 Very good afternoon to all of you. I am giving a view on real-time wave forecasting. I have used using various soft computing tools like artificial neural network, model trees and genetic programming. This wave analysis incorporates the procedure of calculating and diagnosing wave conditions and out of this significant wave height is of immense importance. Traditionally this forecasting was done using by transforming the wind on information to wave information with various empirical equations like SMB, Hesselman's CEM for later numerical models like VAM and SWAN came into existence. With the deployment of data voice around the country's coastline, various time series modeling using stochastic models like ARMA, ARIMA and recently ANN came into use. These time series based models are very recent. I have taken two sites first from the Gulf of Mexico which can be easily downloadable and exhaustive data is available so I have chosen this site. Out of these I have taken four sites, two from the deeper location and two from the shallow location. For the station 42001 I have taken a small data set that is from the deeper location and for the other three stations I have taken an exhaustive set of 12 years early significant height data. Various innovative schemes I have tried for input output combinations where various training schemes, various algorithms to get the best model for the forecasting. When we go for lower lead time predictions like for 3 hours, 6 hours, the predictions from all these models were like quite comparable but when we go for higher lead times like 12 and 24 hours like that, the results deteriorated and up till now so many trials have been done to improve these results. With the study I found that a method in which previous selective values like 3rd, 6th, 12th, 24th I am taking and the predictions is one lead hour like 3rd hour at a time. So the model will be having four inputs like previous 3rd, 6th, 12th and 24th and the output will be one that is 12th, 24th like that. So that I have named it, we give it as a jumping forecast method and these are the results with various error measures of correlation coefficient, truth mean square error and mean absolute error. With this background to the other three stations which I have taken for 12 years the data was exhaustive. So first I did the statistical study and saw that quite a large variation is there in the data even between the months. So 12 month wise networks were developed for each of the boy stations that is 4, 2, 0, 3, 6, 4, 2, 0, 3 and 4, 2, 0, 1, 9 and again various method, various iterations of the input output combinations were done to improve the higher lead time results of 12 and 24. Then going through the data we saw that there exist a lot of gaps in the data. This is a particular example of it. So we try to see how is the percentage of the missing values in the data and we saw that the missing values varied from 6.57 to 42 percent. So it was felt to fill these gaps using some appropriate methods. This is quite logical but this has not been applied to this wave data. So we tried to fill using some appropriate data and we tried various methods available in the literature. Out of this we found that temporal regression was applied to fill these gaps when the gaps were small like for a week or a day but when the gaps were high we used the spatial mapping that is using the other stations data nearby to fill that data and we can see lot of improvement with the help of these error measures in the correlation coefficient before and after the filling. This is for station 4, 2, 0, 3 and 4, 2, 0, 1, 9. We can see the picture. The left hand side is for scatter plot and right hand side is the time series plot for 12 hour and 24 hour for station 4, 2, 0, 3 before filling the gaps and see we can see the improvement in the results when we fill the gaps earlier to the data before applying to the network. The picture would be more clear with the graph. These dotted lines are after filling of the gaps and the full line is for before filling the gaps for 12th and 24th hour data. Again the data mining tools like MT and GP were applied to see if the improvement in the largely times could be possible. These are the results for the three stations with various error measures. The picture would be more clear and when we see that all the three tools that is ANN, MT and GP are compared when all three error measures are taken into account we see that the MT, GP are working parallel or equal to ANN. So, it would be dependent upon the choice of the user to use the tool. Again we took a Indian data site out of which this is the deeper location and these are the two shallow locations. The data set belong to from year 1998 to 2004 and again when we studied the data we could see the gaps and hence these gaps were filled using temporal regression and spatial mapping according to the number of data gaps present and using the jumping forecast method we were able to get very good results even for 24 hour higher lead time results the error measure if we see the error measures it is quite comparable to the lower lead time results which has not been attained by any of the researcher till now. So, we were able to with the these innovative schemes and applying other processing and all we were able to get very good results. These are the scatter plots and time series plots for the stations so for 