Research On Photovoltaic Disaggregation And Incentive-based Demand Response Baseline Load Of Behind-the-meter System

With the depletion of fossil energy and the continuous deterioration of ecological environment,it has become the consensus for countries all over the world to accelerate the renewable energy dominated power systems.In order to solve the stability problems that may occur when a high proportion of renewable energy is connected to the grid,realize the localized renewable energy consumption and promote the active participation of customers in energy regulation,many countries,including the United States and Italy,have started to implement the behind-the-meter（BTM）system that integrates large numbers of distributed renewable energy with the public meter as the physical divider and the customer side as the core.However,under the net energy metering of BTM system,only the net load information coupled with load demand and renewable energy output is presented to the outside world.This will cause a lack of energy information on the front side of the meter,making it difficult to implement accurate energy forecasting,formulate optimal demand response plans and a series of other problems.In this thesis,the author proposed a BTM system photovoltaic generation disaggregation technique to disaggregate the net load for a BTM system containing distributed photovoltaic generation devices,and conduct a study related to energy prediction and incentive-based demand response baseline load based on the disaggregation results.The main contributions are as follows:（1）The BTM system photovoltaic generation disaggregation technique is summarized and a deep learning based photovoltaic generation disaggregation method is proposed to address the possible defects of existing disaggregation techniques,which is free from the dependence of most photovoltaic generation disaggregation models on the establishment of energy proxies sites and photovoltaic system model construction.In the modeling process,firstly,the net load data are matched and filtered according to the similarity of electricity consumption behavior and the similarity of irradiance to extract photovoltaic generation difference features and load demand difference features.Then,the extracted features are used as supervised targets,and the photovoltaic generation difference fitted model and load demand difference fitted model are constructed by combining the meteorological data with a deep long short-term memory network.Finally,the photovoltaic conversion characteristics are used to transform the photovoltaic generation difference fitted model into photovoltaic generation disaggregation model.The experimental results show that the proposed BTM photovoltaic generation disaggregation method has higher disaggregation accuracy than the benchmark algorithm when only historical net load data and meteorological data are required.（2）To address the problems that the proposed initial BTM photovoltaic generation disaggregation method is difficult to account for the time-shifted characteristics of irradiance and poorly interprets electricity consumption of customers behavior,a behind-the-meter photovoltaic generation disaggregation method based on cross-iteration update strategy is proposed.Firstly,the solar trajectory features are added as an additional input feature in the modeling stage of the generation difference fitted model,which can explain the time-shift phenomenon of the irradiance.Secondly,based on the idea of load forecasting,the historical load demand disaggregation value is introduced as an additional time series feature in the construction of the consumption difference fitted model to solve the problem that meteorological data are difficult to effectively explain the changes of load demand behavior of customers.Finally,a cross-iteration update strategy is proposed to realize the interactive update of the model by iteratively correcting the supervisory objectives of mutual fitted models through the outputs of the constructed photovoltaic generation difference fitted model and load demand difference fitted model.The experimental results show that the difference modeling property of the proposed method drives the model extremely robust and still has good disaggregation performance in the face of long-term stability problems caused by uncalibrated meters.By utilizing its own net load to form the supervision objective,the proposed model can be applied to a variety of BTM system photovoltaic generation disaggregation scenarios,such as different PV penetration rates,different PV panel manufacturing processes and different PV equipment types.（3）In response to the problem that the net load data presented to the front side of the meter under the net energy metering policy of the BTM system will result in the mutual masking of load demand information and PV output information,the short-term net load indirect prediction strategy and the baseline load indirect estimation strategy are proposed.Under the above strategy,based on the net load disaggregation results,when constructing the short-term net load prediction model and the demand response（DR）customer baseline load（CBL）estimation model,different parts of the prediction or estimation results will be combined to obtain the final calculation results after making targeted predictions or estimates of the disaggregated energy information separately.This strategy firstly makes a targeted short-term prediction or estimation of the energy information obtained from the disaggregation separately,and then secondly combines the results of different parts of the forecast or estimation.The prediction results obtained from the models built by temporal convolutional networks show that the net short-term load indirect prediction strategy possesses lower prediction errors than the direct prediction strategy.In the modeling process of CBL indirect estimation strategy,net load disaggregation is used to correct the photovoltaic generation part of CBL in the settlement stage of incentive-based DR based on irradiance information,which makes the CBL estimation results more stable and accurate and provides an objective and fair data basis for incentive-based DR implementation.（4）CBL is an important basis for calculating customer incentive compensation when implementing incentive-based DR by load aggregators.Under the net energy metering policy of BTM system,only the net load based CBL（NL-CBL）can be used as the metering standard to implement incentive-based DR,but at this time,it is prone to the situation that customers do not participate in incentive-based DR due to the uncertainty of photovoltaic generation.In response to the above problem,a strategy to implement incentive-based DR using load-based customer baseline load（L-CBL）as a measurement standard is proposed to solve it.The constructed economic model of incentive-based DR considering the bias of CBL estimation shows that for customers of consumption reduction ratio adjustable type,using L-CBL as a metering metric can effectively avoid the reduced willingness of customers to participate in incentive-based DR caused by the uncertainty of photovoltaic generation.At the same time,the load aggregator can still control the load demand reduction of customers by implementing different CBL averaging estimation（High5of10,Mid4of6 and Low4of5）.The constructed economic model of incentive-based DR considering the uncertainty of photovoltaic generation show that the adoption of L-CBL can avoid the risk caused by the uncertainty of photovoltaic generation in incentive-based DR for the implementer and the participant,increase the amount of incentive compensation provided by the load aggregator and increase the curtailment capacity of users to achieve a win-win situation.