Prediction method for water taking temperature of circulating water of coastal power plant

Abstract

The invention relates to the technical field of machine learning, in particular to a method for predicting the water taking temperature of circulating water of a coastal power plant. Specifically, the invention provides a coastal power plant circulating water taking temperature time sequence prediction method for modeling by combining monitoring data of surrounding stations and real-time data of a power plant. At the initial stage of model online, the water taking temperature prediction error can be obviously reduced through the model trained by station data around the power plant; along with gradual accumulation of data such as the real water taking temperature of the power plant, an original model is finely adjusted, and the model prediction error is further reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method for predicting the temperature of circulating water intake of a coastal power plant. Background Art

[0002] For coastal nuclear power plants or thermal power plants, circulating cooling water is directly taken from seawater, heated in the condenser and discharged into the ocean. For these power plants, the intake water temperature directly affects the heat exchange performance of the power plant condenser and the power consumption of the circulating water pump under the corresponding working conditions. If the changing trend of the circulating water intake temperature can be predicted, the power plant can adjust the speed of the circulating water pump in real time according to the temperature change to ensure the operation of the entire plant.

[0003] Existing temperature prediction methods mainly focus on the prediction of seawater temperature or the prediction of other industrial process temperature parameters. There is currently a lack of direct prediction methods for power plant water intake temperature. At the same time, seawater temperature prediction is modeled using a time series neural network model, which requires the accumulation of a large amount of historical data. For special working conditions, some patents use a small sample-based transfer learning algorithm to fine-tune the original model. This method still uses the same measurement point data at different stages of model training, which can easily lead to a decrease in the accuracy of the model in the early stage of going online.

[0004] Based on this, there is an urgent need for a prediction method for the circulating water intake temperature of coastal power plants to solve the above technical problems. Summary of the invention

[0005] The invention provides a method for predicting the temperature of circulating water intake of a coastal power plant, which can reduce the model prediction error.

[0006] An embodiment of the present invention provides a method for predicting the temperature of circulating water intake of a coastal power plant, comprising:

[0007] Collect monitoring data from stations in the sea area near the power plant and water intake temperature data of the power plant. The data measurement points include parameters such as longitude and latitude, air temperature, and seawater temperature. The data sampling frequency needs to be consistent, or the sampling interval of the water intake temperature data of the power plant is lower than the sampling interval of the station monitoring data.

[0008] When the power plant is in shutdown condition or the temperature sensor measurement point fails, the missing values ​​of the temperature data need to be filled for the temperature time series prediction data. At the same time, in order to deal with the problem of bad points in the data transmission process, the abnormal temperature points are determined by thresholds and eliminated;

[0009] The temperature series data monitored by the stations in the historical -t-0 period, including at least the air temperature and the seawater temperature, are used as the model input, and the seawater temperature series data monitored by the stations in the 0-t period are used as the output to train a neural network model based on time series, including but not limited to LSTM model, GRU model, and Transformer model;

[0010] When the temperature measurement point data within a period of at least 2t is accumulated, the data is organized into a single sample, input into the trained model, and the prediction error is calculated. The error evaluation method can be the root mean square error, etc., which is not limited here;

[0011] As the amount of power plant data continues to accumulate, the model error is constantly evaluated. When the model error cannot meet the set requirements, the existing power plant temperature data is used to organize it into data samples and fine-tune the original trained model;

[0012] After the model is launched, the trained or fine-tuned model is used to make predictions based on the real-time temperature measurement point data of the power plant, and the prediction output is used by other control optimization models of the power plant.

[0013] It can be seen from the above scheme that the method for predicting the circulating water intake temperature of coastal power plants provided by the present invention combines the monitoring data of surrounding stations and the real-time data of power plants to model the time series prediction method for circulating water intake temperature of coastal power plants. In the early stage of the model online, the model trained by the data of stations around the power plant can significantly reduce the prediction error of the intake temperature; as the amount of data such as the actual intake temperature of the power plant gradually accumulates, the original model is fine-tuned to further reduce the model prediction error. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A schematic flow chart of a method for predicting the temperature of circulating water intake of a coastal power plant provided by an embodiment of the present invention;

[0016] Figure 2 A data graph of the historical 5-year monitoring data of a station near a power plant provided in an embodiment of the present invention;

[0017] Figure 3 A time series trend diagram of seawater temperature in different years provided by an embodiment of the present invention;

[0018] Figure 4LSTM model framework diagram provided for an embodiment of the present invention;

[0019] Figure 5 A schematic diagram of a model loss reduction curve provided by an embodiment of the present invention;

[0020] Figure 6 This is a temperature variation trend diagram of the outlet main pipe of the open-circulation cooling water pump provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] See also Figure 1 The embodiment of the present invention provides a method for predicting the temperature of circulating water intake of a coastal power plant, the method comprising:

[0023] Data collection: Collect monitoring data from stations in the sea area near the power plant and water intake temperature data of the power plant. The data measurement points include parameters such as longitude and latitude, air temperature, and seawater temperature. The data sampling frequency needs to be consistent, or the sampling interval of the water intake temperature data of the power plant should be lower than the sampling interval of the station monitoring data.

