An EEMD neural network based real-time data abnormal value detection method is provided, and the problem that the existing real-time data abnormal value detection method does not consider the historical data abnormal value is taken into account. The method comprises: obtaining historical time series data, and sorting the historical data according to a chronological order; using a median method to carry out preliminary detection on the historical data; finely detecting the historical data by using an EEMD method, and replacing the detected abnormal value with a zero value; then using the curve fitting method to fill the zero value, that is, correcting the abnormal value, and obtaining historical data closer to the objective reality after carrying out abnormal value detection and correction; and finally, by learning the historical data, using the neural network method so that the to-be-reported real-time data can be more accurately predicted, comparing the predicted value with the real-time reported monitored value so that whether there is abnormality can be determined, and correcting the abnormality. The method can be used to detect the real-time data abnormal value in one-dimensional time series, and is applicable to a wide range of fields such as water resources, traffic, weather, thermal power generation and other real-time monitoring data abnormal value detection.