A method for predicting fluid outlet temperature of a boiler heat exchanger

Abstract

The application provides a boiler heat exchanger fluid outlet temperature prediction method, which comprises the following steps: step one, establishing a mixed expert model with heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree as inputs and heat exchanger cold fluid outlet temperature prediction as output; step two, optimizing the mixed expert model through a database construction learning strategy to form a parameter-adapted mixed expert model; and step three, inputting the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree data under the current state, and automatically outputting the cold fluid outlet temperature prediction result under the current heat exchanger working state through the mixed expert model.

Description

TECHNICAL FIELD

[0001] The present application relates to a boiler heat exchanger fluid outlet temperature prediction method, in particular to the technical field of intelligent prediction of heat exchanger cold fluid outlet temperature. BACKGROUND

[0002] The heat exchanger is an important equipment of the boiler heat energy system, and its heat exchange efficiency and stability directly determine the industrial production efficiency, operation safety and health status of the boiler system. However, the heat exchange mechanism of the heat exchanger is complex, and has the characteristics of nonlinearity, delay and multiple disturbances, so it is generally difficult to ideal temperature control adjustment for the required control object. On the one hand, due to the complex heat exchange mechanism and nonlinearity of the heat exchanger, it is difficult to establish a very accurate mathematical model. At present, most heat exchanger models cannot accurately reflect the internal heat exchange dynamic process under actual operating conditions. In addition, due to the fast and slow flow of liquid in the heat exchange process, the temperature change has a certain pure lag. When the control effect is applied, it often takes a period of time to reflect the control effect, so that the system cannot timely find the interference of the outside world, and the temperature regulation is easy to appear the phenomenon of excessive overshoot or oscillation, which affects the stability of the production process and the service life of the equipment.

[0003] Due to the difficulty of real-time and accurate description of the energy conversion process of the heat exchanger by the existing mechanism model, there is no reliable boiler heat exchanger fluid outlet temperature prediction method based on the mechanism model. Some research attempts to explore the data-driven temperature prediction method, such as a multi-model intelligent combination algorithm for establishing a data-driven model of the furnace temperature in the boiler incineration process, and a fuzzy modeling method for variable frequency control of a compression refrigeration unit. The heat exchanger cold fluid outlet temperature control of the boiler is still blank, especially for the data-driven model taking the heat exchanger hot fluid flow data, the heat exchanger cold fluid valve opening data and the heat exchanger hot fluid temperature data as input variables, and taking the cold fluid outlet temperature prediction result as output. SUMMARY

[0004] The present application provides a boiler heat exchanger fluid outlet temperature prediction method to solve the above problems in the prior art.

[0005] Technical scheme: A boiler heat exchanger fluid outlet temperature prediction method, comprising:

[0006] Step one, establishing a mixed expert model taking the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree as input, and the heat exchanger cold fluid outlet temperature prediction as output;

[0007] Step two, optimizing the mixed expert model by a database construction learning strategy to form a parameter-adapted mixed expert model;

[0008] Step three, input the current state of the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening data, automatically output the current heat exchanger working state of the cold fluid outlet temperature prediction results through the mixed expert model.

[0009] In further embodiments, the step one, the mixed expert model takes the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening multi-modal data as model input, obtains multi-modal data features through normalization processing and two-layer convolution layer calculation, according to the time-varying characteristics of different data, the two-layer convolution kernel size used for heat exchanger hot fluid temperature data feature extraction is 1x9 and 1x5,

[0010] The two-layer convolution kernel size used for heat exchanger hot fluid flow data feature extraction is 1x5 and 1x3,

[0011] The two-layer convolution kernel size used for heat exchanger cold fluid valve opening feature extraction is 1x5 and 1x3, after normalization and two-layer convolution, data features are obtained to represent heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening parameters.

[0012] In further embodiments, the first expert model in the mixed expert model is a 9-layer long short-time neural network regression prediction model, the second expert model is a 7-layer long short-time neural network regression prediction model, and the third expert model is a 5-layer long short-time neural network regression prediction model. Then, the prediction results of the mixed expert model are adaptively fused to obtain the prediction results of the heat exchanger cold fluid outlet temperature, and the outlet temperature of the cold fluid under the current state is predicted.

