Photovoltaic energy storage scheduling method and system considering source load uncertainty

Abstract

The invention discloses a photovoltaic energy storage scheduling method and system considering source load uncertainty, and relates to the technical field of energy scheduling. A building load prediction model and a photovoltaic power generation prediction model are established in advance; based on the building load prediction model and the photovoltaic power generation prediction model, generating a plurality of typical source load scenes through a Latin hypercube sampling method; and determining an energy scheduling strategy of the photovoltaic energy storage system according to each source load scene. According to the method, photovoltaic and load prediction is performed by using a machine learning model, a plurality of typical scenes are generated by using Latin hypercube sampling based on a prediction result, and the scene uncertainty is fully considered in the generation process of the typical scenes, so that the energy scheduling strategy of the photovoltaic energy storage system is generated by comprehensively considering the typical scenes; the method can effectively cope with the uncertainty of source load output, especially extreme working conditions, so that the robustness of an energy scheduling strategy is effectively enhanced, the energy utilization rate is further improved, and the power utilization cost of a user is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of energy scheduling, and in particular to a photovoltaic energy storage scheduling method and system taking into account source-load uncertainty. Background Art

[0002] In recent years, with the increasingly serious environmental pollution caused by traditional power production and the continuous development of new energy technologies, the scale of renewable energy such as wind and light connected to the energy Internet has gradually expanded. Although wind and light energy have advantages such as cleanliness and renewability, they are random, intermittent and volatile in actual deployment, which often makes it difficult to achieve reasonable energy scheduling, resulting in low energy utilization efficiency, which in turn leads to energy waste and increased electricity costs.

[0003] At present, in order to improve energy utilization efficiency, dispatching algorithms that can respond to changes in energy supply and demand in real time are usually developed, such as dispatching algorithms based on optimization theory and artificial intelligence. However, existing dispatching algorithms fail to fully consider the uncertainty of source and load output and are less practical.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a photovoltaic energy storage scheduling method and system that takes into account the uncertainty of source and load in view of the above-mentioned defects of the prior art, aiming to solve the problem that the existing energy scheduling algorithm fails to fully consider the uncertainty of source and load output and has low practicality.

[0006] The technical solution adopted by the present invention to solve the problem is as follows: In a first aspect, an embodiment of the present invention provides a photovoltaic energy storage scheduling method considering source-load uncertainty, the method comprising: Pre-establishing a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; Based on the building load prediction model and the photovoltaic power generation prediction model, a number of typical source-load scenarios are generated by a Latin hypercube sampling method; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; The energy dispatch strategy of the photovoltaic energy storage system is determined according to each of the source-load scenarios.

[0007] In a second aspect, an embodiment of the present invention further provides a photovoltaic energy storage scheduling method system considering source-load uncertainty, the system comprising: A model prediction module is used to pre-establish a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; A power consumption scenario generation module, used to generate several typical source-load scenarios through a Latin hypercube sampling method based on the building load prediction model and the photovoltaic power generation prediction model; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; The optimization scheduling control module is used to determine the energy scheduling strategy of the photovoltaic energy storage system according to each source-load scenario.

[0008] Beneficial effects of the present invention: The embodiment of the present invention uses a machine learning model to perform photovoltaic and load forecasting, and uses Latin hypercube sampling based on the forecast results to generate several typical scenarios. The generation process of these typical scenarios fully considers the scenario uncertainty. Therefore, the energy scheduling strategy of the photovoltaic energy storage system generated by comprehensively considering these typical scenarios can effectively deal with the uncertainty of source and load output, especially to deal with extreme working conditions, thereby effectively enhancing the robustness of the energy scheduling strategy, thereby improving energy utilization and reducing users' electricity costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0010] Figure 1 It is a flow chart of a photovoltaic energy storage scheduling method considering source-load uncertainty provided by an embodiment of the present invention.

[0011] Figure 2 It is a schematic diagram of parameter settings of the inverter provided by an embodiment of the present invention.

[0012] Figure 3 It is a logic diagram of electric energy dispatching control in load priority mode provided by an embodiment of the present invention.

[0013] Figure 4 This is a logic diagram of electric energy dispatch control in battery priority mode provided by an embodiment of the present invention.

[0014] Figure 5 It is a logic diagram of electric energy dispatching control in the grid priority mode provided by an embodiment of the present invention.

[0015] Figure 6 It is a battery charging and discharging effect diagram of a photovoltaic energy storage scheduling strategy taking into account source-load uncertainty provided by an embodiment of the present invention.

[0016] Figure 7 It is a module schematic diagram of a photovoltaic energy storage scheduling system taking into account source-load uncertainty provided by an embodiment of the present invention.

