Neural Network Demand Forecasting via Mesh Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing demand forecasting systems face errors when using a single model for predicting service demands across different geographical areas, particularly due to deviations in demand frequency distributions and noise in input features.
Innovation Solution
A learning system that normalizes demand features for each geographical area before inputting them into a neural network, reducing errors by accounting for discrepancies in feature scales and distributions, thereby enabling high-accuracy forecasts using a single model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a forecast model is created for each mesh, then the model can capture local peculiarities, but errors increase when noise is present in input features
Solution Approach 1:
The patent segments the learning data by mesh to capture local characteristics while maintaining a unified model structure. The normalization process is applied separately for each mesh, allowing local adaptation without creating separate forecast models for each mesh.
Solution Approach 2:
The patent applies local quality by performing normalization specific to each mesh based on its historical demand distribution. This allows the model to adapt to local characteristics (different demand frequencies across meshes) while using a single universal forecast model, thereby capturing local peculiarities without sacrificing robustness.
2Reliability
If a single forecast model is used for all meshes, then the model is robust against noise and universally applicable, but errors increase when there is deviation in demand frequency distribution between meshes
Solution Approach 1:
The patent changes the parameter scale by normalizing the number of demands for each mesh based on its own historical distribution. This transformation allows the single forecast model to process data from different meshes uniformly, eliminating the negative impact of distribution deviations while maintaining model robustness.
Solution Approach 2:
The patent creates equipotential conditions by normalizing demand data across different meshes so that they all operate on a comparable scale. This allows the single forecast model to treat all meshes equally, preventing any single mesh's distribution characteristics from dominating or causing errors in the forecast.
3Device complexity
If demand data with different scales is input directly to the model, then the processing is simple, but the model cannot accurately capture local characteristics
Solution Approach 1:
The patent performs preliminary normalization of the demand data for each mesh before inputting it to the forecast model. This pre-processing step adjusts the scale of demand data according to each mesh's historical characteristics, enabling the model to accurately capture local features without requiring complex mesh-specific model structures.
Data Source
AI summary
A learning system includes a learning data acquisition unit that generates a neural network included in a single forecast model for performing a forecast of demand for a service through machine learning and acquires, as learning data, an actual demand value of the service and a feature amount for learning for each mesh associated with a past period, the feature amount for learning including the number of demands for the service in each mesh in each period as a feature amount, a learning data normalization unit that normalizes at least the number of demands in the feature amount for learning for each mesh, and a generation unit that performs machine learning on the basis of an actual demand value and a demand forecast value obtained by inputting the feature amount for learning including the normalized number of demands to the neural network, and generates the neural network.