A boiler wall temperature prediction and control method based on a deep learning model

Abstract

The present invention relates to a boiler wall temperature prediction and control method based on a deep learning model. This method establishes a deep learning model for the wall temperature of the key monitored heating surfaces of the boiler system, calculates the changes in the boiler wall temperature at future moments under different control quantities based on this model, and selects the control quantity corresponding to the optimal control result as the output of the controller to perform wall temperature intervention control. Compared with the prior art, the present invention has the advantages of simple engineering application calculation, preventing the wall temperature from exceeding the temperature, and thus improving the safe and stable operation of the unit, etc.

Description

Technical Field

[0001] The present invention relates to the field of thermal intelligent control and protection, and particularly to a boiler wall temperature prediction control method based on a deep learning model. Background Art

[0002] With the deep promotion of China's economic structure adjustment and transformation and upgrading, as well as the vigorous implementation of the power system reform, under the policy drive of energy conservation, consumption reduction, and emission reduction, smart power plants integrate technologies such as the Internet, big data, and artificial intelligence, and have become the main trend of power plant development through measures such as promoting intelligent operation management, intelligent maintenance safety, and intelligent control. The operation of smart power plants can effectively enhance the core competitiveness of power plants and promote the sustainable development of power plants.

[0003] Due to the large delay and inertia in the energy production process of the boiler system of coal-fired thermal power units, and due to the large number of complex coupling factors in the boiler combustion process, as well as the significant differences in operating characteristics under different operating conditions, there are problems such as difficult operation adjustment and poor automatic control regulation performance for objects such as the wall temperature and steam temperature of the coal-fired unit boiler system. The traditional DCS control method cannot solve this problem well. With the rapid development of intelligent control technology and data mining technology, applying deep learning technology to the modeling and prediction of complex objects in thermal power units has become an important research and application field.

[0004] Deep learning models can better apply to the application scenarios of complex high-dimensional, nonlinear, time-varying, and massive data in thermal power units because they can combine low-level features into more abstract high-level features. However, due to the high dimension of deep learning models, which are non-linear mathematical models, the deep learning model cannot be directly used as a prediction model for optimal control solution. Usually, intelligent optimization algorithms are adopted. However, due to the complex calculation of deep learning models and optimization algorithms, it cannot meet the requirements of engineering applications (Patent: A pH value control method and system for a desulfurization system based on predictive control (202210672456.7)). Currently, in the actual engineering control of complex objects in thermal power units, deep learning models mainly adopt warning or intelligent feed-forward control methods (Patent: A denitration system for thermal power units based on deep learning and an optimized control method (202011261792)), and it is not convenient to embed the deep learning model into the control solution to fully utilize the predictive control function of the deep learning model. At the same time, since the wall temperature regulation target is to prevent overheating and there is no specific operating set value, less predictive control work has been carried out for boiler wall temperature in previous thermal control, and the intervention of boiler wall temperature is more carried out in the way of manual operation adjustment. Summary of the Invention

[0005] The purpose of the present invention is to provide a boiler wall temperature prediction control method based on a deep learning model to overcome the above-mentioned defects existing in the prior art.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, there is provided a boiler wall temperature prediction and control method based on a deep learning model. The method establishes a deep learning model for predicting the wall temperature of the key monitored heating surfaces of the boiler system. When it is predicted by the model that the wall temperature may exceed the limit in the future, the method calculates the changes in the boiler wall temperature at future moments under different control quantities through the model, and selects the control quantity corresponding to the optimal control result as the output of the controller to perform wall temperature intervention control.

[0008] As a preferred technical solution, the method specifically includes the following steps:

[0009] Step S1: Use the relevant parameters of the boiler coal pulverizing system, combustion system, and air and flue gas system as characteristic input parameters, select historical operation data, and train a deep learning model for predicting the future changes in the boiler wall temperature using a recurrent neural network;

[0010] Step S2: Read the parameter values of the characteristic variables from the DCS system in real time and establish a dynamic data storage matrix queue;

[0011] Step S3: Extract the input data matrix of the current moment of the model from the dynamic data storage matrix queue according to the sampling period of the deep learning model training data and the model time compensation;

[0012] Step S4: Input the data matrix in step S3 into the deep learning model to calculate the future changes in the boiler wall temperature;

[0013] Step S5: Obtain the wall temperature over-temperature warning limit T s , and compare it with the calculation result of step S4. If the over-temperature situation occurs, perform wall temperature prediction control intervention and execute step S6. Otherwise, return to step S3;

[0014] Step S6: Determine the change amount limit [-Δp a , Δp b of the feed water flow rate p2 within a control period. When the deep learning model predicts that the wall temperature exceeds the limit, replace the actual feed water flow rate p2(k) at the current moment with the newly set feed water flow rate , and re-perform the model prediction calculation to obtain the wall temperature change value at future moments;

[0015] Step S7: Calculate the wall temperature prediction control performance index J under different i inputs;

[0016] Step S8: Select the minimum J i corresponding to As the feed water flow rate command for the next moment, wall temperature overheating intervention is performed.

