Vectorial-Criterion-BasedAssessment of Ecological State ofthe Natural-Engineering System

Abstract

The article deals with the issues of assessing the ecological state of the territory of natural-engineering systems as the critical indicator of their stability. A brief analysis of the natural environment pollution assessment criteria based upon the scalar convolutions of ecological monitoring data arrays is provided. It is stated that neglecting mutual correlation of the pollution indicators being measured in the course of ecological monitoring reduced informative value of the mentioned criteria. Instructional aspects of plotting the territory pollution assessment criterion based upon dimension reduction and de-correlation of the ecological monitoring data array is presented. The results of monitoring the soil mantle pollution at the territory of the local natural-engineering system comprising the chemical weapons destruction facility and the natural environment within the boundaries of the protective action zone are taken as the example. The results of use of the Principal Components Analysis for dimension reduction and de-correlation of the monitoring data array are presented. The possibility of substituting three principal components for eleven pollution indicators without sacrificing the acceptable informative value of description is presented. Switch to orthogonal basis in the principal components’ space allows reviewing the values of the latter at the point as the projection in general case of a multi-dimensional vector which is called the damage vector and can be used as the integrated indicator of pollution. The options of pollution assessment on the basis of comparison of the current and initial moduli of the pollution vector as well as with the use of the surface of critical pollutions is considered. The possibility to simplify the structure of neural networks for forecasting the pollution degree by means of the neural network’s single output parameter in the form of the pollution vector is provided. The example of using the neural network trained on the soil mantle background monitoring data for forecasting the current soil pollution at the points of the protective action zone is provided. A perceptron having one hidden layer describes satisfactorily the distribution of the pollution vector over the territory five years after the man-made facility has been commissioned is demonstrated.