[0013] In further embodiments, the step two, first collect the working condition data of the heat exchanger for a certain time period, collect the heat exchanger hot fluid flow data, heat exchanger cold fluid valve opening data, heat exchanger hot fluid temperature data and cold fluid outlet temperature for a certain time period, construct a database, input the trained database into the mixed expert model for model optimization;

[0014] The mixed expert model includes threshold model optimization and expert model optimization, the mixed expert model is optimized through database construction learning strategy, a parameter adaptive cold fluid outlet temperature prediction mixed expert model is formed, then the loss function of the ith expert model is calculated, and the loss function is obtained, the expression is as follows:

[0015]

[0016] Wherein, E i represents the loss function, g iis the i-th threshold calculation unit, T represents the threshold value, γ is the true value of the outlet temperature of the cold fluid in the training set, O i is the output of the i-th hybrid expert model;

[0017] The threshold model in the i-th expert model is optimized, and the optimization process is represented as:

[0018]

[0019] wherein, is the probability of the i-th expert model predicting the true value of the outlet temperature of the cold fluid, η is the learning rate, is the output of the i-th expert model hidden layer, for the first expert model is the output of the fifth hidden convolutional layer, for the second expert model is the output of the fourth hidden convolutional layer, for the third expert model is the output of the second hidden convolutional layer.

[0020] In further embodiments, if the optimal value of the i-th threshold The branch corresponding to the i-th expert model is discarded to form a new hybrid expert model, and the new hybrid expert model is retrained until the network converges and the optimal value of each threshold function is greater than the threshold value T.

[0021] In further embodiments, the outlet temperature of the cold fluid of the heat exchanger at the current time is related to the working state of the heat exchanger at n consecutive time points, and then the data of n consecutive time points is selected to form a time sequence as input:

[0022] First, X t = {x t …x t-n} is the heat fluid temperature data of the heat exchanger at n consecutive time points at time t, Y t = {y t …y t-n} is the heat fluid flow data of the heat exchanger at n consecutive time points at time t,

[0023] Z t = {z t …z t-n} is the cold fluid valve opening data of the heat exchanger at n consecutive time points at time t;

[0024] Heat exchanger heat fluid temperature data features: X′ t = f conv (X t )

[0025] Heat exchanger heat fluid flow data features: Y t ′ = f conv (Yt )

[0026] Heat exchanger cold fluid valve opening data characteristics: Z′ t =f conv (Z t )

[0027] Among them, f conv () is the calculation of two convolutional layers;

[0028] The data features of temperature, flow rate and valve opening are spliced ​​and fused to form multimodal fusion features: the expression is as follows:

[0029] I′ t ={X′ t ,Y t ′,Z′ t}

[0030] Among them, I′ t Represents the multimodal fusion feature. The output of the hybrid expert model is the outlet temperature O of the cold fluid of the heat exchanger at time t. t .

[0031] In a further embodiment, the multimodal fusion features are predicted by a hybrid expert model to form the prediction result of the i-th expert model at time t, which is expressed as follows:

[0032]

[0033] Among them, g i () is the i-th threshold calculation unit, f i () is the calculation unit of the i-th expert model, when f i When () is the 1st, 2nd and 3rd expert model calculation unit, f1() is a 9-layer long-short time neural network regression prediction model, f2() is a 7-layer long-short time neural network regression prediction model, and f3() is a 5-layer long-short time neural network regression prediction model.

[0034] In a further embodiment, the hot fluid temperature, hot fluid flow rate and cold fluid valve opening parameters of the heat exchanger under the actual state are input, and the cold fluid outlet temperature prediction result under the current state is automatically output through the prediction of the cold fluid outlet temperature, and the weighted fusion value of the expert model prediction results is calculated as the cold fluid outlet temperature prediction result under the current state;

[0035]

[0036] in, is the optimal value of the threshold function obtained through training, O i,t is the cold fluid outlet temperature prediction result obtained by the i-th expert model based on the input at time t.

