of the biomass cogeneration system with the trained prediction model; the method can be applied to the biomass cogeneration system, and by inputting a number of variable data, the predicted pollutant emission concentration of the system is output by the prediction model and the correlation function system.

[0044] The beneficial effects of the present invention are as follows:

[0045] (1) Simple design, easy to understand, strong practicability, wide practicability;

[0046] (2) With extremely strong interpretability;

[0047] (3) Compared with the support vector machine algorithm, it can greatly shorten the running time on the premise of ensuring the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of the present invention;

[0049] Figure 2 is a flow chart of the application of the present invention;

[0050] Figure 3 is an implementation flow chart of the biomass cogeneration system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] The following further describes the present invention in detail with reference to embodiments, but the protection scope of the present invention is not limited thereto.

[0052] The main execution part of the present invention runs and is implemented on the process control computer for predicting the pollutant concentration of the biomass cogeneration system.

[0053] The application process of the present invention is divided into three stages:

[0054] Parameter initialization, setting appropriate initial values for the hyperparameters in the proposed method, importing the variable data related to the biomass cogeneration system and the corresponding pollutant emission concentration in the import interface of the control computer model; setting the accuracy of the initial random error of the biomass cogeneration system in the parameter setting interface;

[0055] Offline training: First, write the algorithm in the form of code and input it into the control computer to initially build the model framework. Then, input the variable data related to the biomass cogeneration system and the corresponding pollutant emission concentration data into the algorithm to train the algorithm to obtain the final appropriate hyperparameter values. Finally, a trained algorithm model can be obtained;

[0056] Online prediction: Start the CPU in the computer to read the parameter values. By online predicting the variable data of the biomass cogeneration system and executing the corresponding algorithm program, the corresponding pollutant concentration can be predicted in real time;

[0057] Due to the rapidity of the algorithm, the program running time can be approximately ignored, and the algorithm can ultimately fully achieve real-time performance.

[0058] As Figure 1 shown, the present invention relates to a method for predicting the pollutant emission concentration of a biomass cogeneration system. First, a prediction model for the emission concentration is initially established, and based on this, the parameters involved in the model are further solved by the Bayesian method. The prior probability distribution, posterior probability distribution of the weights related to the emission concentration, the logarithmic likelihood function and the fast marginal likelihood function of the precision of the weights related to the emission concentration and the precision of the random noise are constructed in sequence. After taking the derivative of the fast marginal likelihood function and setting it to zero, the values of the parameters in the model can be finally obtained. Finally, the predicted value of the emission concentration can be obtained by combining the formula with the obtained parameter values.

[0059] The method includes the following steps:

[0060] Step 1: Initially construct a prediction model for the pollutant emission concentration of the biomass cogeneration system;

[0061] Assume that the given dataset of pollutant emission concentrations is {(x n ,t n ), n = 1, 2,..., N}, where x n ∈R N×N represents the input matrix of each variable of the biomass cogeneration system, and t n ∈R NDenote the pollutant emission concentration of the biomass cogeneration system; \(n\) represents the sample index, with a total of \(N\) samples; here \(N = 23\), corresponding to 23 variables, including the active power of the generator, the instantaneous value of the main steam flow, the outlet temperature of the main steam header of the boiler, the flow rate of the primary air main pipe, the outlet temperature of the primary hot air, the flow rate of the secondary air main pipe, the outlet temperature of the secondary hot air, the flue gas temperature (left), the flue gas temperature (right), the temperature before the boiling temperature in the combustion chamber (the first 6), the temperature after the boiling temperature in the combustion chamber (the last 3), the temperature on the left side of the middle of the combustion chamber (the first one), the temperature on the right side of the middle of the combustion chamber (the first one), the temperature on the left side of the middle of the combustion chamber (the third one), the temperature on the right side of the middle of the combustion chamber (the third one), the furnace outlet temperature (left), the furnace outlet temperature (right), the oxygen content at the outlet of the low-temperature superheater (left), the oxygen content at the outlet of the low-temperature superheater (right), the cyclone outlet temperature (left), the cyclone outlet temperature (right), the oxygen content of the flue gas at the outlet of the air preheater (left), the oxygen content of the flue gas at the outlet of the air preheater (right).

[0062] According to the relevant knowledge of probability theory, consider representing the model of the pollutant emission concentration of the biomass cogeneration system as shown in Equation (1)

[0063] t n = y(x n ; w) + ε n (1)

[0064] In Equation (1), ε n ∈(0,1) represents the random noise of the biomass cogeneration system, and it is assumed to follow a Gaussian distribution with a mean of 0 and a precision (the inverse of the variance) of β; w represents the weight vector, w = [w 0 , w 1 , …, w N T , w i ∈(0,1), and \(i\) is an integer from 1 to \(N\).

