金元鲃鱼养殖生长预测数据

解决养殖过程中 “精准投喂”“环境调控” 问题，实现金元鲃鱼“滇优1号”体重、日投喂量及生长等级的精准预测，为养殖生产提供数据支撑，提升金元鲃鱼“滇优1号”养殖效益与生长品质。本数据集合适用于金元鲃鱼工厂化规模化养殖的生长管理与风险预警，降低养殖损耗和成本，助力养殖主体提质增效、增强市场竞争力，同时可推动金元鲃鱼养殖模式向绿色高效可持续转型。记录输入变量：养殖天数（监测时间 - 鱼苗投放起始时间）、水温、溶解氧、pH、氨氮浓度、亚硝酸盐浓度、日投喂量、鱼体重。
1、使用回归算法分析输入变量与鱼体重、日投喂量的非线性关联，确定特征重要性权重（养殖天数对体重权重 57.5%、溶解氧对日投喂量权重 86.9%）；参数配置：随机森林回归设 100 棵决策树、最大深度 8 层，梯度提升树学习率 0.1，迭代终止条件为预测误差达标；评估指标：体重预测目标为 R²≥0.98、RMSE＜3g，实际达标 R²=0.986、RMSE=2.57g；日投喂量预测目标为 R²≥0.99，实际达标 R²=1.0。
2、使用多分类算法分析输入变量与5级生长等级（按体重分位数划分）的类别关联，构建概率判别规则，取置信度最高的等级为最终结果；评估指标：等级预测目标为准确率≥70%、平均置信度≥0.4，实际达标准确率= 73.93%，平均置信度0.45，仅当置信度≥30% 时判定为有效预测。训练与优化规则：优化目标：以“回归指标最优、分类指标达标”为核心，明确体重预测 R²≥0.98、RMSE＜3g，日投喂量 R²≥0.99，生长等级准确率≥70%；优化方式：利用体重、日投喂量实际值及生长等级实际类别，通过网格搜索调优超参数，结合5折交叉验证避免过拟合。

This dataset addresses the challenges of "precision feeding" and "environmental regulation" in aquaculture, enabling accurate prediction of body weight, daily feeding amount and growth grade of Jinyuan barbel fish "Dianyou No.1", providing data support for aquaculture production and improving the breeding benefits and growth quality of this strain. This dataset is suitable for growth management and risk early warning in large-scale industrialized aquaculture of Jinyuan barbel fish "Dianyou No.1", reducing aquaculture losses and costs, helping aquaculture entities improve quality and efficiency and enhance market competitiveness, while promoting the green, efficient and sustainable transformation of Jinyuan barbel fish aquaculture models.

Recorded input variables include: culture days (monitoring time minus the initial stocking time of fish fry), water temperature, dissolved oxygen, pH, ammonia nitrogen concentration, nitrite concentration, daily feeding amount, and fish body weight.

1. Regression algorithms were used to analyze the nonlinear correlation between input variables and fish body weight and daily feeding amount, and the feature importance weights were determined: culture days contributes 57.5% to body weight prediction, and dissolved oxygen contributes 86.9% to daily feeding amount prediction. Parameter settings: Random Forest Regression was configured with 100 decision trees and a maximum depth of 8 layers; Gradient Boosting Tree was set with a learning rate of 0.1, and the iteration terminated when the prediction error met the standards. Evaluation metrics: For the body weight prediction task, the target is R² ≥ 0.98 and RMSE < 3g; the actual achieved performance is R² = 0.986 and RMSE = 2.57g. For the daily feeding amount prediction task, the target is R² ≥ 0.99, and the actual achieved R² = 1.0.

2. Multi-classification algorithms were employed to analyze the categorical correlation between input variables and 5-level growth grades (divided based on body weight quantiles), and a probability discrimination rule was constructed, taking the grade with the highest confidence as the final prediction result. Evaluation metrics: For growth grade prediction, the target is accuracy ≥ 70% and average confidence ≥ 0.4; the actual achieved performance is accuracy = 73.93% and average confidence = 0.45. Valid predictions are only determined when the confidence level is ≥ 30%.

Training and optimization rules: The core optimization objectives are "optimal regression metrics and qualified classification metrics", specifically requiring that body weight prediction achieves R² ≥ 0.98 and RMSE < 3g, daily feeding amount prediction achieves R² ≥ 0.99, and growth grade prediction achieves accuracy ≥ 70%. Optimization method: Grid search was used to tune hyperparameters with the actual values of body weight, daily feeding amount and actual growth grade categories, combined with 5-fold cross-validation to avoid overfitting.

搜集汇总

背景与挑战

背景概述

该数据集专注于金元鲃鱼养殖的生长预测，通过采集养殖天数、水温、溶解氧等环境变量，利用回归和分类算法精准预测鱼体重、日投喂量和生长等级，以支持养殖过程中的精准投喂和环境调控。数据集在预测性能上表现优异，例如体重预测的R²达到0.986，生长等级预测准确率为73.93%，适用于规模化养殖的管理优化和风险预警，有助于提升养殖效益和可持续性。