[0017] As a preferred technical solution, the data sampling period in step S1 is t, the time step is d, and a total of 6 models are established to predict the changes in the boiler wall temperature in the next 30s, 60s, 90s, 120s, 150s, and 180s.

[0018] As a preferred technical solution, the dynamic data storage matrix queue in step S2 is specifically:

[0019]

[0020] where k represents the current moment, and p j (k) is the parameter of the jth characteristic variable at the kth moment, where j = 1, 2,..., n, t is the data sampling period, and d is the time step.

[0021] As a preferred technical solution, in step S2, after reading the real-time data of the characteristic variables, a shift operation is performed on the dynamic data storage matrix, and the real-time data is assigned to the parameter corresponding to the current moment.

[0022] As a preferred technical solution, the input data matrix in step S3 is specifically:

[0023]

[0024] where k represents the current moment, and p j (k) is the parameter of the jth characteristic variable at the kth moment, where j = 1, 2,..., n, t is the data sampling period, and d is the time step.

[0025] As a preferred technical solution, the future change situation of the boiler wall temperature in step S4 includes the change situations of the boiler wall temperature in the next 30s, 60s, 90s, 120s, 150s, and 180s.

[0026] As a preferred technical solution, in step S5, if any one of the 6 future wall temperature change data exceeds the overheating warning limit T s , wall temperature prediction control intervention is performed.

[0027] As a preferred technical solution, the newly set feed water flow rate is specifically calculated as follows:

[0028]

[0029] Δp i = min(-Δp a + i·Δp, Δp b ), i = 0, 1,..., m

[0030] where Δp is the feed water flow control range, i is the number of calculations, and Ceiling is the ceiling operation;

[0031] As a preferred technical solution, the wall temperature prediction control performance index J i is specifically calculated as follows:

[0032]

[0033] where ω is the calculation coefficient of the change in feed water flow rate, is the predicted value of the wall temperature at time l under the action of.

[0034] Compared with the prior art, the present invention proposes a deep learning prediction control method convenient for practical engineering applications. By establishing a deep learning model of the wall temperature of the key monitored heating surface of the boiler system, based on this model, the changes in the boiler wall temperature at future times under different control quantities are calculated respectively, and the control quantity corresponding to the optimal control result is selected as the output of the controller to perform wall temperature intervention control to prevent the wall temperature from exceeding the limit, thereby improving the safe and stable operation of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is the specific flow chart of the boiler wall temperature prediction control method based on the deep learning model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0037] As Figure 1 shown, for the prediction control of the wall temperature at the outlet of the vertical pipe on the rear wall of the boiler of a 1000MW ultra-supercritical unit as follows:

[0038] The prediction control of the wall temperature 15 at the outlet of the vertical pipe on the rear wall of the boiler of a certain 1000MW ultra-supercritical unit is as follows:

[0039] 1) Take the relevant parameters of the boiler coal pulverizing system, combustion system, and air and flue gas system as characteristic input parameters. Select historical operation data and use a recurrent neural network to train a deep learning model for predicting the future change of the boiler wall temperature. The data sampling period is 20 s, the time step is 10 steps, and a total of 6 models for predicting the changes of the boiler wall temperature in the next 30 s, 60 s, 90 s, 120 s, 150 s, and 180 s are established. The input and output characteristic parameters of the model are shown in Table 1:

[0040] Table 1

[0041] Serial number Relevant variable Unit 1 Unit power MW 2 Feed water flow t / h 3 Coal mill A-F coal feed amount and total coal amount t / h 4 Total air volume t / h 5 Primary air volume of mill A-F t / h 6 Primary air pressure kPa 7 Opening of secondary air damper of each layer of burners ％ 8 Feed water enthalpy value kJ / kg 9 Swing valve position of each corner burner ％ 10 Main steam pressure MPa 11 Intermediate point temperature ℃ 12 Furnace area soot blowing signal -

[0042] 2) Read the parameter values of the characteristic variables in Table 1 from the DCS system in real time. Considering the large difference between the model sampling period and the DCS data period, in order to drive the model in real time, establish a dynamic data storage matrix queue;

[0043]

[0044] where k represents the current moment, and p j (k)(j = 1, 2,..., n) is the parameter of the jth characteristic variable at the kth moment. After reading the real-time data of the characteristic variables, perform a shift operation on the dynamic data storage matrix and assign the real-time data to the parameter corresponding to the current moment (k moment).