[0017] Figure 8 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention discloses a photovoltaic energy storage scheduling method and system considering source-load uncertainty. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0020] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0021] In view of the above-mentioned defects of the prior art, the present invention provides a photovoltaic energy storage scheduling method considering the uncertainty of source and load. The method pre-establishes a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; based on the building load prediction model and the photovoltaic power generation prediction model, a number of typical source-load scenarios are generated by the Latin hypercube sampling method; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; and the energy scheduling strategy of the photovoltaic energy storage system is determined according to each of the source-load scenarios. The present invention uses a machine learning model to perform photovoltaic and load prediction, and uses Latin hypercube sampling based on the prediction results to generate a number of typical scenarios. The generation process of these typical scenarios fully considers the scene uncertainty. Therefore, the energy scheduling strategy of the photovoltaic energy storage system generated by comprehensively considering these typical scenarios can effectively deal with the uncertainty of source-load output, especially to deal with extreme working conditions, thereby effectively enhancing the robustness of the energy scheduling strategy, thereby improving energy utilization and reducing the user's electricity cost.

[0022] like Figure 1 As shown, the method specifically comprises the following steps: Step S100, pre-establishing a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future load power consumption at each time, and the photovoltaic power generation prediction model is used to predict the future photovoltaic power generation at each time.

[0023] Specifically, this embodiment will collect relevant data of the research object in advance, and the research object can be a specified building and its photovoltaic energy storage system. Two prediction models are constructed by collecting data, one is the building load prediction model, and the other is the photovoltaic power generation prediction model. The former is used to predict the future load power consumption at different times, and the latter is used to predict the future photovoltaic power generation at different times.

[0024] In one implementation, the input data of the building load prediction model are meteorological data, date data and historical data of building electricity consumption, and the output data are future hourly load electricity consumption. The date data of the building load prediction model includes characteristic data of whether it is a weekday; the input data of the photovoltaic power generation prediction model are meteorological data, date data and historical data of photovoltaic power generation, and the output data are future hourly photovoltaic power generation.

[0025] Specifically, this embodiment collects meteorological data, date data, building electricity consumption history data, and photovoltaic power generation history data of the research object, and establishes a building load prediction model and a photovoltaic power generation prediction model respectively, both of which are hourly prediction models. Since the value of building electricity load is quite different on working days and non-working days, the building load prediction model adds an input feature of whether it is a working day compared to the photovoltaic power generation prediction model.

[0026] For example, the original data of photovoltaic power generation and building electricity consumption of a building over a period of time, as well as the meteorological data and date data during this period, are obtained, and feature engineering is performed. The data collection interval can be once an hour. Among them, the original data can include hourly photovoltaic power generation and building electricity load data, and the meteorological data can include outdoor temperature, humidity, and solar shortwave radiation data. In order to facilitate the model to understand the data, the date data generates features such as time, date, and whether it is a working day. Both photovoltaic power generation and building electricity load have obvious time periodicity. The historical data of photovoltaic power generation and building electricity consumption in the previous period are extracted as lag features to predict future photovoltaic power generation and load electricity consumption.

[0027] Furthermore, the input data of the building load prediction model are the number of hours in a day, whether it is a weekday, the outdoor temperature, the solar radiation, and the building usage in the previous hour, and the output data are the hourly load power consumption in the future day; the input data of the photovoltaic power generation prediction model are the number of hours in a day, the outdoor temperature, the solar radiation, and the photovoltaic power generation in the previous hour, and the output data are the hourly photovoltaic power generation in the future day.

[0028] Specifically, for the prediction process of load power consumption, the xgboost algorithm (an ensemble learning algorithm based on gradient boosting decision trees) can be used to take the number of hours in a day, whether it is a weekday, outdoor temperature, solar radiation (or solar shortwave radiation data) and historical data of building power consumption as feature inputs to establish a building load prediction model, and finally output the hourly load power consumption (or building power consumption, building power load) for the next day. For the prediction process of photovoltaic power generation, the xgboost algorithm can be used to take the number of hours in a day, outdoor temperature, solar radiation and historical data of photovoltaic power generation as feature inputs to establish a photovoltaic power generation prediction model, and finally output the hourly photovoltaic power generation for the next day.

