Meteorological conditions such as air density, temperature, solar radiation etc. strongly affect the power generation from solar, and thus, the prediction and estimation process should consider weather conditions as critical inputs. The nature of weather forecast is highly unpredictable, so many applications use meteorological data from in-place on-site sensors to add to the forecast and some use complex networks with complicated mapping. The in-situ sensor approach and dense mapping methods, however, present several drawbacks. First, the use of sensors give rise to extra operational, installation and maintenance cost. Second, it requires significant amount of time to capture and accumulate data for various occasions and scenarios, and in addition, sensor itself can be the cause of error measurements. The complex methods are computational inefficient and may present suboptimal convergence. This paper presents a sensorless solar output power forecasting based on historical weather (publicly available from met office) and PV data. The algorithm uses simple to implement neural networks with few neurons and hidden layers for its training and allows for day a head forecast. The proposed methodology presents a guideline on how to select the relevant data from weather and how it affects the accuracy and training time of neural network.