[0037] Beneficial effects: the present application proposes a boiler heat exchanger fluid outlet temperature prediction method based on a hybrid expert model, which can fill the existing cold fluid outlet temperature prediction; by establishing a heat exchanger cold fluid outlet temperature prediction with heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree multi-modal data source as characteristic variables, the database construction learning strategy optimizes the hybrid expert model, forms a parameter adaptive cold fluid outlet temperature prediction hybrid expert model, through multi-modal feature extraction and fusion, hybrid expert model prediction and threshold weighted fusion, the intelligent prediction result of cold fluid outlet temperature is calculated; the heat exchanger cold fluid outlet temperature intelligent prediction model, the prediction model takes the multi-modal data of heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree as model input, extracts multi-modal data features through normalization processing and two convolution layers, and obtains fusion features for cold fluid outlet temperature prediction through multi-modal feature fusion; the fusion features are input into the hybrid expert model, and the adaptive threshold module is used to optimize the adaptive fusion of the prediction results of the hybrid expert model, and the outlet temperature of the cold fluid under the current heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree state is adaptively predicted. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is the flowchart of the present application.

[0039] Figure 2 is the heat exchanger cold fluid outlet temperature intelligent prediction diagram of the present application. DETAILED DESCRIPTION

[0040] The applicant believes that, taking the heat exchanger cold fluid outlet temperature prediction as an example: based on the operation mechanism of the heat exchanger, the cold fluid outlet temperature is affected by the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree, among which the heat exchanger hot fluid temperature and hot fluid flow fluctuate with the change of boiler load, and the cold fluid valve opening degree is an important boiler control element for adjusting the heat process, with obvious fluctuation frequency. At present, the data modeling for affecting the cold fluid outlet temperature is still insufficient, and the data-driven cold fluid outlet temperature prediction accuracy is insufficient. The present application innovatively constructs a mathematical model based on a hybrid expert model for three modal data of heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening degree, and predicts the current heat exchanger cold fluid outlet temperature according to historical data.

[0041] The present application will be further specifically described below by examples and in combination with the drawings.

[0042] In the present application, a boiler heat exchanger fluid outlet temperature prediction method is proposed, which comprises:

[0043] Step one, a mixed expert model is established, which takes the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening as input, and the heat exchanger cold fluid outlet temperature prediction as output; the mixed expert model takes the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening multi-modal data as model input, obtains multi-modal data features through normalization processing and two-layer convolution layer calculation; according to the time-varying characteristics of different data, the two-layer convolution kernel sizes used for heat exchanger hot fluid temperature data feature extraction are 1x9 and 1x5,

[0044] The two-layer convolution kernel sizes used for heat exchanger hot fluid flow data feature extraction are 1x5 and 1x3,

[0045] The two-layer convolution kernel sizes used for heat exchanger cold fluid valve opening feature extraction are 1x5 and 1x3, and the data features used to represent the heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening parameters are obtained after normalization and two-layer convolution.

[0046] The first expert model in the mixed expert model is a 9-layer long short-time neural network regression prediction model, the second expert model is a 7-layer long short-time neural network regression prediction model, and the third expert model is a 5-layer long short-time neural network regression prediction model. Then, the prediction results of the mixed expert model are adaptively fused to obtain the prediction results of the heat exchanger cold fluid outlet temperature, and the outlet temperature of the cold fluid under the current state is predicted.

[0047] Step two, the mixed expert model is optimized by database construction learning strategy to form a parameter-adapted mixed expert model; first, the working condition data of the heat exchanger is collected for a certain period of time, and the working condition data (heat exchanger hot fluid flow data, heat exchanger cold fluid valve opening data, heat exchanger hot fluid temperature data, cold fluid outlet temperature) of the heat exchanger in a certain period of time (the rated value is 90 days) is collected. Database is constructed, and the trained database is input into the mixed expert model for model optimization;

[0048] The mixed expert model includes threshold model optimization and expert model optimization, the mixed expert model is optimized by database construction learning strategy to form a parameter-adapted cold fluid outlet temperature prediction mixed expert model,

[0049] Then the loss function of the ith expert model is calculated, and the loss function is obtained, which is expressed as follows:

[0050]

[0051] Where, E i is the loss function, g i () is the ith threshold calculation unit, T represents the threshold value, γ is the true value of the cold fluid outlet temperature in the training set, and Oi Output of the ith mixed expert model; E i is minimized, the optimal expert model is obtained;

[0052] The threshold model in the ith expert model is optimized, and the optimization process is represented as:

[0053]

[0054] wherein, is the probability of the ith expert model predicting the true value of the cold fluid outlet temperature, η is the learning rate, is the output of the hidden layer of the ith expert model, for the first expert model is the output of the fifth hidden convolutional layer, for the second expert model is the output of the fourth hidden convolutional layer, for the third expert model is the output of the second hidden convolutional layer.

