Deep learning-based short-term power load prediction method and apparatus, and processing device

Abstract

The invention provides a short-term power load prediction method and device based on deep learning and processing equipment, which are used for introducing meteorological data to participate in prediction processing under the condition of predicting a short-term power load based on a deep learning model and also providing a specific meteorological factor configuration scheme. The configured short-term power load prediction model can realize a high-precision prediction effect on the target equipment cluster, and can provide high-quality data support for the target equipment cluster and the management work of power supply.

Claims

1. A short-term power load forecasting method based on deep learning, characterized in that: The method comprises: Obtaining power load data of the target device cluster in a past time period, wherein it is assumed that the power usage demand of different devices involved in the target device cluster remains stable in the past time period, and the past time period refers to a time period consisting of historical time points with a preset time span from the current time point forward and the current time point; Acquire meteorological data for the past time period, wherein the meteorological data includes high-altitude layer data and ground-layer data, the high-altitude layer data includes a first air temperature, a potential height, a relative humidity, a u component of a first wind, and a v component of the first wind, and the ground-layer data includes sea level pressure, a second air temperature, a dew point temperature, a u component of a second wind, a v component of a second wind, and solar shortwave radiation; The power load data and the meteorological data are used as inputs into a preset short-term power load forecasting model, wherein the short-term power load forecasting model is a deep learning model, and the short-term power load forecasting model is trained by sample power load data and sample meteorological data; The short-term power load forecasting result output by the short-term power load forecasting model is extracted.

2. The method according to claim 1, characterized in that The first temperature is the temperature of 1000hPa, 925hPa, 850hPa, 700hPa, 500hPa and 200hPa; The potential heights are the potential heights of 1000 hPa, 925 hPa, 850 hPa, 700 hPa, 500 hPa and 200 hPa; The relative humidity is 1000hPa, 925hPa, 850hPa, 700hPa, 500hPa and 200hPa; The u component of the first wind is the u component of wind at 1000 hPa, 925 hPa, 850 hPa, 700 hPa, 500 hPa and 200 hPa; The v component of the first wind is the v component of wind at 1000 hPa, 925 hPa, 850 hPa, 700 hPa, 500 hPa and 200 hPa; The relative humidity is 1000hPa, 925hPa, 850hPa, 700hPa, 500hPa and 200hPa; The second temperature is the temperature at a height of 2 m; The dew point temperature is the 2m high dew point temperature; The u component of the second wind is the u component of the wind at a height of 10 m; The v component of the second wind is the v component of the wind at a height of 10 m.

3. The method according to claim 1, characterized in that The power load data, the meteorological data, the sample power load data and the sample meteorological data are processed using embedding coding.

4. The method according to claim 1, characterized in that: The time resolution of the power load data, the meteorological data, the sample power load data and the sample meteorological data is 3 hours.

5. The method according to claim 1, characterized in that The short-term power load forecasting model specifically adopts a multi-input structure deep learning model based on a one-dimensional convolutional neural network. The short-term power load forecasting model includes multiple convolutional layers. After the data is input into the short-term power load forecasting model, it enters the multiple convolutional layers for processing. The convolution kernel size of the convolutional layer is 3×3, and the corresponding convolution operation adopts a local connection method and a weight sharing method. The receptive field is expanded by superimposing multiple convolutional layers. Each convolutional layer does not perform pooling, and only changes the number of convolution kernels. There is a BatchNorm layer and a softPlus nonlinear activation function layer after each convolutional layer. The number of convolution kernels of the last convolutional layer is 10. After the output of the last convolutional layer is flattened into one dimension, it is spliced ​​and fused with the power load data, and then input into the dense layer to predict the load data of 96 points on the next day as the short-term power load forecasting result.

6. The method according to claim 1, characterized in that The method further comprises: Acquiring the sample power load data and the sample meteorological data; Preprocessing the sample power load data and the sample meteorological data; The short-term power load forecasting model is trained based on the sample power load data and the sample meteorological data.

7. The method according to claim 6, characterized in that The training of the short-term power load forecasting model based on the sample power load data and the sample meteorological data comprises: Taking into account the impact of working day factors and holiday factors, the short-term power load forecasting model is trained based on the sample power load data and the sample meteorological data.

8. A short-term power load forecasting device based on deep learning, characterized in that: The device comprises: A first acquisition unit is used to acquire power load data of a target device cluster in a past time period, wherein it is assumed that power usage demand of different devices involved in the target device cluster remains stable in the past time period, and the past time period refers to a time period consisting of a historical time point with a preset time span from a current time point forward and the current time point; a second acquisition unit, configured to acquire meteorological data of the past time period, wherein the meteorological data includes high-altitude layer data and ground-layer data, the high-altitude layer data includes a first air temperature, a potential height, a relative humidity, a u component of a first wind, and a v component of the first wind, and the ground-layer data includes a sea level pressure, a second air temperature, a dew point temperature, a u component of a second wind, a v component of the second wind, and solar shortwave radiation; A prediction unit, used to input the power load data and the meteorological data into a preset short-term power load prediction model, wherein the short-term power load prediction model is a deep learning model, and the short-term power load prediction model is trained by sample power load data and sample meteorological data; The extraction unit is used to extract the short-term power load forecast result output by the short-term power load forecast model.

9. A processing device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 7 when calling the computer program in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 7.

Images