Research On Unfrozen Water Content Of Frozen Soils Based On Near Infrared Spectroscopy

After the concept of unfrozen water content in frozen soils presented, a lot of researcher carried out a series test and study by calorimetric experiment. At present, the main approaches that have been used include the nuclear magnetic resonance (NMR),the neutron spin-echo (NSE), the time domain reflectometry(TDR)and the dielectric capacity method. But most of them inherent shortcoming, such as expensive equipment, complex operation, needing professional tests, or lengthy testing time. It is desirable to develop a new test method for ordinary researchers in an average laboratory.This text mainly carried on the research at following several respect:In this paper, the near-infrared spectroscopy applied to the measurement of unfrozen water content in frozen soil in order to achieve rapid, non-destructive testing purposes. Built near infrared spectroscopy test platform of the unfrozen water content, use soil moisture sensor SM2801B measure unfrozen water content, temperature sensor DS18B20 measurements the frozen soils real-time temperature and using near infrared spectrometer collected power value calibration of the unfrozen water content. Collected at each temperature corresponding to the time of unfrozen water content and the near-infrared spectral acquisition to power values so the power value can be contented with unfrozen water content. The collected data is stored, the establishment of permafrost Unfrozen water content forecasting model.Variation for frozen soil unfrozen water content and ice content are complex.Frozen soil unfrozen water content and ice content affected by the environment of high nonlinearity,and frozen soil ice content cannot be measured directly.To solve the above problems,we propose two estimation models of unfrozen water and ice content based on BP artificial neural network. BP network is of slow convergence and easy to fall into local minimum point. The design selects five groups of different initial moisture content as training samples of frozen soil. Using the measured data (Near-infrared spectrometer detected power value) as the input and the measured data (unfrozen water content) as the output, getting the estimated model of unfrozen water content by training. Use single hidden layer and double hidden layer built the net. After the train, given unfrozen water content forecasting model with the model validation and testing. Ice forecast model with the power and value of unfrozen water content as input, the initial moisture content of unfrozen water content obtained by subtracting the calculated ice content as output, the BP neural network have a good recognition and can get more accurate estimation results.In this paper, the near-infrared spectroscopy technology have been introduced into unfrozen water content and ice content in frozen soil of Forecasting. Frozen soil unfrozen water content and ice content estimation model designed in this paper can accurately estimate the change trends of frozen soil unfrozen water content and ice content. The method is simple. It provides an effective research method to estimate the frozen soil unfrozen water content and ice content; it provides a reference for our researchers to estimate and forecast construction and laboratory tests. It also provides a reference for research in other regions of frozen soil unfrozen water content and ice content, and promotes the improvement and strengthening of research level of frozen soil areas.