ﺷﻨﺎﺳﺎﯾﯽ ﭘﺎراﻣﺘﺮﻫﺎي ﺑﺎرﻫﺎي اﻟﮑﺘﺮﯾﮑﯽ ﺑﺎ اﺳﺘﻔﺎده از ﺳﺎﺧﺘﺎر ﭼﻨﺪ ﻣﺘﻐﯿﺮه ﻣﺒﺘﻨﯽ ﺑﺮ ﯾﺎدﮔﯿﺮي ﻋﻤﯿﻖ

Electrical Load Parameter Identification using Multi-Variant Structure Based on Deep Learning

ﻣﺪلﺳﺎزي ﺑﺎر ﯾﮑﯽ از وﻇﺎﯾﻒ ﺿﺮوري در ﻣﻄﺎﻟﻌﺎت ﺳﯿﺴﺘﻢﻫﺎي ﻗﺪرت ﻣﺤﺴﻮب ﻣﯽﺷﻮﻧﺪ. ﺑﺎ ﺗﻮﺳﻌﻪ ﺳﯿﺴﺘﻢﻫﺎي ﻗﺪرت اﯾﻦ ﻣﺴﺌﻠﻪ ﺑﯿﺶ از ﭘﯿﺶ ﭘﯿﭽﯿﺪهﺗﺮ ﺷﺪه اﺳﺖ. روشﻫﺎي ﭘﯿﺸﯿﻦ ﻣﺪلﺳﺎزي ﺑﺎر داراي ﻋﯿﻮب اﺳﺎﺳﯽ ﻣﺎﻧﻨﺪ اﻟﻒ( ﺣﺴﺎﺳﯿﺖ ﺑﺎﻻ ﺑﻪ ﻧﻮﯾﺰ، ب( ﻋﺪم ﻟﺤﺎظ ﻫﻤﮕﺮاﯾﯽ ﺑﺎرﻫﺎي اﻟﮑﺘﺮﯾﮑﯽ در ﯾﮏ ﺷﺒﮑﻪ، ج( واﺑﺴﺘﮕﯽ ﺑﻪ ﻣﺪل رﯾﺎﺿﯽ، د( ﺑﺎر ﻣﺤﺎﺳﺒﺎﺗﯽ ﺑﺎﻻ و ه( واﺑﺴﺘﮕﯽ ﺑﻪ اﻧﺪازهﮔﯿﺮي ﻣﺤﻠﯽ ﻫﺴﺘﻨﺪ. ﺑﺮاي رﻓﻊ اﯾﻦ ﻣﺸﮑﻼت، در اﯾﻦ ﻣﻘﺎﻟﻪ ﯾﮏ ﺳﺎﺧﺘﺎر ﻣﺒﺘﻨﯽ ﺑﺮ ﯾﺎدﮔﯿﺮي ﻋﻤﯿﻖ ﺗﻮﺳﻌﻪ داده ﺷﺪه اﺳﺖ ﮐﻪ ﻗﺎدر ﺑﻪ ﺷﻨﺎﺳﺎﯾﯽ ﺗﻌﺪاد زﯾﺎدي از ﭘﺎراﻣﺘﺮﻫﺎي ﺑﺎر ﺑﻪﺻﻮرت ﻫﻤﺰﻣﺎن ﺑﺎ ﺳﺮﻋﺖ و دﻗﺖ ﻣﻄﻠﻮب اﺳﺖ. ﺳﺎﺧﺘﺎر ﻃﺮاﺣﯽ ﺷﺪه ﻗﺎدر ﺑﻪ درك ﮐﺎﻣﻞ وﯾﮋﮔﯽﻫﺎي زﻣﺎﻧﯽ ﺑﺮ ﻣﺒﻨﺎي ﯾﮏ ﺳﺎﺧﺘﺎر ﺣﺎﻓﻈﻪدار ﺑﺎزﮔﺸﺘﯽ اﺳﺖ. ﻫﻤﭽﻨﯿﻦ، ﺑﺮاي ﺗﺨﻤﯿﻦ ﺗﻌﺪاد ﻣﺘﻐﯿﺮﻫﺎي زﯾﺎد ﯾﮏ روش اﺧﺘﺼﺎصدﻫﯽ وزن ﺑﺮاي اﯾﻦ ﻣﺪل ﺗﻮﺳﻌﻪ داده ﺷﺪه اﺳﺖ. ﻧﻬﺎﯾﺘﺄ، ﯾﮏ ﺗﺎﺑﻊ ﺗﻠﻔﺎت ﻓﺮﻣﻮلﺑﻨﺪي ﺷﺪه اﺳﺖ ﺗﺎ ﻣﻘﺎوم ﺑﻮدن ﺳﺎﺧﺘﺎر در ﺑﺮاﺑﺮ ﺑﺎ ﻧﻮﯾﺰ را اﻓﺰاﯾﺶ دﻫﺪ. ﻣﻄﺎﻟﻌﺎت ﻋﺪدي ﺑﺮ روي ﺷﺒﮑﻪ 68-ﺷﯿﻨﻪ IEEE ﻣﻮﺛﺮ ﺑﻮدن و ﺑﺮﺗﺮي روش ﭘﯿﺸﻨﻬﺎدي را در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﺗﻌﺪادي از روشﻫﺎي ﮐﻢ-ﻋﻤﻖ و ﻋﻤﯿﻖ را ﻧﺸﺎن ﻣﯽدﻫﺪ.

Electrical load modeling has been considered an essential task in power system studies. With the recent development of power systems, load modeling is becoming more and more challenging. The previous methods on load modeling are suffered from: i) high sensitivity to noise; ii) neglecting the load correlation in a power system, iii) high computational burden, and iv) dependency on the local measurement devices. To address these problems, this paper develops a deep neural network-based structure that can identify a large number of parameters simultaneously with fast performance as well as high accuracy. The designed network can fully understand the temporal features using a gated recurrent neural network-based structure. Furthermore, to provide the ability to estimate a large number of load parameters, a technique to assign the learning weight has been developed. Consequ-ently, to enhance the robustness of the designed network considering noisy conditions, a loss function has been developed in this paper. The numerical results on the IEEE 68-bus system demonstrate the effectiveness and superiority of the proposed network in comparison with several shallow-based and deep-based structures.