[0045] 3) Extract the input data matrix of the current moment of the model from the dynamic data storage matrix queue according to the sampling period of the deep learning model training data and the model time compensation:

[0046]

[0047] 4) Input the data matrix in step 3) into the deep learning model to calculate the changes of the boiler wall temperature in the next 30 s, 60 s, 90 s, 120 s, 150 s, and 180 s respectively;

[0048] 5) Obtain the over-temperature warning limit T s of the wall temperature. If none of the 6 predicted future wall temperature change data in step 4) shows an over-temperature situation, perform predictive intervention control; if any one of the 6 predicted future wall temperature change data exceeds the over-temperature warning limit T s , perform wall temperature prediction control intervention;

[0049] 6) Determine the change amount limit [-Δp a , Δp b of the feed water flow rate p2 within a control cycle. When the deep learning model predicts that the wall temperature is over-temperature, use the newly set feed water flow rate to replace the actual feed water flow rate p2(k) at the current moment, and re-perform model prediction calculation to obtain the wall temperature change value at the future moment

[0050]

[0051] Δp i = min(-Δp a + i·Δp, Δp b ), i = 0, 1, …, m

[0052] 7) Calculate the wall temperature predictive control performance index under different inputs:

[0053]

[0054] In the formula, ω is the calculation coefficient of the change in feed water flow rate, taken as 0.01.

[0055] 8) Select the minimum J i corresponding as the feed water flow rate command for the next moment to perform wall temperature over-temperature intervention.

[0056] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A boiler wall temperature prediction and control method based on a deep learning model, characterized in that, This method establishes a deep learning model for the wall temperature of the key monitored heating surfaces in the boiler system. When it is predicted by the model that the wall temperature may exceed the normal range in the future, the method calculates the changes in the boiler wall temperature at future moments under different control quantities through this model, and selects the control quantity corresponding to the optimal control result as the output of the controller to perform wall temperature intervention control; This method specifically includes the following steps: Step S1: Use the relevant parameters of the boiler coal pulverizing system, combustion system, and air and flue gas system as characteristic input parameters, select historical operation data, and train a deep learning model for predicting the future changes in the boiler wall temperature using a recurrent neural network; Step S2: Read the parameter values of the characteristic variables from the DCS system in real time and establish a dynamic data storage matrix queue; Step S3: Extract the input data matrix at the current moment of the model from the dynamic data storage matrix queue according to the sampling period of the deep learning model training data and the model time compensation; Step S4: Input the data matrix in Step S3 into the deep learning model to calculate the future changes in the boiler wall temperature; Step S5, obtain the wall temperature over-temperature warning limit T s , and compare it with the calculation result of step S4. If over-temperature occurs, perform wall temperature prediction control intervention and execute step S6; otherwise, return to step S3; Step S6, determine the change limit of the water flow rate p2 within a control cycle [-Δp a ,Δp b ], when the deep learning model predicts that the wall temperature is over-temperature, the newly set water flow rate Instead of the actual water supply flow rate p2(k) at the current moment, re-calculate the model prediction to obtain the wall temperature change value at the future moment; Step S7, calculate different wall temperature predictive control performance index J under the i input; Step S8, select the minimum J i corresponding as the feed water flow rate command for the next moment to perform wall temperature overheating intervention; The newly set feed water flow rate The specific calculation is as follows: Δp i = min(-Δp a + i·Δp, Δp b ), where i = 0, 1, …, m where Δp is the feed water flow control interval, i is the number of calculations, and Ceiling is the ceiling operation; The wall temperature prediction control performance index J i The specific calculation is as follows: where ω is the calculation coefficient of the change in feed water flow rate, is the predicted value of the wall temperature at time l under the action of.

2. The boiler wall temperature prediction and control method based on a deep learning model according to claim 1, wherein, In Step S1, the data sampling period is t, the time step is d, and a total of 6 models for predicting the changes in the boiler wall temperature in the future 30s, 60s, 90s, 120s, 150s, and 180s are established.

3. A boiler wall temperature prediction and control method based on a deep learning model according to claim 1, characterized in that The dynamic data storage matrix queue in Step S2 is specifically: where k represents the current moment, p j (k) is the parameter of the j-th characteristic variable at the k-th moment, where j = 1, 2, …, n, t is the data sampling period, and d is the time step.

4. The boiler wall temperature prediction and control method based on a deep learning model according to claim 3, characterized in that In Step S2, after reading the real-time data of the characteristic variables, perform a shift operation on the dynamic data storage matrix and assign the real-time data to the parameter corresponding to the current moment.

5. A boiler wall temperature prediction and control method based on a deep learning model according to claim 1, characterized in that, The input data matrix in Step S3 is specifically: where k represents the current moment, and p j (k) is the parameter of the j-th characteristic variable at the k-th moment, where j = 1, 2, …, n, t is the data sampling period, and d is the time step.

6. The boiler wall temperature prediction and control method based on a deep learning model according to claim 1, wherein, The future changes in the boiler wall temperature in Step S4 include the changes in the boiler wall temperature in the future 30s, 60s, 90s, 120s, 150s, and 180s.

7. A boiler wall temperature prediction and control method based on a deep learning model according to claim 6, characterized in that, In the step S5, if any one of the six future wall temperature change data points exceeds the overtemperature warning limit T s , wall temperature prediction control intervention is carried out.

Images