[0065] In Equation (1), the mean value \(y(x)\) of the pollutant emission concentration of the biomass cogeneration system is defined as Equation (2),

[0066]

[0067] In Equation (2), \(M = N + 1\), \(\varPhi(x n )\) is the basis vector, \(\varPhi(x n ) = [\varPhi(x 1 ), \varPhi(x 2 ), …, \varPhi(x N )] T ∈R N ; \(\varPhi(x i ) = [1 K(x n , x 1 ) K(x n , x 2 ) … K(x n , x​N )] T , K(x n , x i ) is a kernel function, not restricted by positive definite kernels.

[0068] Step 2: Construct the prior probability density function of the relevant weights of the pollutant emission concentration of the biomass cogeneration system;

[0069] Assume that the pollutant emission concentration t of the biomass cogeneration system n obeys an independent distribution. According to the relevant knowledge of probability theory, the likelihood function of t is Equation (3),

[0070]

[0071] In Equation (3), Φ is an N×(N + 1) matrix, which is the basis vector mentioned in Step 1 and is abbreviated here.

[0072] If the traditional parameter solution method is used to directly take the derivative of Equation (3), overfitting will occur. Therefore, consider constraining the weight vector w. Assume that it follows a prior Gaussian distribution with a mean of 0, and its prior conditional probability density function is Equation (4),

[0073]

[0074] In Equation (4), M = N + 1, α i (i = 1, 2,..., M) represents the precision corresponding to the relevant weight w of the emission concentration i of, and its value directly affects the strength of the prior distribution acting on each weight parameter, and is also the key to ensuring the sparsity of the model.

[0075] Step 3: Construct the posterior probability density function of the relevant weights of the pollutant emission concentration of the biomass cogeneration system;

[0076] Given the prior likelihood function, according to Bayes' formula in probability theory (the posterior distribution is proportional to the product of the prior distribution and the likelihood function), the posterior conditional probability density function can be obtained as Equation (5),

[0077]

[0078] Among them, the covariance and mean of the posterior distribution are respectively

[0079] Σ = (βΦ T Φ + A) -1

[0080] μ = βΣΦ T t

[0081] where the matrix A = diag(α 1 , α2 , …, α M ), diag represents a diagonal matrix.

[0082] Step 4: Construct a log-likelihood function related to the pollutant emission concentration related weight precision and random noise precision of the biomass cogeneration system;

[0083] Based on the given pollutant emission concentration data x n , the marginal likelihood function of the emission concentration weight precision and random noise precision can be obtained by integrating the weight w. First, establish Equation (6),

[0084]

[0085] where the matrix C = β -1 I + ΦA -1 Φ T is an N×N matrix. Introducing this matrix can make the subsequent operations of solving the weight precision and random error precision faster. Take the logarithm of Equation (6) to obtain the log-likelihood function of the weight precision and random noise noise precision, as shown in Equation (7),

[0086]

[0087] Step 5: Construct a fast marginal likelihood function related to the pollutant emission concentration related weight precision and random noise precision of the biomass cogeneration system;

[0088] To make the hyperparameters quickly reach stable values to obtain the weights, the present invention introduces a fast marginal likelihood algorithm. This algorithm quickly estimates the hyperparameters of the given data samples, removes the irrelevant vectors of the training samples, ensures the sparsity of the model, and thus reduces the training time. Transform Equation (7) to obtain Equation (8),

[0089]

[0090] where s i is the sparsity factor, used to measure the overlap degree between the basis vector Φ i and the basis vector Φ i already existing in the model. The expression is q i is the quality factor, which is an important parameter to measure the error generated by the model after removing the basis function Φ i . The expression is

[0091] Take the derivative of L(α i ) with respect to the hyperparameter α i and set it to zero to obtain Equation (9),

[0092]

[0093] Similarly, the expression for the random error precision can be solved. Maximizing Equation (6) and setting its derivative to zero gives Equation (10).

[0094]

[0095] where Σ mm is the m-th diagonal element of the posterior variance Σ.

[0096] Step 6: Obtain the predicted concentration of pollutants emitted by the biomass cogeneration system;

[0097] After learning a large amount of pollutant emission data, most hyperparameters will tend to infinity, and their corresponding weights will be 0, making the entire model highly sparse. If a data value of a variable of the biomass cogeneration system is given as x * , the predicted value t * of the corresponding pollutant concentration follows the distribution of Equation (11).

