The issue of comfort in subway stations is typically analyzed using the predicted mean vote (PMV) and the predicted percent dissatisfied (PPD) indices. Based on the PMV-PPD comfort indices calculation model, the weight proportions of different environmental parameters on comfort were studied. The PMV calculation model was improved by considering the spatiotemporal characteristics of passengers' clothing and activity levels. The PPD calculation model was enhanced by taking into account the impact of drastic environmental temperature changes on comfort. The environmental comfort of public areas in stations was then analyzed using the improved PMV-PPD calculation model. On this basis, the feasibility of predicting environmental comfort using long short-term memory(LSTM) networks was explored. The research results indicate that the weight proportions of metabolic rate, air temperature, clothing thermal resistance, and humidity on environmental comfort are 0.558, 0.260, 0.113, and 0.069, respectively. At a given time, the maximum differences in PMV and PPD at different monitoring points on the platform are approximately 15% and 60%, respectively. The improved PMV-PPD calculation model is found to be more universally applicable compared to the traditional PMV-PPD calculation model. The neural network is shown to accurately predict PMV and PPD values, with a maximum error of 8% for PMV and 14% for PPD between the actual and predicted values.