Research on communication performance indexes in the key area based on stream processing and LSTM algorithm

In existing research, the analysis of network performance in the key area are is based on offline statistical data. This method has the disadvantages of slow speed in summarizing cell indexes, long time consumption for calculating and slow positioning of problematic cells. In addition, this method cannot predict the trend of index changes within a region and evaluate the optimization effect on communication cells. To solve the above problems, this study proposes using stream processing and LSTM algorithm to analyze indexes of the key region. First, this study obtains the list information of all communication cells in the key area and downloads 15-minute granularity performance files required for six indexes calculations. The Java program completes parsing and sends them to Kafka. Next, Spark Steaming reads data from Kafka, calculates the index values of communication cell, and writes them to ClickHouse database. If value of the index is abnormal, information about the communication cell in poor quality will be displayed as a report. If the number of exceptions for cell exceeds the set value, a text message reminder will be sent for network optimization. In addition, using Spring Boot programs to summarize the indexes in the key area into daily granularity. Based on the historical data of daily granularity, an LSTM model is established to predict the trend of indexes in the key area. The optimization effect of the communication cell is evaluated based on this method, and if the prediction index exceeds a set value, an SMS alert will be triggered.