[0098]

[0099] where α * , β * are the values of the weight precision and random noise precision related to the emission concentration obtained through the fast marginal likelihood function.

[0100] The method of the present invention can be applied to a biomass cogeneration system. By inputting several variable data, the predicted concentration of pollutants emitted by the system can be output.

[0101] As Figure 2 shown, it is a schematic diagram of the specific operation steps for predicting the pollutant emission concentration of a biomass cogeneration system using the method of the present invention. First, initialize the random noise precision of the biomass cogeneration system, set the number of iterations and parameter thresholds, find a candidate basis function set through a certain method, then select one of the basis functions and initialize the precision of the system weights using a formula; next, calculate the posterior mean, covariance, sparsity, quality, etc. for all candidate basis functions; secondly, start the screening of candidate basis functions. The screening principle is determined by the difference between sparsity and quality to further clarify the situation of candidate basis functions and whether to update the precision of the system weights. Repeat this screening process until the parameters reach the threshold or the algorithm reaches the number of iterations; finally, the appropriate values of the hyperparameters can be obtained.

[0102] The operation steps include:

[0103] (1) On the basis of fully understanding and mastering the algorithm theory knowledge, initialize the random noise precision β of the biomass cogeneration system. Generally, the initial value of β is taken as 0.1×var[t] according to experience; reasonably set the number of iterations, hyperparameter thresholds, etc., and select a candidate set of basis vectors by a certain method;

[0104] (2) Select a basis function Φ from the candidate basis vectors i (i = 1, 2, …, N), initialize the precision α of the weights related to the emission concentration of the biomass cogeneration system i , and initialize α according to the formula This formula is the calculation method of the best initial value for the experimental results after verification. The remaining weight precisions α i (m ≠ i) are all initialized to infinity; m (3) For all candidate basis functions, calculate the corresponding posterior covariance Σ and mean μ using the following formula;

[0105] Σ = (βΦ

[0106] Φ + A) T -1

[0107] μ = βΣΦ T t

[0108] Next, for all candidate basis functions, calculate the corresponding s m and q m (m = 1, 2, …, M),

[0109]

[0110]

[0111] Since the matrices involved in the above formula are relatively complex to calculate in the actual biomass cogeneration system, they can be replaced by C in the actual industrial process -1 , and at this time, S m and Q m are respectively

[0112]

[0113]

[0114] The replaced S m and Q m have the following relationship with the unreplaced s m and q m ;

[0115]

[0116] ​

[0117] (4) Select a candidate basis function Φ from all M basis functions Φ m set; i ;

[0118] (5) Calculate θ i is a parameter set for application and is the difference related to s m and q m ;

[0119] (6) If θ i > 0 and α i < ∞, it means that the basis function Φ i is in the model and does not need to be added to the model. Only use the formula to re - estimate α i ;

[0120] (7) If θ i > 0 and α i = ∞, it means that the basis function Φ i is not in the model. Add Φ i to the model, calculate the corresponding parameters and then use the formula to re - estimate α i ;

[0121] (8) If θ i ≤ 0 and α i < ∞, delete Φ i and set α i = ∞;

[0122] (9) Estimate the variance of the stochastic noise of the biomass cogeneration system where N is the number of data and M is the number of basis functions;

[0123] (10) Recalculate the covariance matrix Σ, the mean matrix μ and s m and q m in the corresponding iteration process;

[0124] (11) If the weights accuracy related to the emission concentration and the stochastic noise accuracy converge (no longer change) or reach the maximum number of iterations, save the weight values and bias values and terminate the program; otherwise, go to step (4).

[0125] Such as Figure 3As shown in the figure, it is a flow chart of a biomass cogeneration system. The biomass cogeneration process includes technological links such as biomass storage and transportation, biomass combustion, steam-water cycle, steam power generation, steam and heat supply, flue gas treatment, and flue gas emission. Biomass fuel is usually transported by specialized transport vehicles to the front furnace bin for storage, and during use, it is conveyed to the boiler for combustion through a conveyor belt. The primary air and secondary air required for combustion are respectively fed into the furnace from the bottom and side walls of the furnace. Water-cooled walls are arranged around the furnace to absorb part of the heat generated by combustion. Most of the solid materials carried out of the furnace by the gas flow are separated and collected in the separator, and then sent back to the furnace through the return device for multiple cycle combustions. A large amount of high-temperature steam generated during the combustion process passes through the superheater, reheater, economizer, etc. After treatment, most of it is used to drive the steam turbine generator to supply power and to heat the cold water in the condenser tower, and a small part is used for steam and heat supply. The high-temperature flue gas generated by combustion enters the dust collector for dust removal, and finally is discharged into the atmosphere through the induced draft fan to the chimney.

