Power system regional load prediction method based on environmental characteristic self-adaption

Abstract

The invention relates to the technical field of power system load prediction, in particular to a power system regional load prediction method based on environmental feature self-adaption, which comprises the following steps: S1, constructing an environmental feature extraction layer, accessing a data source in real time, and constructing'photovoltaic output fluctuation ratio-temperature gradient 'joint features by the environmental feature extraction layer; s2, constructing a hybrid prediction layer, processing time sequence dependence by using an LSTM neural network, constructing a random forest regression sub-model, learning a nonlinear relationship between environment characteristics and loads, establishing a dynamic weight distribution module, calculating an environment characteristic similarity matrix through an attention mechanism, and calculating an environment characteristic similarity matrix; the method can effectively solve the problems that an existing prediction method cannot fully consider the non-linear coupling relation between environment characteristics and power loads, so that the power grid prediction precision is insufficient; the problems of insufficient adaptability to the difference of environmental factors, poor generalization ability, high prediction failure rate, high probability of periodic distortion, data access lag, high response delay and low real-time performance are solved.

Description

Technical Field

[0001] This invention relates to the field of power system load forecasting technology, specifically to a regional load forecasting method for power systems based on environmental characteristics. Background Technology

[0002] In recent years, with the rapid growth of my country's economy, the power industry has continued to develop and expand. Power load forecasting has become increasingly important for the safe and economical operation of the power system, serving as the basis for economic dispatch, energy storage management, future energy contracts, and power plant maintenance plans. The accuracy of power load forecasting is directly related to the operation of the power system. If the forecast value is too high, it will lead to excessive power production, resulting in energy waste and wear and tear on production machinery; while if the forecast value is too low, it will lead to insufficient power production, affecting people's lives and causing economic losses.

[0003] However, existing traditional regional load forecasting methods still have shortcomings. Specifically, existing forecasting methods, such as ARIMA and single LSTM, do not fully consider the nonlinear coupling relationship between environmental characteristics such as photovoltaic power output fluctuations and seasonal temperature changes and power load, resulting in insufficient grid forecasting accuracy. At the same time, existing methods are not adaptable enough to the differences in environmental factors such as meteorology and photovoltaic installed capacity density between mountainous and plain areas across regions. The load is affected by nonlinear factors such as meteorology, geography, and energy structure. Existing models have poor generalization ability, and fixed-weight models have a high failure rate in extreme weather. In areas with high penetration of new energy sources and a large proportion of photovoltaic power, the forecasting results of traditional models are prone to periodic distortion due to frequent power output fluctuations. In addition, the access of meteorological data and new energy power generation data is lagging, and manual feature calibration leads to high response delay and low real-time performance.

[0004] Therefore, a regional load forecasting method for power systems based on environmental characteristics is needed to solve the problems mentioned in the background. Summary of the Invention

[0005] The purpose of this invention is to provide a power system regional load forecasting method based on environmental characteristics to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A regional load forecasting method for power systems based on environmental characteristics adaptively includes the following steps: S1, construct an environmental feature extraction layer, access the data source in real time, and construct a joint feature of "photovoltaic power output fluctuation rate-temperature gradient" in the environmental feature extraction layer; S2 constructs a hybrid prediction layer, uses an LSTM neural network to handle time series dependencies, constructs a random forest regression sub-model and learns the nonlinear relationship between environmental features and load, establishes a dynamic weight allocation module, calculates the environmental feature similarity matrix through an attention mechanism, and automatically adjusts the sub-model weights according to changes in environmental features.

[0007] As a preferred embodiment of the present invention, the data sources accessed in real time in S1 include meteorological bureau API, photovoltaic power station SCADA system and geographic information system.

[0008] As a preferred embodiment of the present invention, the meteorological bureau API includes data such as temperature, humidity, and irradiance; the photovoltaic power station SCADA system includes output fluctuation rate and inverter status data; and the geographic information system includes altitude and regional power consumption structure data.

[0009] As a preferred embodiment of the present invention, the time series dependence in S2 includes the impact of continuous temperature changes on the load.

[0010] As a preferred embodiment of the present invention, the nonlinear relationship between environmental characteristics and load in S2 includes the sudden increase in air conditioning load under high temperature weather and the degree of influence of temperature rise on load increase.

[0011] As a preferred embodiment of the present invention, the adjustment of sub-model weights in S2 includes increasing the weights of LSTM and decreasing the weights of random forests during cold weather.

[0012] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, an environmental feature extraction layer is constructed, and data sources are accessed in real time. The environmental feature extraction layer constructs a joint feature of "photovoltaic power output volatility - temperature gradient". A hybrid prediction layer is constructed, and an LSTM neural network is used to process time series dependencies. A random forest regression sub-model is constructed and learns the nonlinear relationship between environmental features and load. A dynamic weight allocation module is established, and an environmental feature similarity matrix is ​​calculated through an attention mechanism. The weights of the sub-model are automatically adjusted according to changes in environmental features. This technical solution introduces the joint feature of "photovoltaic power output volatility - temperature gradient" to solve the prediction distortion problem in areas with high penetration of new energy. At the same time, a lightweight model is deployed at the edge node of the power grid dispatch, reducing the response delay from 15 minutes to 45 seconds. The model parameters are automatically updated within a fixed time to adapt to sudden weather conditions such as short-term heavy rainfall causing photovoltaic power output to drop to zero. This solves the problems of existing prediction methods, which do not fully consider the nonlinear coupling relationship between environmental features and power load, resulting in insufficient power grid prediction accuracy, insufficient adaptability to differences in environmental factors, poor generalization ability, high prediction failure rate, easy occurrence of periodic distortion, data access lag, high response delay, and low real-time performance. Detailed Implementation

