109年度跨部會化學物質資訊服務平台(化學雲)應用計畫
| 中文摘要 | 本年度資料擴增及系統維護方面，擴增介接3個系統資料與更新1個系統資料，持續維護9部會49個化學物質管理資訊系統定期拋轉。新增資料交換稽催自動比對機制，並發送722封稽核通知信件。完成「使用進出口申報通關資料管理要點」及「使用財稅資料管理要點」管控機制; 開立 246 個化學雲帳號; 回饋消防署及毒化物系統廠商座標資料錯誤共294筆以及回饋毒化物系統廠商災防圖資錯誤共8家。 部會需求開發方面，提供臺北市消防局與新北市消防局Web Service 介接工廠危險物品申報網資料; 與消防單位研商後建立廠商運作背景資訊摘要版格式及化學物質優先呈現排序原則。並分別在化學雲平台提供廠商運作背景資訊完整版、摘要版跟自選版查詢功能與提供3個應用程式介面(API)予消防署派遣APP介接廠商運作背景資訊摘要版。 大數據分析方面，本計畫使用貝氏網路模式、分類與迴歸樹及隨機森林三方法建構單一的化學物質運作廠商風險模型，比較三種模式績效優劣外，也使用集成學習（ensemble learning）技術以精進風險分析模式的建構，集成模式可產出較單一模式更佳的分類正確率。使用人工智慧技術進行國內外化學物質新聞事件自動化分類整理新聞，並完成新聞事件輿情分析，提供主管單位瞭解新聞討論熱度。建立6類食品製造廠商依類別分析化學物質流向及風險之方法，便於關注各類廠商特定的食品添加物或歷史事件的毒性化學物質所造成的風險，並據以建立各類食品廠商風險模型。 其他分析工作方面，完成區塊鏈應用於化學物質管理的可行性評估報告，經過參考「世界經濟論壇」所建議的初步評估方法，所得的初步評估結果為正面可行，建議仍須進一步小規模試作，以實際測試驗證。此外，完成災防圖資PDF檔案辨識切割程式開發，並據以提出災防圖資標準格式建議方案。 | ||
|---|---|---|---|
| 中文關鍵字 | 化學物質、跨部會合作、警示與快報 | ||
基本資訊
| 專案計畫編號 | 經費年度 | 109 | 計畫經費 | 14300 千元 | |
|---|---|---|---|---|---|
| 專案開始日期 | 2020/03/09 | 專案結束日期 | 2020/12/31 | 專案主持人 | 李曜全 |
| 主辦單位 | 化學局 | 承辦人 | 林桂如 | 執行單位 | 景丰科技股份有限公司 |
成果下載
| 類型 | 檔名 | 檔案大小 | 說明 |
|---|---|---|---|
| 期末報告 | 109化學雲期末全文上傳.pdf | 10MB |
2020 Inter-ministerial Chemical Substance Information Service Platform (Chemical Cloud) Application Project
| 英文摘要 | In terms of data augmentation and maintenance this year, we’ve maintained 49 systems under continuous running, registered 246 ChemiCloud accounts, and expanded access to 3 data sources. We’ve developed auto data form checking and sent 722 notification mail. Furthermore, we’ve given feedback of 294 mistakes of vendors’ location coordinates to two systems. In terms of the development and interface of department requirements, we’ve developed web service functions for Taipei city fire department and New Taipei city fire department to use factory hazardous materials declaration network information. Moreover, we discussed with several fire departments to identify the form of quick report of chemical substance for disaster rescue and developed three kinds of API for National Fire Agency, Ministry of the Interior to get the quick report from ChemiCloud. In terms of big data analysis, we’ve used Bayesian network model, classification and regression tree and random forest to construct a single chemical substance operation manufacturer risk model, compared the performance of the three models, and also uses ensemble learning technology to refine the risk analysis model. The ensemble learning model can produce a better classification accuracy rate than the single model. In addition, use artificial intelligence technology to automate the classification and sorting of news about chemical substance news events at home and abroad. We have also completed public opinion analysis of news events and provided competent units to understand the popularity of news discussions. Furthermore, the establishment of a method for six types of food manufacturers to analyze the flow and risk of chemical substances by category is convenient for focusing on the risks caused by specific food additives of various manufacturers or toxic chemical substances of historical events, and establishing risk models for various food manufacturers based on them. In terms of other analytical work, the feasibility assessment report of the application of block-chain to chemical substance management was completed. After referring to the preliminary assessment method recommended by the World Economic Forum, the preliminary assessment results obtained are positive and feasible. It is recommended that further small-scale trials to be implemented and verified with actual test. In addition, we completed the development of the identification and cutting program for PDF files of disaster prevention maps, and put forward suggestions on the standard format of disaster prevention maps based on them. | ||
|---|---|---|---|
| 英文關鍵字 | Chemical Substance, Inter-Ministerial Cooperation, Alert and Quick Report | ||