[0029] For example, the meteorological data and date data of the research object are collected as feature inputs. For the building load prediction model, since the load power consumption has a large difference in value between working days and non-working days, a new feature "is weekday" is constructed based on whether it is on a working day. For example, when the date is on a working day, this feature value is 1, and it is 0 on a non-working day. Secondly, because the load power consumption is greatly affected by historical values, the load power consumption in the previous hour is added as a model feature. Finally, the input data of the building load prediction model includes: the number of hours in a day, whether it is on a working day, outdoor temperature, solar radiation, and load power consumption in the previous hour, a total of 5 input features. For the photovoltaic power generation prediction model, since the photovoltaic power generation is less affected by working days and non-working days, the feature "is weekday" is not added. Compared with the building load prediction model, the photovoltaic power generation prediction model uses the photovoltaic power generation in the previous hour, the latter uses the load power consumption in the previous hour, and the former does not have the feature "is weekday", and the other input features of the two are the same. Finally, the output data of the building load prediction model is the hourly load power consumption in the future day, and the output data of the photovoltaic power generation prediction model is the hourly photovoltaic power generation in the future day.

[0030] Furthermore, for the training process of the building load prediction model and the photovoltaic power generation prediction model, the collected raw data can be divided into a training set and a test set with a division ratio of 9:1. The training set is used for model training, and the test set is used for the evaluation of the final model. For example, the model is trained using the historical data of the previous 31 days and predicts the photovoltaic power generation or load power consumption of the next day.

[0031] By comprehensively analyzing the training time and determination coefficient ( ) to evaluate the overall prediction performance of the prediction model.

[0032] The calculation formula of the evaluation index is: ; In the formula, is the predicted output of the ith data point, is the actual value of the ith data point, is the sample mean, that is, the average of all actual values. ) is closer to 1, indicating that the prediction model can better explain the volatility of the observed data and has a better fit to the real data.

[0033] For example, the historical data from June 19, 2018 to July 19, 2018 are used for model training to predict the photovoltaic power generation data and load power consumption data from July 20, 2018 to August 19, 2018. The prediction results show that the determination coefficient of the photovoltaic power generation prediction model ( ) is 0.91, and the root mean square error (RMSE) is 0.62; the determination coefficient of the building load prediction model ( ) is 0.95 and the root mean square error (RMSE) is 0.53.

[0034] Step S200, based on the building load prediction model and the photovoltaic power generation prediction model, generate several typical source-load scenarios through the Latin hypercube sampling method; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model.

[0035] This embodiment uses a multi-scenario approach to describe the uncertainty of renewable energy (i.e., photovoltaic power generation) and load power consumption. This embodiment specifically uses the Latin hypercube sampling method: the possible value range of each input variable is divided into intervals of equal probability, and then a sample is drawn from each interval to ensure that the sampling of all variables covers their entire range. Applying the hypercube sampling method to the scenario generation process can effectively reduce sample repeatability and achieve better results with a relatively small number of samples.

[0036] In one implementation, based on the building load prediction model and the photovoltaic power generation prediction model, several typical source-load scenarios are generated by a Latin hypercube sampling method, including: For each prediction model of the building load prediction model and the photovoltaic power generation prediction model, obtaining prediction data of a target historical time period through the prediction model; Calculate the probability distribution function of the error percentage per hour in the target historical time period based on the predicted data and the real data corresponding to the target historical time period; Determine a number of sampling points according to the probability distribution function, determine a scene according to the sample value of each of the sampling points, and obtain a number of scenes; Performing scene reduction according to each scene to obtain a number of target scenes and scene probabilities of each target scene; wherein the target scene of the building load prediction model is the load power consumption scene, and the target scene of the photovoltaic power generation prediction model is the photovoltaic power generation scene; Each of the load power consumption scenarios and each of the photovoltaic power generation scenarios are combined in pairs to obtain several source-load scenarios.

[0037] Specifically, the principle of generating scenes based on the building load prediction model or the photovoltaic power generation prediction model in this embodiment is the same, so this embodiment takes a prediction model as an example to illustrate the scene generation process. First, a target historical time period (for example, one month) is selected, and the target historical time period needs to have real data related to the prediction performance of the prediction model. The prediction data of the target historical time period is obtained through the prediction model, and the prediction data is subtracted from the known real data and divided by the real data to obtain the probability distribution function of the error percentage of the target historical time period (for example, a normal distribution function). The Latin hypercube sampling method is used to sample the probability distribution of the error percentage per hour, and several target scenes are obtained after scene reduction. For the convenience of distinction, this embodiment defines the target scene of the building load prediction model as the load power consumption scene, and the target scene of the photovoltaic power generation prediction model as the photovoltaic power generation scene. Randomly combine all load power consumption scenes with all photovoltaic power generation scenes in pairs to obtain several typical source-load scenes, which fully consider the scene uncertainty. The scene generation method in this embodiment has low requirements on the amount of data, and the model has strong interpretability, avoids complex calculations, and improves the operation efficiency and speed.

