6 results about "Genetic programming" patented technology

This invention relates to the fields of artificial intelligence and computer vision, and provides a few-shot fine-grained image classification method based on multi-scale feature extraction, attention, and relation enhancement using genetic programming. The method includes: extracting multi-scale deep features using a pre-trained deep feature extraction network (preferably ResNet-18); designing a strongly typed genetic programming tree structure comprising an input layer, a multi-scale feature extraction layer, a channel selection layer, an attention layer, a feature concatenation layer, and an output layer; enhancing discriminative regions and suppressing background noise through a spatial attention mechanism; developing a prototype-query relation enhancement module to amplify the regions of common interest between the support set and the query set to enhance feature representation; and using the sum of classification accuracy and cluster separation score as a fitness function to guide the evolutionary search. This invention combines deep neural network representation with genetic programming evolutionary search, achieving excellent classification performance on fine-grained few-shot image classification tasks while maintaining interpretability.

The application discloses a photovoltaic power station power prediction method based on prior knowledge guided genetic programming, first, obtaining training data preprocessed by symbol normalization, learning photovoltaic power generation symbol rules through self-supervised pre-training of a Transformer model, and outputting a prior symbol vector of a target function; a double fitness collaborative optimization mechanism containing numerical fitting accuracy and physical mechanism similarity is designed, and the prior vector is used to strengthen and select individuals with high physical matching degree; a Pareto multi-objective optimization is used to balance the double-dimensional conflict, the prior knowledge is transferred to the whole evolution process of genetic programming, and finally the optimal explainable prediction model is extracted from the Pareto frontier solution set. The application solves the problems of traditional genetic programming evolution blindness, easy falling into local optimum and poor physical consistency of generated model, realizes explicit physical explainability of the model while ensuring prediction accuracy, and can be widely applied to power generation scheduling and operation optimization of photovoltaic power stations.

This invention discloses a method and system for constructing a rainfall-runoff model. The method involves collecting and preprocessing relevant daily sequence data within a watershed; constructing a hydrological model component library; pre-assembling the hydrological model structure; defining a parameter space; performing a collaborative search using genetic programming; simulating daily runoff using the optimal model structure and parameters obtained through the search and calibration; evaluating the model's generalization performance; and outputting the optimal model structure type, optimized parameter vector, calibration set, validation set performance indicators, and runoff simulation sequence. This invention can automatically generate rainfall-runoff models with adaptive model structures and interpretable hydrophysical mechanisms, effectively improving the automatic discovery capability of model structures and the interpretability of hydrophysical mechanisms, while reducing manual workload while avoiding overfitting.

The application discloses a kind of multi-objective synthetic oversampling methods for unbalanced data, based on genetic programming, the synthesis of minority class sample is modeled as constraint multi-objective optimization problem, including: genetic programming driven batch synthesis: individual is coded as a complete sample synthesis rule, a batch of high-quality samples is automatically generated in single run by evolution search, capture complex nonlinear relationship beyond linear interpolation;Constraint multi-objective evaluation mechanism: the sample generation process is modeled as constraint multi-objective optimization problem, while optimizing inter-class separation and local neighborhood structure, and from classification boundary, distribution consistency, sample diversity, sample quality four aspects explicitly constraint sample, ensure the effectiveness of generated sample;Constraint processing mechanism: design special constraint processing strategy to select feasible solution to participate in sample generation, promote population feasibility and realize batch synthesis of high-quality minority class sample;The application significantly improves sample generation efficiency, is advantageous to process large-scale unbalanced data.

This invention discloses a genetic programming mutation probability optimization method based on the maximum mutual information coefficient. In traditional genetic programming, feature selection tends to shift from initial completely random selection to selective selection with bias. This invention constrains the search direction of genetic programming, thereby improving search efficiency. The method includes: Step S1, using the maximum mutual information coefficient to measure the correlation between each feature and the target in the dataset, and merging the correlations of each feature and the target into a correlation vector; Step S2, determining the probability distribution of genetic programming when selecting new features through mutation based on the correlation vector; Step S3, performing genetic programming evolution, during which the probability distribution remains fixed and is unaffected by the number of features selected within the population. This invention reduces the impact of random initialization on the overall performance of genetic programming, enhances the robustness of genetic programming on fundamental problems, improves training efficiency and accuracy, and enhances model generalization performance.