CN-122021243-A - Thermal power generating unit boiler combustion effect prediction method and device based on LSTM model

CN122021243ACN 122021243 ACN122021243 ACN 122021243ACN-122021243-A

Abstract

The application provides a thermal power unit boiler combustion effect prediction method and device based on an LSTM model, wherein the method comprises the steps of obtaining operation related information of a thermal power unit boiler in a preset time with the current time as the last time; and predicting the combustion effect value of the thermal power generating unit boiler in a single time step with the current moment as a starting point by taking the time sequence characteristic as the input information of a preset LSTM model. The technical scheme of the application can improve the prediction capability of the combustion effect of the boiler of the thermal power unit, increase the control capability of the boiler of the thermal power unit, and is beneficial to supplying energy to the thermal power unit with lower cost and high efficiency.

Inventors

YE XIANG
ZHANG XUNKUI
LI JIANHUA
ZHU XIANRAN
YANG LEI
ZHOU YANAN
MA SHAOWU
LI YOU
WANG ZIMO

Assignees

中国大唐集团科学技术研究总院有限公司华北电力试验研究院
中国大唐集团科学技术研究总院有限公司

Dates

Publication Date: 20260512
Application Date: 20251223

Claims (10)

1. A thermal power generating unit boiler combustion effect prediction method based on an LSTM model is characterized by comprising the following steps of: Acquiring operation related information of a thermal power unit boiler in a preset time with the current time as the last time, wherein the operation related information is used for reflecting dynamic characteristics which are expressed by the thermal power unit boiler in the preset time and can influence the combustion effect of the thermal power unit boiler; Constructing the operation related information into time sequence features according to the acquisition time of the operation related information and a preset time step, wherein the time sequence features are used for reflecting dynamic feature distribution conditions of the thermal power generating unit boiler, which change along with time development in a plurality of preset time steps in the preset time length; And predicting a combustion effect value of the thermal power generating unit boiler in a single time step with the current moment as a starting point through the LSTM model by taking the time sequence characteristics as input information of a preset LSTM model, wherein the LSTM model is used for reflecting the association relation between operation association information of the thermal power generating unit boiler in the preset time length and the combustion effect of the thermal power generating unit boiler in the first time step after the preset time length.
2. The method of claim 1, wherein the operation-related information includes one or more of load, total coal amount, primary air amount, secondary air amount, over-fire damper opening, oxygen amount, furnace temperature, and exhaust gas temperature; The combustion effect values include boiler efficiency and nitrogen oxide emission concentration.
3. The method according to claim 2, further comprising, before the acquiring the operation-related information of the thermal power generating unit boiler for a predetermined period of time ending with the current time,: Acquiring a training sample set, wherein each group of training samples in the training sample set comprises historical operation related information of the thermal power generating unit boiler in the preset time length and historical boiler efficiency and historical nitrogen oxide emission concentration of the first historical time step of the thermal power generating unit boiler after the preset time length; For each group of training samples, according to the collection time of the historical operation associated information and the historical time step, constructing the historical operation associated information into a historical time sequence feature, wherein the historical time sequence feature comprises the historical operation associated information in each of a plurality of historical time step; And taking the historical time sequence characteristics as input information, taking the historical boiler efficiency and the historical nitrogen oxide emission concentration of the first historical time step after the historical time sequence characteristics as output information, iteratively training an initial LSTM model until the mean square error of the predicted combustion effect and the actual combustion effect output by the LSTM model is minimized, and stopping iteration.
4. A method according to claim 3, further comprising, prior to said constructing said historical running correlation information as a historical time series feature according to said time of acquisition of said historical running correlation information and said historical time step: Any historical operation associated information in the training sample set is rejected, and if the historical operation associated information is not in a preset safety range, the rejection position is determined to be a missing value; If the history operation related information is identified as a missing value, filling the missing value by adopting a linear interpolation method or an adjacent data average method based on the adjacent value of the history operation related information in a time sequence; and carrying out normalization processing on all the history operation related information after filling.
5. The method according to any one of claims 1 to 4, further comprising, before the acquiring the operation-related information of the thermal power plant boiler for a predetermined period of time ending with the current time,: Acquiring power generation cost information of the boiler of the thermal power generating unit and power supply demand information of the thermal power generating unit; Determining a target combustion effect value in an achievable section of the combustion effect value of the thermal power generating unit boiler based on the power generation cost information and the power supply demand information, so that the power generation cost information can be matched with the power supply demand information; Determining a corresponding prediction accuracy level based on the target combustion effect value; And respectively setting the prediction basis time length and the prediction time step length corresponding to the prediction precision grade as the preset time length and the time step length when the combustion effect of the thermal power generating unit boiler is predicted.
6. Thermal power generating unit boiler combustion effect prediction unit based on LSTM model, its characterized in that includes: The operation related information acquisition unit is used for acquiring operation related information of the thermal power unit boiler in a preset time with the current time as the last time, wherein the operation related information is used for reflecting dynamic characteristics which are expressed by the thermal power unit boiler in the preset time and can influence the combustion effect of the thermal power unit boiler; The time sequence feature construction unit is used for constructing the operation associated information into time sequence features according to the acquisition time of the operation associated information and a preset time step, wherein the time sequence features are used for reflecting dynamic feature distribution conditions of the thermal power generating unit boiler, which change along with time development in a plurality of preset time steps in the preset time length; The combustion effect prediction unit is used for predicting a combustion effect value of the thermal power unit boiler in a single time step with the current moment as a starting point through the LSTM model by taking the time sequence characteristics as input information of a preset LSTM model, wherein the LSTM model is used for reflecting the association relation between operation association information of the thermal power unit boiler in the preset time length and the combustion effect of the thermal power unit boiler in the first time step after the preset time length.
7. The apparatus of claim 6, wherein the operation-related information includes one or more of a load, a total coal amount, a primary air amount, a secondary air amount, a burnout door opening, an oxygen amount, a furnace temperature, and a smoke discharge temperature; The combustion effect values include boiler efficiency and nitrogen oxide emission concentration.
8. The apparatus as recited in claim 7, further comprising: The training set acquisition unit is used for acquiring a training sample set before the operation association information acquisition unit acquires the operation association information, wherein each group of training samples in the training sample set comprises historical operation association information of the thermal power unit boiler in the preset time length and historical boiler efficiency and historical nitrogen oxide emission concentration of the thermal power unit boiler in a first historical time step after the preset time length; A historical time sequence feature construction unit, configured to construct, for each set of training samples, the historical operation association information as a historical time sequence feature according to the collection time of the historical operation association information and the historical time step, where the historical time sequence feature includes respective historical operation association information in a plurality of historical time steps; And the LSTM model training unit is used for iteratively training an initial LSTM model by taking the historical time sequence characteristics as input information and taking the historical boiler efficiency and the historical nitrogen oxide emission concentration of the first historical time step after the historical time sequence characteristics as output information until the mean square error of the predicted combustion effect and the actual combustion effect output by the LSTM model is minimized, and stopping iteration.
9. A computer device comprises at least one processor, and a memory communicatively coupled to the at least one processor; Wherein the memory stores instructions executable by the at least one processor, the instructions being configured to cause the processor to perform the method of any one of the preceding claims 1 to 5.
10. A computer-readable storage medium, characterized in that computer-executable instructions are stored, which are configured to perform the method of any one of claims 1 to 5.

