Environmental emergency decision support system based on Artificial Neural Network

We present a general methodology for developing environmental emergency decision support systems (EEDSS) based on an Artificial Neural Network (ANN). We highlight the method for developing the system using an illustrative example of an unexpected atmospheric accident with an ANN prototype system for a district in Shanghai. The network architecture of the ANN is introduced. Then the development process and key technologies are addressed. The procedures for matching the environmental emergency decision support characteristics are as follows: (1) digitization (coding) of case information and emergency measures, in which the information of cases are divided into the input attributes and decision-making information, and standardized and digitized through the Feature Evaluation (FE) method and the Intensity Hierarchical (IH) method, respectively; (2) construction of environmental emergency ANN, in which Gradient Descent with Momentum and Adaptive Learning Rate (GDMALR) method (traingdx function), a modified back-propagation algorithm, is employed to do training; and (3) translation (decoding) of decision-making information, in which output data of ANN is interpreted into practical contingency measures with Translation Based on Conventional Import Ratios (TBCIR) method. The training features, time, errors, accuracy, and input attribute weights of the prototype system are analyzed. The usage of the prototype system is demonstrated through a hypothetical case. This article encounters the challenge of ANN’s own lack of training samples. We discuss to the concept of integrating Case-Based Reasoning (CBR), Genetic Algorithm (GA), and ANN to overcome this difficulty and form a technology system for generating useful decision support information for environmental emergency response.