In recent years, due to the demand of global sustainable development, renewable energy have attracted increasing attention. Solar energy is prized and praised in all kinds of renewable energies by virtue of availability and cleanliness. Photovoltaic (PV) system, a power generation technology based on the photovoltaic effect of semiconductors to convert solar energy, plays a more and more important role in energy area. The rapid development of all kinds of PV power generation systems has given rise to the need for fine operation and maintenance (O&M) of PV systems, especially large scale PV systems.

As most of the PV modules are exposed to the air and have inclined angle, dust deposits on the surface of modules, which is block to the solar radiation. Whereupon, the PV power generation efficiency is reduced by the dust deposition. In view of this problem, cleaning is a vital part of O\&M for PV systems. For large scale PV systems, cleaning can improve the power generation efficiency significantly, but at the same time, it has to pay a high cost. In addition to innovative cleaning technologies, cleaning scheduling and routing can be optimized based on the historical and forecasting data of power generation, dust deposition and meteorological parameters to reduce the economic loss caused by dust and cleaning. However, at present, cleaning optimization is still under extensive management, whether in industrial application or academic research. Cleaning scheduling in time dimension solely focus on the optimization of fixed frequency or dust deposition threshold; and cleaning routing in space dimension was paid little attention. Considering the benefits and costs of cleaning, to minimize the total economic loss caused by dust, this thesis does research of temporal-spacial cleaning scheduling optimization for large scale PV systems and makes main contributions as follows:

1. The backgrounds of PV systems, dust deposition and cleaning scheduling; research status of dust deposition monitoring, temporal scheduling and route planning; and the challenges that motivate this thesis are introduced.

2. An weakly-supervised data-driven approach: improved Similarity-Based Modeling (SBM) approach for dust deposition monitoring is proposed. This approach increases the performance of the existing approaches in aspects of accuracy, response rate, universality and cost. An experiment of dust deposition effect is conducted. The proposed approach and some other existing approaches are employed to the dust deposition monitoring and verified by the experimental data.

3. A method to design cleaning schedules for PV systems is proposed. Different from the previous studies focused solely on the fixed cleaning intervals, our method designs specific cleaning schedules considering the daily variation, rather than the general situation of power generation and dust deposition. The cleaning schedules are updated everyday in this method, utilising the forecasting accuracy increase with time. The performance of our proposed method operating under various conditions is evaluated in a case study and compared with classical scheduling methods.

4. A hybrid cleaning scheduling policy with periodic planning and dynamic adjustment stages is proposed. Specifically, the periodic planning stage aims for medium-term schedule while the dynamic adjustment stage is tailed for short-term fine-tuning. In the former stage, we show that when the number of cleaning times is fixed, the periodic cleaning strategy is optimal. Moreover, we derive the optimality condition under which the optimal cleaning interval can be determined. In the latter stage, based on the determined cleaning interval, we dynamically adjust the cleaning schedule with the forecast of meteorological parameters, PV power generation, and dust deposition in order to further minimize economic losses. In addition, we take the forecasting uncertainty into account and propose a new custom parameter called risk-taking tendency (RTT) which is able to quantify the risk preference of decision makers and analyze its influence on the scheduling policy. A case study is provided to illustrate the proposed strategy.

5. The PV cleaning route planning problem is modeled and a joint temporal-spacial coordination optimal method for PV cleaning scheduling is proposed. Considering the cleaning capacity limits and a more realistic cleaning process, based on the temporal cleaning scheduling policy, a method to solve the problems of cleaning resources dispatch, cleaning workload allocation and cleaning route planning is presented. We model the problem as a production-driven traveling salesman problem with time-dependent cost (PD-TSP-TC), which is an extended TSP we proposed. A nonlinear 0-1 Integer Programming Problem is modeled and we solve it by genetic algorithm (GA). The total economic loss is reduced further than the methods just considering temporal or spacial optimization.

The thesis concludes with a research recap and future research outlook.