To ensure the successful execution of on-orbit space missions, the operational data of space payloads must be continuously monitored. Due to the characteristics of interval-scaled variable parameters that change and fluctuate randomly with telemetry parameters, the traditional interval range anomaly detection method has low accuracy, and the deep learningbased approach is limited in the study of parameter changes with commands. In this paper, an anomaly detection method based on LSTM network model is proposed. First, encode command (CMD) into CMD Time Series and construct the corresponding dataset. Then, generate the parameter expectation values using LSTM and design the anomaly detection discriminative parameters and rules. By training the network model and setting the thresholds of the discriminative parameters according to the performance, we demonstrate the feasibility of the method by taking the actual parameters of an experimental payload as an example.