[0024] Data preprocessing: When the power plant is in shutdown condition or the temperature sensor measurement point fails, the missing values ​​of the temperature data need to be filled for the temperature time series prediction data. At the same time, in order to deal with the problem of bad points during data transmission, the abnormal temperature points are determined by thresholds and eliminated;

[0025] Model training: Use the temperature series data monitored by the station in the historical -t-0 period, including at least air temperature and seawater temperature, as the model input, and the seawater temperature series data monitored by the station in the 0-t period as the output, and train the neural network model based on time series. The model includes but is not limited to LSTM model, GRU model, and Transformer model;

[0026] Model evaluation: When the temperature measurement point data within a period of at least 2t is accumulated, the data is organized into a single sample, input into the trained model, and the prediction error is calculated. The error evaluation method can be the root mean square error, etc., which is not limited here;

[0027] Model fine-tuning: As the amount of power plant data continues to accumulate, the model error is constantly evaluated. When the model error cannot meet the set requirements, the existing power plant temperature data is used to organize it into data samples and fine-tune the original trained model;

[0028] Model prediction: After the model is launched, the trained or fine-tuned model is used to make predictions based on the real-time temperature measurement point data of the power plant, and the prediction output is used by other control optimization models of the power plant.

[0029] In this embodiment, a method for predicting the time series of circulating water intake temperature of coastal power plants is modeled by combining the monitoring data of surrounding stations and the real-time data of power plants. In the early stage of the model, the model trained by the data of stations around the power plant can significantly reduce the prediction error of the intake temperature; as the amount of data such as the actual intake temperature of the power plant gradually accumulates, the original model is fine-tuned to further reduce the model prediction error.

[0030] The following is an introduction to a specific technical solution implementation plan using a newly built coastal power plant as an example.

[0031] Data collection: Obtain the historical monitoring data of a station near the power plant for 5 years. Figure 2 As shown;

[0032] Data preprocessing: The abnormal data with air temperature of 999.9 is removed, and the missing values ​​of seawater temperature are filled by interpolation method. The trend of seawater temperature after processing is as follows: Figure 3 As shown;

[0033] Model training: The time series prediction neural network model is used to predict the seawater temperature. The independent variables of the model are the seawater temperature and air temperature every hour of the day, and the dependent variable is the seawater temperature every hour in the next 24 hours. The model framework is as follows: Figure 4 As shown;

[0034] Model evaluation: Model error reduction trend and verification results are as follows Figure 5 As shown, the RMS error is 0.55°C and the R2_score is 99.4%;

[0035] Model fine-tuning: Collect the temperature measurement point data of the power plant for one year, such as Figure 6 As shown in the figure, the model was fine-tuned after data preprocessing. On the same power plant verification data set, the root mean square error of the prediction was 0.32℃, and the model error was significantly reduced.

[0036] Based on the above analysis, this patent proposes a time series prediction method for circulating water intake temperature of coastal power plants that combines monitoring data from surrounding stations and real-time data from power plants for modeling. In the early stage of the model launch, the model trained with data from stations around the power plant can significantly reduce the prediction error of the intake temperature; as the amount of data such as the actual intake temperature of the power plant gradually accumulates, the original model is fine-tuned to further reduce the model prediction error.

[0037] Finally, it should be noted that the above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for predicting the temperature of circulating water intake of a coastal power plant, characterized in that: include: Collect monitoring data from stations in the sea area near the power plant and water intake temperature data of the power plant. The data measurement points include parameters such as longitude and latitude, air temperature, and seawater temperature; The data sampling frequency needs to be consistent, or the sampling interval of the power plant water intake temperature data needs to be lower than the sampling interval of the station monitoring data; When the power plant is in shutdown condition or the temperature sensor measurement point fails, the missing values ​​of the temperature data need to be filled for the temperature time series prediction data. At the same time, in order to deal with the problem of bad points in the data transmission process, the abnormal temperature points are determined by thresholds and eliminated; The temperature series data monitored by the stations in the historical -t-0 period, including at least the air temperature and the seawater temperature, are used as the model input, and the seawater temperature series data monitored by the stations in the 0-t period are used as the output to train a neural network model based on time series, including but not limited to LSTM model, GRU model, and Transformer model; When the temperature measurement point data within a period of at least 2t is accumulated, the data is organized into a single sample, input into the trained model, and the prediction error is calculated. The error evaluation method can be the root mean square error, etc., which is not limited here; As the amount of power plant data continues to accumulate, the model error is constantly evaluated. When the model error cannot meet the set requirements, the existing power plant temperature data is used to organize it into data samples and fine-tune the original trained model; After the model is launched, the trained or fine-tuned model is used to make predictions based on the real-time temperature measurement point data of the power plant, and the prediction output is used by other control optimization models of the power plant.

Images