基于朴素贝叶斯增量学习算法的个体热舒适预测模型
Personal thermal comfort prediction model based on naive Bayesian incremental learning algorithm
摘要：
目前通过机器学习建立个体热舒适模型需要在建模前获得大量数据，不符合实际应用场景，并且更新模型以适应个体热舒适偏好随季节等因素变化的成本太高。基于朴素贝叶斯算法，使用增量学习方法建立和更新了热舒适模型，修改了模型中的先验概率和后验概率。通过实验数据验证，该方法平均学习26组样本后，预测结果即可与个体实际热期望保持高度一致（Kappa系数为0.699 4）。并且相比已有的模型更新方法，本文提出的模型建立与更新方法节约了17%的时间成本和90%以上的空间成本，更加适合部署到计算与存储资源较少的轻量化智能终端。
Abstract：
The establishment of personal thermal comfort model through machine learning requires the acquisition of a large amount of data before modeling, which is not in line with the actual application scenarios. Moreover, it costs too much to update the model to adapt to the change of personal thermal comfort preference with seasons and other factors. Therefore, based on the naive Bayesian algorithm, establishes and updates the thermal comfort model using the incremental learning method, in which the prior probability and the posterior probability of the model is modified. Through experimental data verification, after learning 26 samples averagely by this method, the predicted results can be substantial agreement with the true thermal preference (Kappa coefficient is 0.6994). Moreover, compared with the existing update methods, the establishment and update method proposed in this paper saves 17% of the time cost and more than 90% of the space cost. Thus, it is more suitable for deployment to lightweight intelligent terminals with fewer computing and storing resources.
Keywords:personal thermal comfort model, machine learning, incremental learning, naive Bayesian algorithm, Kappa coefficient