Wind Power Probability-Density Forecasting With Time-Variant Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wind power forecast methods, particularly those using time-invariant neural networks, suffer from low accuracy and inability to accurately predict the uncertainty of wind power, leading to inefficiencies in power system dispatching.
Innovation Solution
A time-variant deep feed-forward neural network model is constructed with multiple layers, incorporating input and output layers that take wind power data and probability density distribution at adjacent moments, trained using a mixed density network and alternate output method, allowing for multi-step forecasting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a time-invariant neural network model (RNN or CNN) is used for wind power forecasting, then the model structure is simple and easy to implement, but the forecast accuracy is low and the uncertainty of wind power cannot be accurately forecasted
Solution Approach 1:
The patent applies the dynamics principle by transforming the traditional time-invariant neural network into a time-variant neural network. The model parameters are no longer fixed but vary with time, allowing the network to adapt to the changing statistical characteristics of wind power data. This is achieved by introducing time-varying parameters that capture the temporal evolution of wind power uncertainty, thereby improving forecast accuracy while maintaining model interpretability.
Solution Approach 2:
The patent implements parameter changes by modifying the neural network parameters from constant values to time-dependent functions. The model parameters are updated dynamically based on the input data sequence, enabling the network to adapt to different operational conditions and time periods. This parameter transformation allows the model to capture the non-stationary nature of wind power generation, resolving the contradiction between model simplicity and forecasting precision.
2Measurement precision
If a time-variant deep feed-forward neural network model is constructed to improve forecast accuracy, then the forecast accuracy and uncertainty prediction improve, but the model complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the complex time-variant forecasting problem into multiple manageable components through a multi-layer neural network architecture. Each layer processes specific features and transformations, breaking down the overall complexity into modular units. This segmented approach allows the model to handle time-variant characteristics systematically while maintaining computational tractability and facilitating training through localized gradient propagation.
3Ease of operation
If traditional point forecast method is used, then the forecast value is certain and easy to obtain, but the uncertainty of wind power cannot be quantitatively described
Solution Approach 1:
The patent applies dimensionality change by transitioning from a one-dimensional point forecast to a multi-dimensional probability density forecast. Instead of predicting a single value, the model outputs a full probability density function that captures the distribution characteristics of wind power. This dimensional expansion preserves uncertainty information while maintaining operational simplicity through the use of parametric distribution assumptions and efficient computation methods.
Data Source
AI summary
A forecast method and system of wind power probability density. The forecast method includes: acquiring wind power data, preprocessing the wind power data, establishing a data set; then, constructing a time-variant deep feed-forward neural network forecast model, where the model includes multiple layers of neural networks, and each layer of neural network includes an input layer, a hidden layer and an output layer which are connected in sequence; taking wind power data at adjacent moments as an input of two input layers of two adjacent layers of neural networks, taking probability density distribution of wind power at adjacent moments as an output of two output layers of two adjacent layers of neural networks, and training and testing the model; inputting the wind power data to be forecasted into the trained time-variant deep feed-forward neural network forecast model for forecasting to obtain a more accurate and reliable wind power forecast result.