

















摘" 要: 為檢測(cè)電力系統(tǒng)中的網(wǎng)絡(luò)攻擊行為,文中提出一種基于電力數(shù)據(jù)采集與監(jiān)視控制(SCADA)系統(tǒng)的攻擊檢測(cè)方法,探討了機(jī)器學(xué)習(xí)方法作為檢測(cè)電力系統(tǒng)攻擊的可行性,并評(píng)估了其性能,討論了機(jī)器學(xué)習(xí)模型作為攻擊檢測(cè)方法的意義。此外,還提出一種基于機(jī)器學(xué)習(xí)的投票分類(lèi)模型(RES),其由RF、ET和SVM三種基本分類(lèi)器構(gòu)成,使用投票分類(lèi)中的軟投票方法,并且考慮了基本分類(lèi)器的權(quán)重對(duì)投票分類(lèi)模型的影響。通過(guò)在密西西比州立大學(xué)和橡樹(shù)嶺國(guó)家實(shí)驗(yàn)室的電力系統(tǒng)攻擊數(shù)據(jù)集上進(jìn)行實(shí)驗(yàn)和分析,結(jié)果表明,與其他方法相比,RES模型在電力系統(tǒng)的攻擊檢測(cè)方面準(zhǔn)確率得到大幅提升,在電力系統(tǒng)攻擊數(shù)據(jù)集上的二分類(lèi)準(zhǔn)確率達(dá)到了98.40%,能夠準(zhǔn)確地檢測(cè)電網(wǎng)中的網(wǎng)絡(luò)攻擊行為。
關(guān)鍵詞: SCADA系統(tǒng); 投票分類(lèi)模型; 電力系統(tǒng); 網(wǎng)絡(luò)攻擊; 機(jī)器學(xué)習(xí); 入侵檢測(cè)
中圖分類(lèi)號(hào): TN915.08?34; TP391.4; TP769" " " " " 文獻(xiàn)標(biāo)識(shí)碼: A" " " " " " " "文章編號(hào): 1004?373X(2025)04?0018?06
Power system attack detection technology based on SCADA and voting classification model
GENG Zhenxing, WANG Yong
(School of Computer Science and Technology, Shanghai University of Electric Power, Shanghai 200120, China)
Abstract: In order to detect cyber?attack behaviors in power systems, a method of attack detection based on power SCADA (supervisory control and data acquisition) system is proposed, the feasibility of machine learning method for detecting power system attacks is discussed and its performance is evaluated, and the significance of machine learning model as an attack detection method is discussed. The machine learning based voting classification model (RES) is proposed, which is composed of three basic classifiers: random forest (RF), extra tree (ET), and support vector machine (SVM), the soft voting method in voting classification is adopted, and the influence of the weight of the basic classifier on the voting classification model is considered. Through experiments and analysis on the power system attack dataset from Mississippi State University and Oak Ridge National Laboratory, the results show that in comparison with other published methods, the RES model has substantially higher accuracy in attack detection in the power system, and the binary classification accuracy on the power system attack dataset can reach 98.40%, which is capable of accurately detecting cyber?attacks in the power grid.
Keywords: SCADA system; voting classification model; power system; cyber attack; machine learning; intrusion detection
0" 引" 言
隨著電力系統(tǒng)的數(shù)字化轉(zhuǎn)型和智能化發(fā)展,自動(dòng)控制、網(wǎng)絡(luò)通信、人工智能等技術(shù)被廣泛地應(yīng)用于新型電力系統(tǒng),并在新型電力系統(tǒng)中承擔(dān)著重要作用[1]。電力系統(tǒng)的網(wǎng)絡(luò)化程度越來(lái)越高,電力系統(tǒng)中的傳感設(shè)備、控制設(shè)備等連接到互聯(lián)網(wǎng),使得系統(tǒng)容易受到來(lái)自網(wǎng)絡(luò)的攻擊[2]。2015年烏克蘭大停電是一個(gè)典型案例,其主要原因是黑客通過(guò)入侵計(jì)算機(jī)設(shè)備向物理設(shè)備發(fā)送大量惡意跳閘指令,造成了數(shù)小時(shí)的大面積停電事故[3]。2003年加拿大和美國(guó)部分地區(qū)遭受攻擊而停電,導(dǎo)致經(jīng)濟(jì)損失上百億美元,這表明對(duì)電力系統(tǒng)的攻擊可能會(huì)對(duì)國(guó)家經(jīng)濟(jì)造成巨大影響[4]。……