[0013] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] To facilitate understanding of the present invention, a more comprehensive description of the invention will be provided below, along with several embodiments. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0015] In this embodiment, the present invention provides a technical solution: A regional load forecasting method for power systems based on environmental characteristics adaptively includes the following steps: S1, construct an environmental feature extraction layer, access the data source in real time, and construct a joint feature of "photovoltaic power output fluctuation rate-temperature gradient" in the environmental feature extraction layer; S2 constructs a hybrid prediction layer, uses an LSTM neural network to handle time series dependencies, constructs a random forest regression sub-model and learns the nonlinear relationship between environmental features and load, establishes a dynamic weight allocation module, calculates the environmental feature similarity matrix through an attention mechanism, and automatically adjusts the sub-model weights according to changes in environmental features.

[0016] Furthermore, the data sources accessed in real time in S1 include meteorological bureau API, photovoltaic power plant SCADA system, and geographic information system.

[0017] Furthermore, the meteorological bureau API includes data such as temperature, humidity, and irradiance; the photovoltaic power station SCADA system includes output fluctuation rate and inverter status data; and the geographic information system includes altitude and regional power consumption structure data.

[0018] Furthermore, the time series dependence in S2 includes the impact of continuous temperature changes on the load.

[0019] Furthermore, the nonlinear relationship between environmental characteristics and load in S2 includes the sudden increase in air conditioning load under high temperature weather and the degree of influence of temperature rise on load increase.

[0020] Furthermore, the adjustment of sub-model weights in S2 includes increasing the weights of LSTM and decreasing the weights of random forest during cold weather.

[0021] Workflow of this invention: When using the regional load forecasting method for power systems based on environmental characteristics designed in this scheme, an environmental feature extraction layer is constructed, and data sources are accessed in real time. The data sources include meteorological bureau API, photovoltaic power plant SCADA system and geographic information system. Meteorological bureau API includes data such as temperature, humidity and irradiance. Photovoltaic power plant SCADA system includes output fluctuation rate and inverter status data. Geographic information system includes altitude and regional power consumption structure data. An environmental feature extraction layer constructs a joint feature of "photovoltaic power output fluctuation rate - temperature gradient"; A hybrid prediction layer is constructed, and an LSTM neural network is used to process time series dependencies, including the impact of continuous temperature changes on the load. A random forest regression sub-model was constructed to learn the nonlinear relationship between environmental characteristics and load. The nonlinear relationship includes the sudden increase in air conditioning load under high temperature weather and the degree of influence of temperature rise on load increase. A dynamic weight allocation module is established, which calculates the environmental feature similarity matrix through an attention mechanism and automatically adjusts the sub-model weights according to changes in environmental features, including increasing the weights of LSTM and decreasing the weights of random forests during cold weather. Through multi-environmental feature coupling analysis, the load forecasting error rate is low in areas with high penetration of new energy sources, and the forecasting stability under extreme weather conditions is greatly improved. At the same time, the dynamic weighting mechanism can adapt to load changes under different weather temperatures, avoid the failure problem of traditional fixed weight models, improve the forecasting stability under extreme weather conditions, support the provincial power grid dispatching system to access meteorological bureau and new energy power plant data in real time, eliminate the need for manual intervention in feature calibration, and reduce the configuration of reserve capacity, save provincial power grid peak-shaving costs, and reduce photovoltaic curtailment rate.

[0022] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A regional load forecasting method for power systems based on environmental characteristics, characterized in that, Includes the following steps: S1, construct an environmental feature extraction layer, access the data source in real time, and construct a joint feature of "photovoltaic power output fluctuation rate-temperature gradient" in the environmental feature extraction layer; S2 constructs a hybrid prediction layer, uses an LSTM neural network to handle time series dependencies, constructs a random forest regression sub-model and learns the nonlinear relationship between environmental features and load, establishes a dynamic weight allocation module, calculates the environmental feature similarity matrix through an attention mechanism, and automatically adjusts the sub-model weights according to changes in environmental features.

2. The regional load forecasting method for power systems based on environmental characteristics according to claim 1, characterized in that: The data sources accessed in real time in S1 include meteorological bureau API, photovoltaic power station SCADA system and geographic information system.

3. The regional load forecasting method for power systems based on environmental characteristics according to claim 2, characterized in that: The meteorological bureau API includes data such as temperature, humidity, and irradiance; the photovoltaic power station SCADA system includes output fluctuation rate and inverter status data; and the geographic information system includes altitude and regional power consumption structure data.

4. The regional load forecasting method for power systems based on environmental characteristics according to claim 1, characterized in that: The time series dependency in S2 includes the impact of continuous temperature changes on the load.

5. The regional load forecasting method for power systems based on environmental characteristics according to claim 1, characterized in that: The nonlinear relationship between environmental characteristics and load in S2 includes the sudden increase in air conditioning load under high temperature weather and the degree of influence of temperature rise on load increase.

6. The regional load forecasting method for power systems based on environmental characteristics according to claim 1, characterized in that: The adjustment of sub-model weights in S2 includes increasing the weight of LSTM and decreasing the weight of random forest during cold weather.