基于HMM和BP神经网络组合模型的用水行为识别

在当前水资源短缺以及用水量不断增加的背景下,识别农村居民用水行为,对于农村地区居民用水安全和管理、缓解水资源短缺具有重要的意义。为此,提出了一种隐马尔可夫模型(Hidden Markov Model,HMM)和BP神经网络(Back Propagation,BP)相结合的组合模型,模型综合了BP网络优秀的分类识别能力和HMM强大的时域建模能力的优点。该模型首先为居民用水行为的6个事件分别建立1个HMM,然后计算各个模型的最佳状态的输出概率,再将此概率和期望输出共同训练BP神经网络,最后选取测试数据和已建立的组合模型进行匹配,得到识别结果。研究结果表明:该组合模型在用水行为识别准确度上比单独应用HMM模型高8. 78%,比单独应用BP神经网络高8. 92%,具有一定的应用价值。

Water-using behavior recognition based on HMM and BP neural network mixing model

Recogniton of water-using behavior of rural residents is of great significance to alleviate the shortage of water resources and improve the safety and management of water consumption in rural areas, where the water is in short and water consumption keeps increasing. A mixing model combining the HMM and BP neural network is proposed, based on the excellent classification ability of BP network and the powerful modeling capabilities in time domain of HMM. This model established a HMM for the six events of water-using behavior and the output probability of each model were calculated. Then the BP neural network was trained through the probability and expected output, and the test data and the established combination model were selected to match and the recognition result was obtained. The results showed that the new mixing model was 8.78% more accuracy in water-using behavior recognition than the HMM model alone, and 8.92% more accuracy than the BP neural network alone, showing its application value.