[0055] If the optimal value of the ith threshold is The branch corresponding to the ith expert model is discarded to form a new mixed expert model, and the new mixed expert model is retrained until the network converges and the optimal value of each threshold function is greater than the threshold T.

[0056] Step three, input the heat fluid temperature, heat fluid flow rate and cold fluid valve opening degree data of the heat exchanger under the current state, and automatically output the cold fluid outlet temperature prediction result of the heat exchanger under the current working state through the mixed expert model; the cold fluid outlet temperature of the heat exchanger at the current time is related to the working state of the heat exchanger at n consecutive time points, and then the data of n consecutive time points are selected to form a time sequence as input:

[0057] First, X t ={x t …x t-n} is the heat fluid temperature data of the heat exchanger at n consecutive time points at t time, Y t ={y t …y t-n} is the heat fluid flow rate data of the heat exchanger at n consecutive time points at t time,

[0058] Z t ={z t …z t-n} is the cold fluid valve opening degree data of the heat exchanger at n consecutive time points at t time;

[0059] Heat fluid temperature data characteristics: X′ t =f conv (X t )

[0060] Heat exchanger hot fluid flow data feature: Y t ′=f conv t

[0061] Heat exchanger cold fluid valve opening data feature: Z′ t =f conv t

[0062] Wherein, f conv () is the calculation of two convolution layers;

[0063] The data features of temperature, flow and valve opening are spliced and fused to form a multi-modal fusion feature: the expression is as follows:

[0064] I′ t ={X′ t ,Y t ′,Z′ t}

[0065] Wherein, I′ t represents a multi-modal fusion feature, and the output of the mixed expert model is the outlet temperature O t of the heat exchanger cold fluid at time t.

[0066] The multi-modal fusion feature is predicted by the mixed expert model to form the prediction result of the i-th expert model at time t, and the expression is as follows:

[0067]

[0068] Wherein, g i () is the i-th threshold calculation unit, f i () is the i-th expert model calculation unit, and considering the complementarity of the expert model, when f i () is the first, second and third expert model calculation unit, then f1() is a 9-layer long short-time neural network regression prediction model, f2() is a 7-layer long short-time neural network regression prediction model, and f3() is a 5-layer long short-time neural network regression prediction model.

[0069] The heat exchanger hot fluid temperature, hot fluid flow and cold fluid valve opening parameters under the input real state are input, the cold fluid outlet temperature is predicted, the current state cold fluid outlet temperature prediction result is automatically output, and the weighted fusion value of the expert model prediction result is calculated as the current state cold fluid outlet temperature prediction result;

[0070]

[0071] Wherein, is the optimal value of the threshold function obtained by training, and O i,t ​​​​The cold fluid outlet temperature prediction result obtained by the i-th expert model according to the input at time t.

[0072] As mentioned above, although the application has been represented and described with reference to particular preferred embodiments, it is not to be construed as being limited thereto, and various changes in form and details can be made thereto without departing from the spirit and scope of the application as defined in the appended claims.

Claims

1. A method for predicting the outlet temperature of a fluid in a boiler heat exchanger, characterized in that: The application relates to a heat exchanger cold fluid outlet temperature prediction method based on a hybrid expert model. Step one: a mixed expert model is established, which takes heat fluid temperature, heat fluid flow and cold fluid valve opening degree as inputs and takes heat exchanger cold fluid outlet temperature prediction as output; Step two: the mixed expert model is optimized through a database construction learning strategy to form a parameter-adaptive mixed expert model; Step three: heat exchanger heat fluid temperature, heat fluid flow and cold fluid valve opening degree data under the current state are input, and the mixed expert model automatically outputs the cold fluid outlet temperature prediction result under the current heat exchanger working state; The first expert model in the mixed expert model is a 9-layer long short-time neural network regression prediction model, the second expert model is a 7-layer long short-time neural network regression prediction model, and the third expert model is a 5-layer long short-time neural network regression prediction model; then the mixed expert model prediction result is adaptively fused to obtain the heat exchanger cold fluid outlet temperature prediction result, and the cold fluid outlet temperature under the current state is predicted; The mixed expert model includes threshold model optimization and expert model optimization, the mixed expert model is optimized through a database construction learning strategy to form a parameter-adaptive cold fluid outlet temperature prediction mixed expert model, Then the loss function of the i-th expert model is calculated, and the loss function is obtained, and the expression is as follows: ， wherein, represents a loss function, is an i-th threshold calculation unit, T represents a threshold value, is a true value of the outlet temperature of the cold fluid in the training set, is an output of the i-th mixed expert model; The threshold model in the i-th expert model is optimized, and the optimization process is represented as: ， wherein, is the probability of the real value of the outlet temperature of the cold fluid predicted by the i-th expert model, is the learning rate, is the output of the i-th expert model hidden layer, for the 1st expert model is the output of the 5th hidden convolutional layer, for the 2nd expert model is the output of the 4th hidden convolutional layer, for the 3rd expert model is the output of the 2nd hidden convolutional layer.