[0038] In one implementation, calculating the probability distribution function of the error percentage per hour in the target historical time period according to the predicted data and the real data corresponding to the target historical time period includes: The predicted data is subtracted from the actual data corresponding to the target historical time period and divided by the actual data to obtain a probability distribution function of the error percentage per hour in the target historical time period.

[0039] For example, the building load prediction model and photovoltaic power generation prediction model are first used to predict the photovoltaic power generation and load power consumption from July 20, 2018 to August 19, 2018. Then the predicted values ​​of photovoltaic power generation and load power consumption are subtracted from the true values ​​and divided by the true values ​​to obtain the probability distribution of the error percentage per hour for this month. Considering that the generation of a large number of scenarios will increase the computational complexity, the synchronous back-substitution elimination method can be used to reduce the scenarios, and 5 photovoltaic power generation scenarios and 5 load power consumption scenarios and the corresponding scenario probabilities of each scenario are obtained. The reduced typical scenario set can well reflect the probability distribution of the original scenario set.

[0040] In one implementation, determining a plurality of sampling points according to the probability distribution function, and determining a scene according to a sample value of each of the sampling points includes: Divide a number of probability intervals equally according to the probability distribution function; Determine the sampling points corresponding to each of the probability intervals according to the random numbers in each of the probability intervals; Performing an inverse transformation on the probability distribution function to obtain sample values ​​corresponding to each of the sampling points; A scene is determined according to the sample value of each sampling point.

[0041] Specifically, when using the Latin hypercube sampling method for multi-scenario generation, the probability distribution function is first divided into n probability intervals, and then the random numbers in each probability interval are As sampling points, finally the probability distribution function Perform inverse transform to obtain the sample value of the sampling point .

[0042] For example, taking photovoltaic power generation as an example, assuming that the error between the predicted value and the actual value of photovoltaic power generation per hour from July 20, 2018 to August 19, 2018 follows a normal distribution N ( , ), is the mean of the errors from July 20 to August 19, The percentage of its fluctuation. The probability distribution of the error percentage per hour is divided into 100 intervals for random sampling at equal intervals to ensure that the value range of each variable is evenly covered and improve the representativeness of the sample. The error sampling of load power consumption can be realized in the same way, which will not be repeated here.

[0043] In one implementation, scene reduction is performed according to each scene to obtain a number of target scenes and scene probabilities of each of the target scenes, including: Calculate the Euclidean distance between each scene; For each scene, select another scene with the smallest Euclidean distance to the scene as a pairing scene to obtain a scene combination, and calculate the product of the Euclidean distance and the scene probability of the pairing scene; Filter out the scene combination with the smallest product from each scene combination, delete the scene with the smallest scene probability in the scene combination, and superimpose the scene probability of the deleted scene on another scene in the same scene combination; It is determined whether the total number of remaining scenes reaches a preset value. If not, the step of calculating the Euclidean distance between each scene is continued until the total number of remaining scenes reaches the preset value, and the remaining scenes are used as target scenes.

[0044] Specifically, considering that the generation of a large number of scenes will increase the computational complexity, this embodiment uses the synchronous back-substitution method to reduce the scenes, and the reduced typical scene set can well reflect the probability distribution of the original scene set.

[0045] For example, assuming that the number of photovoltaic power generation scenarios generated by the Latin hypercube sampling method is N, the number of scenarios after reduction is n, and the steps of scenario reduction are: Step (1) Initialization: The probability value of each scene is , the initial reduction number of scenes is .

[0046] Step (2) Calculate each scene Euclidean distance , the calculation method is: ; In the formula, is the photovoltaic power generation value of scene i at time t, is the photovoltaic power generation value of scenario j at time t.

[0047] Step (3) Selection and scenario The scene with the smallest distance , calculate the product of Euclidean distance and scene probability, recorded as: ; The deleted scene is Determined, Euclidean distance Reflects the similarity between two scenes, scene probability Indicates the possibility of the occurrence of the scene. If two scenes and If the product of the Euclidean distance and the scene probability is the smallest, the scene with the smallest scene probability in the scene combination is deleted; Step (4) Delete the above identified scenes , and accumulate the scene probability of the deleted scene to the sample with the closest Euclidean distance , thus ensuring that the sum of the probabilities is 1. Delete the scene After that, the scene probability can be updated as .

[0048] Step (5) Repeat the above steps (2)-(4) until the number of remaining scenes reaches the set value.

[0049] The load power reduction process is similar to the above steps and will not be repeated here.

