Instrumentation Deep Knowledge Services Based On Cloud Framework

As China has been devoting greater efforts to boost the independent innovationof instrumentation， the supporting work of information service in this realm isenjoying inscreasing attention.The present information service platform has gainedsignificant effects in terms of information share and resource utilization efficiency.Yet，capabilities in the applicability to the instrumentation sphere，service depth and dataprocessing remain deficiencies.Aiming at the above problems，Ministry of Scienceand Technology has launched Innovation Method Special Project which integrates thepractical demand of instrumentation and its relevant information resource，providingit with knowledge service.On the basis of this project，this article conducts atheoretical，technical and methodological research on the instrumentation knowledgeservice，established its system and integrates its pertinent information resource，whichwill offer comprehensive and efficient information service for instrumentationindependent innovation.Instrumentation knowledge service mainly incorporates contents from threedimensions：The first is the construction of knowledge system framework，providingknowledge service with data structural model;The second is the technology of miningand organizing knowledge content，offering knowledge service system frameworkprecise complete data resource to enrich structure content;The third is distributedcomputing framework with efficiency，supporting the data processing and miningfunction of the entire knowledge service system.Therefore，this article subsequentlyperforms the research on the above three aspects.(1) Research on Knowledge Service System in Scientific InstrumentationThis paper builds the framework of instrumentation knowledge service. Itorganizes related literatures and information resources of instrumentation from theview of study， research，manufacture and applications. It also clears the structure ofknowledge information and the correlation and mapping among knowledge contents.The article suggests building universal and extendible instrumentation knowledgedatabase model and equipment data model. According to the form of framework design， processed and analyzed knowledge information is filled into the frameworkto form the ultimate systemic and effective instrumentation knowledge service system。(2) Application of Deep Learning in Knowledge ServiceIn the process of knowledge analysis and extraction， natural languageprocessing and image recognition technologies need to be applied. Common technicalmeasures rely on artificial interference and prior data， and the performance of highlevel feature representation is inadequate. This article introduces the deep learningtechnology into the data processing of knowledge service， and improves thetraditional deep learning model according to different application demands in order toenhance the training performance; it brings about an enhanced denoising auto-encodeneural network model according to the positioning and recognition of words in naturalimages， improving the recognition ability of neural network model by adding theneuron interaction mechanism in the auto-encode stage， resulting in an competitiveadvantage in generality and accuracy; it conducts word segmentation on the Chinesetexts in the information resources， executes training according to the deep neuralnetwork structure， improving the accuracy of word segmentation compared withtraditional technologies; it raised an improved hierarchical log-bilinear deep neuralnetwork statistics language model according to the identity of named entity ofChinese literature in scientific instrumentation， combines the unsupervised andsupervised studies， utilizes the multi-level restricted Boltzmann machines traningtext word vectors， inputs the trained word vectors in the feedforward neuralnetworks for supervised training， and thus finishes the machine learning of Chinesetext， effectively improving the learning ability of language models.(3) Application of Cloud Computing in Knowledge ServiceIn order to increase the efficiency of the large-scale parallel computing ofknowledge information， many parallel computing methodologies have been putforward， but there is still room for improvement in the execution efficiency andexpansibility. This article establishes the cloud computing framework based on largeclusters， and improves the platform’s computing ability with distributed memorycomputing; it designs a distributed memory computing framework forcomputer-oriented clusters，establishes Data fragmentation processing and multi-taskscheduling mechanisms， and thus enable the computations of model parameters andneural network elements to run in parallel in the environment; it avoids the impact of disk I/O on the training rate，conducts training with asynchronous parallel computingfor the deep belief nets model， and prevent over-fitting by the dropout method，effectively improving the training rate of deep neural network.