2. A method of predicting fluid outlet temperature of a boiler heat exchanger as claimed in claim 1, wherein, In step one, the mixed expert model takes heat exchanger heat fluid temperature, heat fluid flow and cold fluid valve opening degree multi-modal data as model input, obtains multi-modal data features through normalization processing and two-layer convolution layer calculation, extracts heat exchanger heat fluid temperature data features by adopting two-layer convolution kernel sizes of 1*9 and 1*5, extracts heat exchanger heat fluid flow data features by adopting two-layer convolution kernel sizes of 1*5 and 1*3, extracts heat exchanger cold fluid valve opening degree features by adopting two-layer convolution kernel sizes of 1*5 and 1*3, and obtains data features for representing heat exchanger heat fluid temperature, heat fluid flow and cold fluid valve opening degree parameters after normalization and two-layer convolution.

3. A method of predicting fluid outlet temperature of a boiler heat exchanger as claimed in claim 1, wherein, In step two, the working condition data of the heat exchanger are collected for a certain time period, the heat exchanger heat fluid flow data, the heat exchanger cold fluid valve opening degree data, the heat exchanger heat fluid temperature data and the cold fluid outlet temperature are collected for a certain time period, a database is constructed, and the trained database is input into the mixed expert model for model optimization.

4. The method of claim 1, wherein, If the optimal value of the i-th threshold , discard the branch corresponding to the i-th expert model, form a new hybrid expert model, retrain the new hybrid expert model, until the network converges and the optimal value of each threshold function is greater than the threshold .

5. A method of predicting fluid outlet temperature of a boiler heat exchanger as claimed in claim 1, wherein, The cold fluid outlet temperature of the heat exchanger at the current time is related to the working states of the heat exchanger at n continuous time points, and then data of the n continuous time points are selected to form a time sequence as input: First is the heat fluid temperature data of the heat exchanger at n continuous time points at t, is the heat fluid flow data of the heat exchanger at n continuous time points at t, t is the time; n is the number of consecutive time points; and is the cold fluid valve opening data at the time point t. Heat exchanger hot fluid temperature data features: , Heat exchanger hot fluid flow data features: , Heat exchanger cold fluid valve opening data features: , wherein, is the computation of two convolutional layers; The data features of the temperature, flow and valve opening degree are spliced and fused to form multi-modal fusion features, and the expression is as follows: ， wherein, represents a multimodal fusion feature, and the output of the mixed expert model is the outlet temperature of the cold fluid of the heat exchanger at time t .

6. A method of predicting fluid outlet temperature of a boiler heat exchanger as claimed in claim 5, wherein, The multi-modal fusion features are predicted through the mixed expert model to form the prediction result of the i-th expert model at t time, and the expression is as follows: ， in, is the i-th threshold calculation unit, is the i-th expert model calculation unit, when When it is the 1st, 2nd, and 3rd expert model calculation unit, It is a 9-layer long-short time neural network regression prediction model. It is a 7-layer long-short time neural network regression prediction model. It is a 5-layer long-short time neural network regression prediction model.

7. The method of claim 1, wherein, Input the heat fluid temperature, heat fluid flow and cold fluid valve opening degree parameters in the real state, automatically output the cold fluid outlet temperature prediction result in the current state through the prediction of the cold fluid outlet temperature, and calculate the weighted fusion value of the expert model prediction result as the cold fluid outlet temperature prediction result in the current state; ， wherein, is the optimal value of the threshold function obtained by training, is the predicted outlet temperature of the cold fluid obtained by the ith expert model according to the input at time t.

Images