[0050] Assume that Latin hypercube sampling is used to generate 100 photovoltaic power generation scenarios and 100 power load scenarios that obey probability distribution, and different categories of scenarios are reduced separately. The initial number of scenarios N = 100, substitute the probability of each scenario in the above step (1) =1 / 100. Using the Euclidean distance to reduce the scenarios, steps (3) and (4) are to find the two closest scenarios, merge them, and then delete one scenario. The probability of the other scenario becomes 2 / 100, and the cycle is repeated to reduce to 5 scenarios and obtain the probabilities of these 5 scenarios. Finally, 5 photovoltaic power generation scenarios and 5 power load scenarios are obtained, and these typical scenarios with corresponding probabilities after scenario reduction are derived.

[0051] Step S300: determining an energy dispatch strategy for a photovoltaic energy storage system according to each of the source-load scenarios.

[0052] Specifically, this embodiment uses a machine learning model to perform photovoltaic and load forecasting, and uses Latin hypercube sampling based on the forecast results to generate several typical scenarios. The generation process of these typical scenarios fully considers the scenario uncertainty. Therefore, comprehensively considering these typical scenarios to generate an energy scheduling strategy for a photovoltaic energy storage system can effectively cope with the uncertainty of source and load output, especially extreme working conditions, thereby effectively enhancing the robustness of the energy scheduling strategy, thereby improving energy utilization and reducing users' electricity costs.

[0053] In one implementation, determining an energy dispatch strategy for a photovoltaic energy storage system according to each of the source-load scenarios includes: Establish an objective function based on the system's daily electricity cost; Set constraints based on the inverter's equipment operation information; Determine the power trend and several working modes in the operation of the system, and set different cut-off power and power trend relationships based on each of the working modes; A genetic algorithm is established using the objective function and the constraint conditions, and the optimal hourly setting parameter combination of the inverter under each source-load scenario is searched through the genetic algorithm; wherein the hourly setting parameter combination includes the optimal working mode and cut-off power setting value per hour; The inverter and / or battery group is dispatched according to the searched hourly setting parameter combination.

[0054] Specifically, this embodiment mainly models the control strategy of the inverter and / or battery pack. First, determine the power trend in the system operation (i.e., the priority of power flow in different working modes), comprehensively consider the multiple typical source-load scenarios generated above, and establish the optimization target with the minimum daily electricity cost of the system (the electricity cost is equal to the electricity purchase fee minus the income from photovoltaic power generation), that is, to obtain the objective function. Secondly, set constraints according to the equipment operation of the inverter to ensure that the results of the optimization solution can be applied to actual working conditions. In addition, the study obtains multiple working modes of the inverter, sets the relationship between different cut-off power and the power trend in the system operation, and uses a genetic algorithm to optimize the hourly setting parameter combination of the inverter in all scenarios of the day.

[0055] Since the research object of this embodiment is a real photovoltaic energy storage system, it is impossible to directly optimize the energy flow between photovoltaic, power grid, battery and load in the real world. It can only indirectly affect the size and direction of the above energy flow by changing the selection of the working mode at a certain moment and the battery charge and discharge amount. Therefore, the variable to be optimized is the setting value of the inverter in the real photovoltaic energy storage system. In this embodiment, the optimization variables are set as the working mode (or power consumption mode) and the cut-off power setting value of the inverter every hour. Finally, the hourly setting parameter combination of the inverter is obtained, which takes into account the uncertainty of the scene and the minimum daily electricity cost. These setting parameters are transmitted to the inverter and / or battery through serial communication to realize intelligent control. In the test phase, the daily electricity strategy can also be formulated by searching for the optimal hourly setting parameter combination, and the operating data of the inverter based on the daily electricity strategy is compared with the operating data of the inverter based on the default electricity strategy to verify the robustness of the daily electricity strategy.

[0056] For example, the inverter in this embodiment may be a hybrid inverter. Figure 2 As shown in FIG. 1 , the inverter can set different values ​​for different working modes, cut-off power, and charging and discharging power within a certain period of time. In this embodiment, the inverter is set to have three working modes: load priority mode, battery priority mode, and grid priority mode. The energy scheduling strategies under the three working modes are as follows: Figure 3 , Figure 4 , Figure 5 As shown, the photovoltaic energy storage scheduling method considering the uncertainty of source and load in this embodiment is also established in this way. The energy scheduling strategy in the load priority mode is that photovoltaics give priority to the use of loads, and the excess electricity is determined to be used for battery charging or Internet access based on the relationship between the cut-off power and the battery power; the energy scheduling strategy in the battery priority mode is that photovoltaics give priority to the power required by the battery, and photovoltaics will only supply power to the load or Internet access when the battery does not need to be charged; the grid priority mode is very similar to the load priority mode, the difference is that in the grid priority mode, when the set cut-off power is less than the battery power, the battery will discharge, while in the load priority mode, the battery will not discharge.

