山崎 高弘

This paper describes a method of readability assessment for web documents. Readability is the ease in which text can be read and understood. We hypothesize that the readability is determined whether a reader can easily grasp text structures. The impression and the complexity of text are significant factors. We extract the features about impression and complexity from a plain text and additional data such as HTML tags.<br>In order to compare effect of extracting features, we are assessing readability rank by machine learning. We conduct 5-fold cross validation for each domain, and calculate the root mean squared error between the actual rank and the estimated rank. The cross validation experiments confirm that the performance of our method is high measured. It shows effectiveness of extracting features about the impression and the complexity for readability assessment.

In this paper, we present and discuss a simple estimation method for the life expectancy of optical disks using resampling (bootstrap) statistical analysis in order to roughly estimate the lifetime of archival-grade disks. The performance of this method was examined experimentally using the Eyring acceleration test with four stress conditions. Results demonstrated the capability of this bootstrap analysis for estimating life expectancy compared with conventional statistical analysis using the Eyring acceleration test.

Recently, Schwartz and Etzion proposed several efficient code construction methods of two-dimensional linear codes capable of correcting a single arbitrary cluster of errors of size b. Some of their methods, however, impose the constraint that the degree m of GF (2(m)) must be even. In this correspondence, we investigate whether such Schwartz-Etzion's methods are applicable to the case in is odd. We provide necessary and sufficient conditions for the original Schwartz-Etzion's methods to be used for constructing two-dimensional cluster-error-correcting linear codes for odd m, and show that the methods are available extensively.

Bluetooth is best suited for an exhibition guidance system, because of its low-power consumption, enabling visitors to use their own mobile phones as browsing devices. High user mobility assumed in an exhibition hall needs an efficient handoff mechanism, which is required to establish connection to the most appropriate access point (AP) with minimized duration and maximized interval. This paper presents an efficient handoff method, by which a mobile device predicts its movement and connects to the most appropriate AP in the moving direction without inquiry. The prediction is based upon location estimation as well as view history. A visitor is considered to be in the location of an exhibit when he/she requests the exhibit's info. The simulation results show the proposed method outperforms the existing methods in data rate and active mode rate.

In this research, a pre-download method in the information service system using Bluetooth is proposed. The pre-download method is to determine the information to the visitor's mobile in advance by using the movement prediction of the visitor. By using this method, the visitor can browse the information smoothly.<br>This method is evaluated by simulation in various room-layouts and walking routes. More than 92% of requested contents are pre-downloaded in case the memory size is large enough to download. The prototype of the information service system installed the proposed method has been developed. By using this system, the effect of the method is confirmed.

Qualitative reasoning is one of the fault detection and diagnosis methods and is an effective way for complex systems. Especially, in the stochastic qualitative reasoning, the probabilistic process is used for state transition of the qualitative models with probabilities. In this paper, the stochastic qualitative reasoning is applied to a modelling method of human qualitative thinking process based on the instrumentation diagram and the control information in building air conditioning systems. However, several subjects remain to be solved for practical uses. This paper proposes real-time simulation using the stochastic qualitative reasoning as a way of practical uses for fault detection and diagnosis of VAV (Variable Air Volume) system that is one of air conditioning systems.

空調システムの異常検知・診断技術を開発するにあたり,フィールドデータを収集して,実際に起きている異常を調査・分析することは重要である.そこで,ある大規模建物の空調システムを対象として,ほぼ全部の空調機系統のコントローラより計測,設定データを収集してグラフ化し,コントローラの制御情報を前提として,現場の自動設備管理者と一緒に評価を実施して,得られた知見を基本症状,病状,対策リスト,基本チェックフローとしてまとめた.この結果は,今後,同じような空調システムの現場で再評価し,有効な診断アルゴリズムを開発することによって自動点検の手法を構築するために有用と考えられる.

空調システムの異常検知・診断システムの研究開発は,IEA-Annex 25を契機に始まったといえる.その後,空気調和・衛生工学会蓄熱最適化委員会のFDD-WGやIEA-Annex 34として研究が進んでいる.確率的定性推論は,計装図と制御内容を基に人間の定性的な思考過程をモデル化し,少ない情報で異常検知・診断を行う手法である.これをVAVシステムで発生させた故障状態データに適用したリアルタイムシミュレーション内容について報告する.今後,異常検知・診断を自動的に行う手法として,確率的定性推論を利用した異常検知・診断アプリケーションが期待される.