融合改进D-S证据理论与超标倍数法的鸡舍环境评估模型

Environmental Evaluation Model for Poultry Houses Integrating Improved D-S Evidence Theory and Exceedance Multiple Method

摘要:

规模化肉鸡舍环境是多因子动态交互的复杂系统,固定权重评估模型难以刻画其不确定性与参数异常的影响,导致评估结果与实际养殖需求出现偏差。 针对不同类型环境参数的特性,采用线性梯形隶属度函数处理温度、 湿度、风速等热环境参数,非线性三角形隶属度函数处理 CO2 、NH3 、PM2. 5 浓度等气体环境参数,将原始监测数据转化为隶属度值并构建基本概率分配(BPA),降低主观因素对概率分配的干扰。 为解决传统 D-S 证据理论在高冲突证据下融合失真的问题,引入邓氏熵(Deng entropy)量化证据的不确定性信息量,结合证据距离构建冲突度量因子,通过相似度矩阵计算各证据的支持度与可信度,实现对传统 D-S 组合规则的改进,提升多源证据融合的稳定性。 同时,考虑到超标参数对鸡群健康的显著影响,采用超标倍数法构建动态权重调整机制:基于各参数的安全阈值,通过线性公式结合遗传算法优化的放大系数(k1 )与非线性调节参数(a),对超标参数的权重进行动态再分配, 强化模型对异常参数的敏感性与惩罚力度。 实验结果表明:对 160 组环境参数进行评估时,模糊层次分析法评估准确率仅为 70. 00% ,改进证据理论评估结果准确率为 87. 50% ,融合超标倍数法后评估准确率为 93. 13% 。 相比之下,本文方法更能体现鸡舍实际环境情况。 本研究为规模化肉鸡舍环境的精准调控提供了理论支撑与技术方案,可应用于智慧养殖环境监控系统的优化升级。

Abstract:

The environment within large-scale broiler houses is a complex system characterized by dynamic, multi-factor interactions. Fixed-weight evaluation models failed to adequately capture its inherent uncertainty and the impact of parameter anomalies, resulting in discrepancies between evaluations and actual farming requirements. Tailored to the distinct characteristics of various environmental parameters, linear trapezoidal membership functions were applied to thermal factors (temperature, humidity, and wind speed), while nonlinear triangular membership functions were used for gas factors (CO2 , NH3 , and PM2. 5 ). By converting raw monitoring data into membership degrees to construct the basic probability assignment (BPA), the interference of subjective factors in probability distribution was effectively minimized. To resolve the fusion distortion inherent in traditional D-S evidence theory when handling highly conflicting evidence, Deng entropy was introduced to quantify the informational uncertainty of the evidence. Combined with evidence distance, a conflict measurement factor was constructed. By calculating the support and credibility of each piece of evidence via a similarity matrix, the traditional D-S combination rule was improved, thereby enhancing the stability of multi-source data fusion. Furthermore, recognizing the severe impact of out-of-limit parameters on flock health, an exceedance multiple method was employed to establish a dynamic weight adjustment mechanism. Based on the safety thresholds of individual parameters, this mechanism used a linear formula alongside an amplification coefficient (k1 )and a nonlinear adjustment parameter (a)—both optimized by a genetic algorithm—to dynamically redistribute the weights of out-of-limit parameters. This amplified the model's sensitivity and penalization of abnormal environmental factors. Experimental results evaluating 160 sets of environmental parameters revealed that the traditional fuzzy analytic hierarchy process (FAHP)achieved an accuracy of only 70. 00% , while the improved evidence theory alone reached 87. 50% . However, upon integrating the exceedance multiple method, the evaluation accuracy rose significantly to 93. 13% . Consequently, the proposed method more accurately reflected the actual conditions within the poultry house. The research result can provide solid theoretical support and a robust technical framework for the precise regulation of large-scale broiler house environments, offering valuable applications for the optimization and upgrading of smart farming environmental monitoring systems.