[0057] Since the charge and discharge power is set to 100%, only the hourly charge and discharge cutoff power value is optimized, which can also meet all situations where the charge and discharge cutoff power and the charge and discharge power are simultaneously used as optimization variables. Therefore, in order to simplify the solution of the model, this embodiment finally sets the variables that need to be optimized to be the selection of hourly working modes in a day and the size of the hourly cutoff power (or hourly battery cutoff charge). Combine all the source and load scenarios obtained above to perform parameter optimization. For example, 5 load power usage scenarios and 5 photovoltaic power usage scenarios can be combined into 5*5=25 source and load scenarios. It is necessary to find the optimal hourly setting parameter combination that meets the 25 source and load scenarios. Finally, the results obtained from the optimization are applied to the real photovoltaic energy storage system to achieve the purpose of saving users' electricity costs. Figure 6 As shown, a diagram showing the battery charging and discharging effect after adopting the method of this embodiment is shown.

[0058] In one implementation, the genetic algorithm is used to: Determine an hourly setting parameter combination to be optimized, and randomly generate a set of initial solutions as a population based on the hourly setting parameter combination to be optimized; wherein each of the initial solutions is an individual in the population; Taking the objective function as the fitness function, calculating the fitness of each individual under each source-load scenario; Determine whether the preset stop condition is currently met; If not, then reproduce according to the fitness of each individual to obtain a new population; the reproduction process includes individual selection, crossover operation, mutation operation and individual replacement; Continue to perform the step of calculating the fitness of each individual under each source-load scenario until the stop condition is met, and select the individual with the highest fitness as the optimal solution.

[0059] Specifically, this embodiment needs to optimize and solve the hourly setting parameter combination of the inverter. First, determine the hourly setting parameter combination that needs to be optimized, including the optimal working mode and the cut-off power setting value per hour, and randomly generate a set of initial solutions, called a population. Each solution is called an individual, usually represented by a chromosome. The objective function established above is used as the fitness function, and the quality of each individual is evaluated by the fitness function, and the individual is selected for reproduction according to the fitness of each individual. The principle of survival of the fittest can be adopted for reproduction, that is, individuals with high fitness are more likely to be selected to ensure the transmission of excellent genes. During reproduction, the two selected individuals are cross-operated to generate new individuals (offspring). The crossover process simulates gene recombination in biological genetics, and adopts a single-point crossover or multi-point crossover method. Furthermore, the new individuals can also be randomly mutated to increase the diversity of the population and prevent falling into a local optimum. The mutation operation can randomly change certain genes of an individual within a small range, for example, each new individual has a 10% probability of undergoing mutation. The newly generated individuals are added to the population, and some better individuals are selected and retained according to the fitness to form a new population. Repeat the above fitness evaluation, selection, crossover, mutation and replacement operations until the stop condition is met and the result is output. In this embodiment, the stop condition can be set to reach a set number of iterations, or the fitness value converges to a stable value (i.e., the optimal solution is found), and the individual with the highest fitness can be taken as the optimal solution.

[0060] In one implementation, the objective function is used as a fitness function to calculate the fitness of each individual in each source-load scenario, including: The objective function value of each individual in each source-load scenario is calculated by the objective function; wherein, for each source-load scenario, the objective function values ​​corresponding to the load power consumption scenario and the photovoltaic power generation scenario are calculated by the objective function, and the objective function value corresponding to the source-load scenario is determined according to the sum of the products of the two objective function values ​​and the corresponding scenario probabilities; The fitness of the individual is determined according to the objective function value under each source-load scenario.

[0061] Specifically, this embodiment takes minimizing the daily electricity cost of the system as the optimization goal, and establishes the objective function as follows: ; In the formula, The cost of electricity for one day; For the scene The corresponding probability; For the scene The amount of electricity purchased from the power grid in the next hour t; is the electricity purchase price at hour t; For the scene The amount of electricity sold to the grid in the next hour t; is the on-grid electricity price at hour t.

[0062] Taking an individual as an example, the fitness of the individual is calculated as follows: for each source-load scenario, the objective function value of each scenario in the source-load scenario is calculated by the above objective function, and then the product of the objective function value of each scenario and the respective scenario probability is taken as the objective function value of the source-load scenario, and the fitness of the individual is determined by combining the objective function values ​​of all source-load scenarios.

