R&D Project 1 – Develop machine learned relief analysis tool

The advancement of technology in remote sensing and image processing enables the analysis of the suitability of sites for the installation of the hydroelectric power plant. These remote sensing techniques, coupled with the machine learning algorithms, enhance the accuracy of the task. The near ocean installation of the hydroelectric power project requires the local scale relief in the topography along which dam can be constructed to increase the kinetic energy of water, because at the mouth of river channels (near the oceans) the velocity of water is significantly dropped and which is not sufficient for the utilization of hydroelectric power generation. The best and recommended type of dams that can be constructed in low lying near ocean areas is hydro pumped storage dams. 

The dam construction is mainly based on several factors, some of which are: Geology, geomorphology of the area, along with its drainage system. These are the main factors that should be observed while selecting the site for the dam construction. Drainage system density, topography, total dissolved solid content, and the curve should also be considered during the site selection for dam construction. Different steps involved in the site selection for the dam are discussed here. In the first step, initial zones are marked that can be suitable for the dam construction. After this geological, geomorphological and climatological factors are mapped by using different softwares. As the data sets are mapped than machine learning techniques, the Analytical hierarchical process and weighted overlay analysis are employed to analyze dam sites. The results from these modern techniques can also be validated by using information from already existing dam sites. 

Machine learning techniques use the set of computational algorithms and statistical models to generate the models without any pre-existing model. The quality of the machine-learned model increases with an increase of inexperience. The machine learning can either be supervised or unsupervised. Supervised learning involves regression and classification problems. On the other hand, unsupervised machine learning deals with clustering problems. Some of the machine learning techniques that are discussed here are unusually named as random forest technique (RF), gradient boosted trees (GBT) technologies, and support vector machine (SVM) technique. 

The random forest technique adopts trees like modeled structures for the prediction and classification by using multiple splitting processes. This technique improves the accuracy of the forecast. The GBT technique also involves a tree-like model that operates in two steps in first step models are generated based on averages, and in the second step, their performance is boosted. The third technique is the support vector machine. It employs the kernel functions to create high dimensional features from the input data. It is best for classification problems. 

Initially, machine learning algorithms are trained by using sample datasets. Modeling can be conducted by using various softwares. One of these softwares is RapidMiner that can be used for modeling. These models are generated by using machine learning techniques. Once the modeling is complete, then the models of different areas are compared, and a suitable site for the dam is then selected accordingly. These locations are later on visited, and the accuracy of machine learning models is cross-checked and verified; hence suitable site for hydroelectric dam can be selected. These machine learning techniques can also be used to check and model other parameters associated with hydroelectricity-generation setup installation, like disturbance of the aquatic ecosystem caused by the construction power generation setup. 

Instead of dams, construction management of local-scale relief can also enhance the kinetic energy of water. The relief studies of the area can be done by using supervised machine learning over the data sets of remotely sensed imageries, e.g., satellite images. In this case, machine learning involves pixel-based analysis techniques using Naïve Bayes classifier. Naïve Bayes classifier is the simple classification tool that is based on Bayes theorem. This type of machine learning tool compares the image pixel to pixel and reveals the information that can be useful for the hydroelectricity generation project construction. So, machine learning is modern technology that can be used to instruct the computing machines to work precisely and rapidly to perform specific tasks. 

Milestone Objectives

The project comprises 3 milestones, where: 

Milestone 1 is divided into 4 milestones: 

Milestone 1A – Identify potential data sources, evaluate and determine best dataset 

Milestone 1B – Setup storage platform and develop access and disaster recovery protocols 

Milestone 1C – Download dataset, perform data cleans 

Milestone 2 is divided into 3 milestones: 

Milestone 2A – Develop R-based rule for analysing global dataset for suitable hydroelectric power locations. 

Milestone 2B – Determine further critical metrics, indicators and factors to be considered by machine learning algorithm to further improve site suitability assessment 

Milestone 2C – Programme machine learning algorithm to refine analysis tool, with iterative improvement of outcomes 

Milestone 3 is divided into 3 milestones: 

Milestone 3A – Run machine learning solution across global dataset 

Milestone 3B – Generate graphical representations and animations of identified sites 

Milestone 3C – Generate final report of prioritised identified locations with supporting data