نوع مقاله : مقاله پژوهشی

نویسندگان

دانشکده‎ مهندسی عمران، دانشگاه تربیت دبیر شهید رجائی، تهران، ایران

چکیده

پیشینه و اهداف: سیلاب‌ها، هر ساله موجب بروز خسارات قابل توجهی در سراسر جهان منجر می­شوند. پیش‌بینی به موقع و دقیق آن‌ها می­تواند میزان خسارات جانی و مالی را به طور چشم‌گیری به حداقل برساند. در سال­های اخیر، مدل­های یادگیری ماشین متعددی به‌منظور پیش‌بینی سیل مورد استفاده قرار گرفته­اند؛ به‌طوری که نتایج آن‌ها حاکی از عملکرد بهتر این مدل­ها نسبت به مدل‌های آماری کلاسیک می­باشد. با این حال، این مدل­ها ویژگی­های مکانی که منجر به ایجاد و تقویت سیلاب­ها می­شوند، مدنظر قرار نمی­دهند. با استفاده از مدل حافظه طولانی کوتاه مدت پیچشی (ConvLSTM)، پیش‌بینی سری­های زمانی با ویژگی­های مکانی و اطلاعات هیدرولوژی ترکیب می­گردد. بدین ترتیب، الگویی جدید از ترکیب عوارض مکانی و پیش‌بینی سری­های زمانی با هدف استخراج خصوصیات زمانی- مکانی طراحی شده است تا چالش اصلی در پیش‌بینی سیل، یعنی ترکیب داده­های مکانی با داده­های هیدرولوژی را تا حد امکان حل کند.
روش‌ها‌: برای تحلیل جامع خصوصیات زمانی- مکانی بارش، ویژگی‌های مکانی را با تحلیل­های سری زمانی ادغام کردیم. به همین منظور از مدل ConvLSTM استفاده گردید که ورودی­های آن شامل طول جغرافیایی، عرض جغرافیایی، ارتفاع، بارش، دبی و سایر اطلاعات ایستگاه­های اندازه‌گیری است. ConvLSTM یک مدل پردازشی سری­های زمانی است که ویژگی­های مکانی را استخراج می­کند. برای دستیابی به پیش‌بینی مکانی-زمانی، از ConvLSTM به عنوان یک بلوک پایه استفاده گردید تا بتوان با طراحی یک شبکه متراکم، ویژگی‌ها را لایه به لایه استخراج کرد تا پس از نگاشت آن‌ها، عمل پیش‌بینی انجام شود. در مرحله شبیه سازی، اندازه بچ و تعداد اپک­ها به‌ترتیب 64 و 30 انتخاب شد. همچنین، از یک شبکه پیچشی سه لایه با تعداد هسته­های 1، 8 و 32 و تعداد عصب­های 20، 40 و 80 در هر لایه به عنوان مدل نهایی استفاده گردید.
یافته‌ها: با تحلیل نتایج، مشخص شد که دقت پیش‌بینی با افزایش زمان بهتدریج کاهش می­یابد. با این حال، هنگامی که زمان پیش‌بینی 10 ساعت پیش از رخداد سیلاب باشد، دقت پیش‌بینی از زمان‌های دیگر کمتر است. علت این است که هنگامی که زمان پیش‌بینی افزایش یابد، میزان اطلاعات دقیق کمتر خواهد بود که منجر به حصول دقت کمتر در یادگیری مدل و در نتیجه کاهش دقت پیش‌بینی می­شود. برای حل این مسأله باید عمق شبکه افزایش یابد که این امر موجب افزایش زمان مدل­سازی می­گردد که نشان از اهمیت مصالحه میان دقت مورد انتظار و زمان پردازش است.
نتیجه‌گیری: ارزیابی نتایج، نشان داد که مدل ConvLSTM قادر است نتایج پیش‌بینی قابل قبولی، به‌ویژه در زمآن‌های کوتاه مدت، ارائه دهد و این مدل یک ابزار مناسب برای پیش‌بینی­های سری زمانی است. با وجود این‌که مدل ConvLSTM، عملکرد قابل ملاحظه­ای برای پیش‌بینی­های کوتاه مدت به‌دست آورد اما هنوز هم برخی محدودیت­ها از جمله پیش‌بینی بلند مدت سیلاب بر پایه داده­های سری زمانی وجود دارد. از جمله محدودیت­های دیگر، می­توان به پیچیدگی و وابستگی مدل ConvLSTM به تعداد نمونه­های آموزشی اشاره کرد که به تبع آن مدل­سازی دقیق­تر نیازمند جمع‌آوری داده­های بیشتر است. بدین ترتیب، در مناطقی که تعداد نمونه­ها برای مدل­سازی کافی نباشد، ممکن است دقت مدل پیش‌بینی تحت تأثیر قرار بگیرد.