[0126] To implement the above embodiments, a computer-readable storage medium is proposed, on which a prediction program for the pollutant emission concentration of a biomass cogeneration system is stored. When the prediction program for the pollutant emission concentration of the biomass cogeneration system is executed by a processor, the prediction method for the pollutant emission concentration of the biomass cogeneration system as described above is implemented; a computer device is also proposed, including a memory, a processor, and a computer program stored on the memory and operable on the processor. When the processor executes the program, the prediction method for the pollutant emission concentration of the biomass cogeneration system as described above is implemented; through the prediction method, medium, device, and application for the pollutant emission concentration of the biomass cogeneration system, the problems of lag, complex calculation, and high economic cost existing in the existing prediction methods for the pollutant emissions of biomass cogeneration systems are overcome, and it is easy to operate and has interpretability.

[0127] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0128] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce an implementation, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in one block or multiple blocks.

[0129] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in one block or multiple blocks.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in one block or multiple blocks.

[0131] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted to include the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0132] Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

[0133] The content described in the embodiments of this specification is only a list of implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments. The protection scope of the present invention also extends to equivalent technical means that those skilled in the art can think of based on the inventive concept of the present invention.

Claims

1. A prediction method for pollutant emission concentration of a biomass cogeneration system, characterized in that: The historical data set of the given pollutant emission concentration of the method is , , , where represents the input matrix composed of the variable data of the biomass cogeneration system represents the pollutant emission concentration of the biomass cogeneration system, n is the sample index, and there are N samples in total; establish a prediction model and an associated function system for the pollutant emission concentration of the biomass cogeneration system; The model is formula (1), (1) Among them, , represents the random noise of the biomass cogeneration system, which follows a Gaussian distribution with a mean of 0 and an inverse variance of , w represents the weight vector, , , is an integer from 1 to N; under the known input variables of the biomass cogeneration system and the observed values of pollutant concentrations, the mean value of the pollutant emission concentration of the biomass cogeneration system is expressed as Equation (2), (2) Among them, , is a basis vector, , ; , is a kernel function, not restricted by positive definite kernels; The pollutant emission concentration correlation function system includes a prior probability density function of emission concentration-related weights, a posterior probability density function of emission concentration-related weights, a log-likelihood function of emission concentration-related weight precision and random noise precision, and a fast marginal likelihood function of emission concentration-related weight precision and random noise precision; Establish a prior probability density function for the weights related to the emission concentration, and the pollutant emission concentration of the biomass cogeneration system obeys an independent distribution, then The likelihood function of is Equation (3), (3) Among them, is a matrix; the weight w is constrained to follow a prior Gaussian distribution with a mean of 0, and its prior conditional probability density function is given by Equation (4). (4) Among them, , represents the precision corresponding to the weight ; ; Based on the prior probability density function of emission-related weights, according to Bayes' theorem, the posterior probability density function of emission concentration-related weights is established as formula (5), (5) Among them, is the covariance of the posterior distribution, is the mean, , , , and diag represents a diagonal matrix; Based on the given pollutant emission concentration data x n , integrate the weight w to establish the log-likelihood function of the weight accuracy and random noise accuracy related to the emission concentration. First, establish Equation (6). (6) Among them, the matrix is a matrix, ; Taking the logarithm of formula (6), formula (7) is obtained, (7) A fast marginal likelihood function of emission concentration with respect to weight precision and random noise precision is established, and formula (7) is transformed to obtain formula (8), (8) Among them, is the sparsity factor, which is used to measure the degree of overlap between the basis function and the basis functions already existing in the model, and the basis function is an element in the basis vector . The expression of is ; is the quality factor, which is used to measure the error generated by the model after removing the basis function . The expression is ; For Regarding weight precision Take the derivative and set it to zero to obtain Equation (9). (9) Similarly, the expression of random noise precision is solved, and formula (6) is maximized by setting its derivative to zero to obtain formula (10). (10) Among them, is the m-th diagonal element of the posterior variance ; Initialize the parameters of the prediction model and the associated function system, and train the prediction model offline to obtain the parameter variables in the prediction model, so as to obtain the predicted concentration of pollutants in the biomass cogeneration system with the trained prediction model; for the trained model, given the data values of the variables of the biomass cogeneration system , the predicted value of the corresponding pollutant concentration satisfies the probability density function of Equation (11). (11) Among them, and are the values of the emission concentration related weight precision and the random noise precision obtained through the fast marginal likelihood function, respectively.

Images