Description

Thermal power generating unit boiler combustion effect prediction method and device based on LSTM model Technical Field The application relates to the technical field of thermal power generation, in particular to a thermal power unit boiler combustion effect prediction method and device based on an LSTM model. Background The combustion process of the boiler of the thermal power generating unit is a complex nonlinear, multivariable and large-lag dynamic process, so to speak, the combustion effect of the boiler directly influences the economy and environmental protection of the thermal power generating unit, wherein the boiler efficiency and the emission concentration of nitrogen oxides are two important indexes of the combustion effect of the boiler. At present, modeling methods for boiler combustion processes mainly comprise mechanism modeling and data-driven modeling. Specifically, the mechanism modeling is based on the physical chemistry principle, and the process is described by establishing a differential equation and an algebraic equation, but the mechanism of the boiler combustion process is complex, the model precision is limited, and the model is difficult to adapt to different working conditions. The data driving modeling method is like a traditional neural network, a support vector machine and the like, and most of the data driving modeling methods adopt static models, so that the time sequence dynamic characteristics of the combustion process are difficult to capture, and the prediction accuracy is limited. Therefore, how to improve the prediction accuracy of the combustion effect of the boiler of the thermal power generating unit and reduce the prediction hysteresis becomes the technical problem to be solved urgently at present. Disclosure of Invention The embodiment of the application provides a thermal power unit boiler combustion effect prediction method and device based on an LSTM model, and aims to solve the technical problem that the accuracy and timeliness of thermal power unit boiler combustion effect prediction in the related art are difficult to meet actual thermal power unit boiler control requirements. In a first aspect, an embodiment of the present application provides a thermal power generating unit boiler combustion effect prediction method based on an LSTM model, including: Acquiring operation related information of a thermal power unit boiler in a preset time with the current time as the last time, wherein the operation related information is used for reflecting dynamic characteristics which are expressed by the thermal power unit boiler in the preset time and can influence the combustion effect of the thermal power unit boiler; Constructing the operation related information into time sequence features according to the acquisition time of the operation related information and a preset time step, wherein the time sequence features are used for reflecting dynamic feature distribution conditions of the thermal power generating unit boiler, which change along with time development in a plurality of preset time steps in the preset time length; And predicting a combustion effect value of the thermal power generating unit boiler in a single time step with the current moment as a starting point through the LSTM model by taking the time sequence characteristics as input information of a preset LSTM model, wherein the LSTM model is used for reflecting the association relation between operation association information of the thermal power generating unit boiler in the preset time length and the combustion effect of the thermal power generating unit boiler in the first time step after the preset time length. In one embodiment of the application, optionally, the operation related information comprises one or more of load, total coal amount, primary air amount, secondary air amount, opening degree of a burnout door, oxygen amount, hearth temperature and exhaust gas temperature; The combustion effect values include boiler efficiency and nitrogen oxide emission concentration. In one embodiment of the present application, optionally, before the acquiring the operation related information of the thermal power generating unit boiler in the predetermined time period taking the current time as the last time, the method further includes: Acquiring a training sample set, wherein each group of training samples in the training sample set comprises historical operation related information of the thermal power generating unit boiler in the preset time length and historical boiler efficiency and historical nitrogen oxide emission concentration of the first historical time step of the thermal power generating unit boiler after the preset time length; For each group of training samples, according to the collection time of the historical operation associated information and the historical time step, constructing the historical operation associated information into a historical time sequence feature, wherein t