کلیدواژه‌ها

موضوعات

عنوان مقاله [English]

Using Convolutional Long Short-Term Memory Model for Flood Prediction in Golestan Province, Iran

نویسندگان [English]

  • S. Ahmadi
  • A. Nabizadeh

Faculty of Civil Engineering, Shahid Rajaee Teacher Training University, Tehran, Iran

چکیده [English]

Background and Objectives: Every year, floods cause significant damages around the world. Timely and accurate prediction can significantly minimize the amount of human and financial losses after flood. In recent years, several machine learning models have been used to predict floods; So that their results indicate the better performance of these models compared to classical statistical models. However, these models do not take into account the spatial features that lead to the creation and strengthening of floods. Using convolutional long-short-term memory model (ConvLSTM), time series prediction is combined with spatial features and hydrological information. Therefore, a new model of combination of spatiotemporal prediction has been designed with the aim of extracting spatiotemporal features in order to solve the main challenge in flood prediction, which combine spatial data with time series hydrological data as much as possible.
Methods: In order to comprehensively analyze the spatiotemporal features of precipitation, we integrated the spatial features with time series analysis. For this purpose, the ConvLSTM model was used, whose inputs include longitude, latitude, altitude, precipitation, discharge and others gathered by ground stations. ConvLSTM is a time series processing model that extracts spatial features. To achieve spatiotemporal prediction, ConvLSTM was used as a basic block so that features can be extracted layer by designing a dense network, so that after mapping them, prediction can be performed. In the simulation stage, the batch size and the number of epochs were selected as 64 and 30, respectively. Also, a three-layer convolutional network with the number of kernels 1, 8 and 32 and the number of neurons 20, 40 and 80 in each layer was used.
Findings: By analyzing the results, it was found that the prediction accuracy gradually decreases with the increase of time. However, when the prediction time is 10 hours before the flood event, the prediction accuracy is lower than other times. The reason is that when the prediction time increases, the amount of accurate information will be less, which leads to less accuracy in learning the model and as a result, the prediction accuracy decreases. To solve this problem, the depth of the network should be increased, which increases the modeling time, which shows the importance of trade-off between the expected accuracy and processing time.
Conclusion: In conclusion, the ConvLSTM model is able to provide suitable prediction results, especially in short-term times, and this model is a suitable tool for time series prediction. Even though the ConvLSTM model achieved a remarkable performance for short-term prediction, there are still some limitations, including long-term flood prediction based on time series data. Moreover, the complexity and dependence of the ConvLSTM model on the number of training samples can be mentioned. Therefore, more accurate model requires the collection of more data in this model. Thus, in regions with limited number of samples, the accuracy of the prediction may be affected.

کلیدواژه‌ها [English]

  • ConvLSTM model
  • Flood prediction
  • Hydrology
  • Spatiotemporal features

COPYRIGHTS 
© 2023 The Author(s).  This is an open-access article distributed under the terms and conditions of the Creative Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) (https://creativecommons.org/licenses/by-nc/4.0/

doi: 10.1016/j.scitotenv.2019.01.021
doi:10.3390/w12041213
https://hal.science/hal-03164840
https://doi.org/10.5194/nhess-20-1149-2020
https://doi.org/10.5194/nhess-21-1179-2021
https://doi.org/10.3390/w11071387
https:// 10.1109/ACCESS.2021.3065939
https://10.1109/ACCESS.2020.2994050
https://doi.org/10.21595/jve.2022.22271