Existing research on emtional response generation only considers the expression of emotion and does not consider the impact of different emotional intensities in responses. To do this end, we proposed a response generation model for specifying emotional intensity, which adds an gated decoding module to capture the changes in emotion intensity during the decoding process. The lexical type selection module is also added to generate different types of words at appropriate positions by controlling the weights of emotion words, degree adverbs, generic words and realize the expression of emotional intensity in text finally. The experiments were based on the open domain multi-round dialogue corpus of NLPCC 2018. The experimental system for generating specified emotion intensity responses was constructed, and experiments on emotional intensity consistency and content association were carried out. Experimental results show that the expression in terms of emotional intensity consistency can be achieved and has good performance in content generation compared to other baseline models. The model improved by 0.41 compared to the optimal model in terms of emotional intensity consistency, was able to generate responses with consistent emotional intensity according to the specified emotional intensity, and had some ability to generate emotional intensity responses.