[0063] The advantages of the present invention are: (1) Efficient optimization: Energy management is performed by considering prediction uncertainty. The Latin hypercube sampling method is used to sample the error percentage of photovoltaic output and power load every hour. The required sampling samples are only one month of data. The corresponding probability distribution for each hour is obtained based on the historical data of the error percentage for this month. Then, the obtained probability distribution is sampled at equal intervals to generate photovoltaic and load scenarios, and scenario reduction is performed to obtain typical photovoltaic and load scenarios. The final scenario fully considers the extreme scenarios when the prediction error is large, and effectively simplifies the scenario generation process, avoiding the large amount of complex calculations in the Monte Carlo simulation, and improving the calculation speed. In addition, the white box nature of this method can also intuitively explain the system operation mechanism. Compared with traditional energy management strategy research, the scheduling strategy obtained by this method has better robustness and can better cope with uncertainties that may occur in the future.

[0064] (2) Strong practicability: Modeling the control logic and equipment operation constraints of real inverters. Genetic algorithms are used to optimize customized working modes and cut-off power settings to ensure that the search results can be directly applied to the energy scheduling methods of real equipment. It has stronger advantages in dealing with non-ideal factors and complex operating conditions, avoiding the inability of simulation models to accurately capture complex dynamic phenomena in the equipment, and can more accurately reflect the actual behavior of the equipment. Compared with simulation, which is difficult to fully consider equipment constraints, this method can better cope with the complexity of actual operation, reduce deviations caused by inaccurate model assumptions, and is more reliable. The optimal parameters obtained by the search can be directly applied to real equipment, reducing the complexity of model conversion and adaptation, and improving the efficiency and benefits of energy management.

[0065] Based on the above embodiments, the present invention also provides a photovoltaic energy storage scheduling system considering source-load uncertainty, such as Figure 7 As shown, the system comprises: A model prediction module is used to pre-establish a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; A power consumption scenario generation module, used to generate several typical source-load scenarios through a Latin hypercube sampling method based on the building load prediction model and the photovoltaic power generation prediction model; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; The optimization scheduling control module is used to determine the energy scheduling strategy of the photovoltaic energy storage system according to each source-load scenario.

[0066] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 8 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a photovoltaic energy storage scheduling method considering source-load uncertainty is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0067] Those skilled in the art will understand that Figure 8 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0068] In one implementation, the terminal has one or more programs stored in its memory, and is configured to be executed by one or more processors. The one or more programs include instructions for performing a photovoltaic energy storage scheduling method that takes source-load uncertainty into consideration.

[0069] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0070] In summary, the present invention discloses a photovoltaic energy storage scheduling method and system considering source-load uncertainty, and relates to the field of energy scheduling technology. The method pre-establishes a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; based on the building load prediction model and the photovoltaic power generation prediction model, a number of typical source-load scenarios are generated by the Latin hypercube sampling method; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; and the energy scheduling strategy of the photovoltaic energy storage system is determined according to each of the source-load scenarios. The present invention uses a machine learning model to perform photovoltaic and load predictions, and uses Latin hypercube sampling based on the prediction results to generate a number of typical scenarios. The generation process of these typical scenarios fully considers the scene uncertainty. Therefore, the energy scheduling strategy of the photovoltaic energy storage system generated by comprehensively considering these typical scenarios can effectively deal with the uncertainty of source-load output, especially to deal with extreme working conditions, thereby effectively enhancing the robustness of the energy scheduling strategy, thereby improving energy utilization and reducing the user's electricity cost.

[0071] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A photovoltaic energy storage scheduling method considering source-load uncertainty, characterized in that: The method comprises: Pre-establishing a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; Based on the building load prediction model and the photovoltaic power generation prediction model, a number of typical source-load scenarios are generated by a Latin hypercube sampling method; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; The energy dispatch strategy of the photovoltaic energy storage system is determined according to each of the source-load scenarios.

2. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 1 is characterized in that: The input data of the building load prediction model are the number of hours in a day, whether it is a weekday, the outdoor temperature, the solar radiation, and the historical data of building electricity consumption, and the output data are the hourly load power consumption for the next day; the input data of the photovoltaic power generation prediction model are the number of hours in a day, the outdoor temperature, the solar radiation, and the historical data of photovoltaic power generation, and the output data are the hourly photovoltaic power generation for the next day.

3. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 1 is characterized in that: Based on the building load prediction model and the photovoltaic power generation prediction model, several typical source-load scenarios are generated by the Latin hypercube sampling method, including: For each prediction model, prediction data for a target historical time period is obtained through the prediction model; wherein the prediction model is the building load prediction model or the photovoltaic power generation prediction model; Calculate the probability distribution function of the error percentage per hour in the target historical time period based on the predicted data and the real data corresponding to the target historical time period; Determine a number of sampling points according to the probability distribution function, determine a scene according to the sample value of each of the sampling points, and obtain a number of scenes; Performing scene reduction according to each scene to obtain a number of target scenes and scene probabilities of each target scene; wherein the target scene of the building load prediction model is the load power consumption scene, and the target scene of the photovoltaic power generation prediction model is the photovoltaic power generation scene; Each of the load power consumption scenarios and each of the photovoltaic power generation scenarios are combined in pairs to obtain several source-load scenarios.

4. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 3 is characterized in that: Calculating a probability distribution function of an error percentage per hour in the target historical time period according to the predicted data and the real data corresponding to the target historical time period includes: The predicted data is subtracted from the actual data corresponding to the target historical time period and divided by the actual data to obtain a probability distribution function of the error percentage per hour in the target historical time period.

5. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 3 is characterized in that: Determining a plurality of sampling points according to the probability distribution function, and determining a scene according to the sample value of each of the sampling points, including: Divide a number of probability intervals equally according to the probability distribution function; Determine the sampling points corresponding to each of the probability intervals according to the random numbers in each of the probability intervals; Performing an inverse transformation on the probability distribution function to obtain sample values ​​corresponding to each of the sampling points; A scene is determined according to the sample value of each sampling point.

6. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 3 is characterized in that: Scene reduction is performed according to each scene to obtain a number of target scenes and scene probabilities of each target scene, including: Calculate the Euclidean distance between each scene; For each scene, select another scene with the smallest Euclidean distance to the scene as a pairing scene to obtain a scene combination, and calculate the product of the Euclidean distance and the scene probability of the pairing scene; Filter out the scene combination with the smallest product from each scene combination, delete the scene with the smallest scene probability in the scene combination, and superimpose the scene probability of the deleted scene on another scene in the same scene combination; It is determined whether the total number of remaining scenes reaches a preset value. If not, the step of calculating the Euclidean distance between each scene is continued until the total number of remaining scenes reaches the preset value, and the remaining scenes are used as target scenes.

7. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 1 is characterized in that: The energy dispatch strategy of the photovoltaic energy storage system is determined according to each source-load scenario, including: Establish an objective function based on the system's daily electricity cost; Set constraints based on the inverter's equipment operation information; Determine the power trend and several working modes in the operation of the system, and set different cut-off power and power trend relationships based on each of the working modes; A genetic algorithm is established using the objective function and the constraint conditions, and the optimal hourly setting parameter combination of the inverter under each source-load scenario is searched through the genetic algorithm; wherein the hourly setting parameter combination includes the optimal working mode and cut-off power setting value per hour; The inverter and / or battery group is dispatched according to the searched hourly setting parameter combination.

8. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 7 is characterized in that: The genetic algorithm is used to: Determine an hourly setting parameter combination to be optimized, and randomly generate a set of initial solutions as a population based on the hourly setting parameter combination to be optimized; wherein each of the initial solutions is an individual in the population; Taking the objective function as the fitness function, calculating the fitness of each individual under each source-load scenario; Determine whether the preset stop condition is currently met; If not, then reproduce according to the fitness of each individual to obtain a new population; the reproduction process includes individual selection, crossover operation, mutation operation and individual replacement; Continue to perform the step of calculating the fitness of each individual under each source-load scenario until the stop condition is met, and select the individual with the highest fitness as the optimal solution.

9. The photovoltaic energy storage scheduling method considering source-load uncertainty according to claim 8 is characterized in that: The objective function is used as a fitness function to calculate the fitness of each individual under each source-load scenario, including: The objective function value of each individual in each source-load scenario is calculated by the objective function; wherein, for each source-load scenario, the objective function values ​​corresponding to the load power consumption scenario and the photovoltaic power generation scenario are calculated by the objective function, and the objective function value corresponding to the source-load scenario is determined according to the sum of the products of the two objective function values ​​and the corresponding scenario probabilities; The fitness of the individual is determined according to the objective function value under each source-load scenario.

10. A photovoltaic energy storage scheduling method system considering source-load uncertainty, characterized in that: The system comprises: A model prediction module is used to pre-establish a building load prediction model and a photovoltaic power generation prediction model, wherein the building load prediction model is used to predict the future hourly load power consumption, and the photovoltaic power generation prediction model is used to predict the future hourly photovoltaic power generation; A power consumption scenario generation module, used to generate several typical source-load scenarios through a Latin hypercube sampling method based on the building load prediction model and the photovoltaic power generation prediction model; wherein each of the source-load scenarios includes a load power consumption scenario corresponding to the building load prediction model, and a photovoltaic power generation scenario corresponding to the photovoltaic power generation prediction model; The optimization scheduling control module is used to determine the energy scheduling strategy of the photovoltaic energy storage system according to each source-load scenario.

Images