12th and 24th hour. Again for with when we got so good good results for higher lead times we tried to apply for higher lead times of 48 and 72 hour and we were able to get quite good results for these three stations. These are the time scatter plot and time series plot for the station SW4 for an example. Again data mining tool of model tree was applied to these data and these are the results for the three stations DS1 SW4 and SW2 with the various lead time hours. This is a genetic programming was applied a comparative comparison can be seen for the three tools that is ANN, MT and GP when these are highly lead time were taken and we can see that they match each other when all the three errors are taken together. In all the forecasting we saw that when it comes to peaks there is a little deviation. So, we try to improve the forecasting scheme using some hybrid models. In this hybrid model we are taking the input data and for each of the lead time hour of 3, 6, 12 and 24 we are taking these output again as a input along with the observed values using ANN and we get for higher lead time as a improved peak prediction model. For various lead times the number of input changes accordingly and with this scatter plot we can see that the scatter reduces when we are using the hybrid model. This is for ANN model comparison for 12th and 24th and this is for 48 and 72 hour. In case of 12th and 24th the results were comparable but when it comes to higher it deteriorates. With all these model formed we try to integrate through a graphical user interface so that a user can use it. An IoT deploy has deployed its boys around the country's coastline and collects the wave information. This wave information they disseminate to various other users like port and harbor other things. So this is the main window of the graphical user interface. In this for all the sites various models has been integrated. So the person who is there in an IoT has to just click the station in which he wants to get the forecasted data, load the previous 24 hours data and run. When he is clicking the run the graphical user interface which we have developed takes care whether if the data is having any missing values. It fills it up using temporal regression or spatial mapping according to the number of gaps present and then run the different models for 3, 6, 12 and 24 hour to get the forecast and give the forecast in the form of a graph or and even the prediction values and similarly if he wants to go for the other station that is also integrated in the same graphical user interface. So with this I will come to the conclusion that various larger sample sizes necessary to get a better results from our study we have seen and for very long duration data monthly forecasting models could be a better choice. Also a simple forward network using LM is good enough and the best method for training we have found is jumping forecast method. Also statistical homogeneity in the data helps in filling in of gaps in the measurements by appropriate method certainly improves the results. For 24 hour lead time it was had remained a difficult problem for several last years. So this has been successfully tackled in the Indian locations by innovative learnings methods and further by improper appropriately filling the gaps. However it has a limitation that the value by wave values were in the range of 2.5 to 6 meters means for higher values it deteriorates and also applications of genetic programming and model trees were almost untried so far for ocean engineering problems and time series prediction in particular. So this has been applied for larger lead times and they are attractive and the choice may of a particular model should will be dependent upon the convenience. The use of technique of GP and MT is yet to be reported widely in ocean engineering and this work indicates that they look very promising for the future applications. The advantage of having large number of controlling mathematical functions like ANN and hence more flexibility in data mining and the built-in capacity to handle a large amount of data are attractive features of GP as well as MT. The technology in terms of wave forecasting models developed in this study has been transferred for its actual implementation to government of India's national data boy collection namely NIOT. This is done by constructing a graphical user interface for online wave forecasting for the benefit of users. So this way we can make the application and we can see how it can be implemented. Thanks. No that have not that have taken non-linear regression model I've developed not any software I've developed by myself means using this Excel I have developed this non-linear regression equation. No pre-processing in the sense I have filled the gaps in the data that only the pre-processing I have done and in the data in Gulf of Mexico data I've done a month-wise data means for January the network was different likewise 12 different models were there for different months. That pre-processing is there. In Indian data actually the dataset was very small which was available so we have not tried for that in Indian data but in Gulf of Mexico data we have developed 12 months-wise networks. Radial basis also have tried. Yeah gradient descent have tried. Means all possible which I could find I have tried and then I've reported the best one which is applicable to my problem.