Photovoltaic power generation intelligent adaptive control system for power grid flexibility demand response

Abstract

The invention discloses a photovoltaic power generation intelligent adaptive control system for power grid flexibility demand response, and the system comprises a data collection module which is used for collecting power grid power characteristic data of n power utilization regions in unit time and time characteristic data of the next unit time; the first analysis module is used for inputting the nth power grid power characteristic data into a pre-trained power prediction model to obtain an active power interval of the nth power utilization area in the next unit time; the second analysis module is used for determining a reactive power support interval of the nth power utilization area in the next unit time according to the active power interval, the load type and the time characteristic data of the next unit time; the stability and smoothness of the power supply voltage of the nth power utilization area in the next unit time are guaranteed, so that the power supply reliability of a power grid system is guaranteed, and the risk of power supply interruption caused by violent voltage fluctuation is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid monitoring, and in particular to a photovoltaic power generation intelligent adaptive control system that responds to power grid flexibility demands. Background Art

[0002] At present, the application of photovoltaic power generation in the reactive power regulation of the power grid is usually adjusted based on the real-time power grid demand. This method has certain advantages. Photovoltaic power generation can quickly respond to the reactive power demand of the power grid, thereby improving the operating efficiency of the power grid to a certain extent.

[0003] However, this process also has a certain lag. Although photovoltaic power generation has the ability to respond quickly, the real-time monitoring and adjustment mechanism often causes the power grid to be unable to keep up with changes in reactive power demand in a timely manner. This lag will increase the gap between power supply capacity and actual demand, affecting the overall performance of the power supply system of the power grid. In addition, the lack of necessary smoothness in the real-time adjustment process can easily lead to violent fluctuations in the power grid when performing reactive power regulation. Such fluctuations will not only affect the stability of voltage, but also cause overload of the power system and power outages, thereby reducing the reliability of power supply.

[0004] Therefore, it is an urgent problem for those skilled in the art to propose an intelligent adaptive control system for photovoltaic power generation that responds to grid flexibility demand to solve the difficulties existing in the prior art. Summary of the invention

[0005] The purpose of the present invention is to provide a photovoltaic power generation intelligent adaptive control system that responds to grid flexibility demands, so as to address the deficiencies in the background technology.

[0006] In order to achieve the above-mentioned object, the present invention provides the following technical solution: a photovoltaic power generation intelligent adaptive control system for grid flexibility demand response, comprising:

[0007] The data acquisition module is used to collect the power characteristic data of the power grid in n power consumption areas per unit time and the time characteristic data of the next unit time. The power characteristic data of the power grid includes load type, environmental data, time characteristics, average active power, and standard deviation of active power; the time characteristic data includes environmental data and time characteristics;

[0008] The first analysis module is used to input the nth power grid power characteristic data into the pre-trained power prediction model to obtain the active power interval of the nth power consumption area in the next unit time;

[0009] The second analysis module determines the reactive power support interval of the nth power consumption area in the next unit time according to the active power interval, the load type, and the time characteristic data of the next unit time;

[0010] The third analysis module is used to determine the reactive power smoothing data of the nth power consumption area in the next unit time according to the reactive power support interval, the load type, and the time characteristic data of the next unit time, where the reactive power smoothing data includes B continuous reactive powers, where B is an integer greater than 1;

[0011] The adaptive control module controls the inverter of the photovoltaic power generation equipment corresponding to the nth power consumption area according to the reactive power smoothing data, and outputs the reactive power in the next unit time.

[0012] Furthermore, the method for determining the reactive power support interval of the nth power consumption area in the next unit time includes:

[0013] Input the active power range, load type, and time characteristic data of the next unit time into the trained reactive power prediction model, and output the reactive power support range in the next unit time;

[0014] The training method of the reactive power prediction model includes:

[0015] Pre-collecting m groups of reactive power training data, where m is an integer greater than 1, the reactive power training data includes reactive power characteristic data and a reactive power support interval corresponding to the reactive power characteristic data, the reactive power characteristic data includes an active power interval, a load type, and time characteristic data of a next unit time;

[0016] Each group of reactive power training data is used as the input of the reactive power prediction model in the form of a vector. The reactive power prediction model uses the reactive power support interval corresponding to a group of reactive power feature data as the output, the reactive power support interval actually corresponding to a group of reactive power feature data as the prediction target, and minimizing the loss function value of the reactive power prediction model as the training target; when the loss function value of the reactive power prediction model is less than or equal to the preset target loss value, the training is stopped to obtain the reactive power prediction model, and the reactive power prediction model is a SVM regression, random forest regression or neural network regression model.

[0017] Furthermore, the method for determining the reactive power smoothing data of the nth power consumption area in the next unit time includes:

[0018] Input reactive power support interval, load type, and time characteristic data of the next unit time into a pre-trained curve generation model, and output reactive power smoothing data of the nth power consumption area in the next unit time;

[0019] The training method of the curve generation model includes:

[0020] The reactive power support interval, load type, and time characteristic data of the next unit time are taken as inputs of a curve generation model, and the curve generation model takes the reactive power smoothing data of the nth power consumption area in the next unit time predicted for each group of reactive power support interval, load type, and time characteristic data of the next unit time as output, and takes the reactive power smoothing data of the actual nth power consumption area in the next unit time corresponding to each group of reactive power support interval, load type, and time characteristic data of the next unit time as a prediction target, and takes minimizing the sum of the prediction accuracies of the reactive power smoothing data of all predicted nth power consumption areas in the next unit time as a training target; the curve generation model is trained until the sum of the prediction accuracies reaches convergence, and the training is stopped; the curve generation model is a random forest regression model or a neural network regression model.

[0021] Furthermore, a constraint module is selected to calculate the reactive power output by Q inverters at each time point in the next unit time based on B continuous reactive powers and operating temperature limits of Q photovoltaic power generation equipment inverters in the nth power consumption area, where Q is an integer greater than 1.

[0022] Furthermore, the method for calculating the reactive power output by the Q inverters at each time point in the next unit time includes:

[0023] Step 1: Evenly distribute the reactive power corresponding to the b-th time point to Q inverter control systems, b∈B, and calculate the temperature corresponding to the reactive power outputted by the Q inverters at the b-th time point;

[0024] Step 2: If the B temperatures are all less than or equal to the corresponding inverter operating temperature, the average reactive power is allocated according to the b-th time point, and the Q inverters are controlled to output the average reactive power at the b-th time point, and step 3 is executed; if the B temperatures are greater than the corresponding inverter operating temperature, step 4 is executed;

[0025] Step 3: Let b=b+1, and repeat steps 1-2 until b is greater than B.

[0026] Step 4: Mark the B inverters whose temperatures are greater than the corresponding inverter operating temperature as abnormal inverters; input the operating temperature, temperature at the previous time point, and time characteristic data of the next unit time corresponding to the abnormal inverter into the pre-built reactive power reconstruction model, and output the reconstructed reactive power corresponding to the abnormal inverter;

[0027] Step 5: subtract the reconstructed reactive power corresponding to the abnormal inverter from the reactive power corresponding to the b-th time point to obtain the reactive power to be allocated, and evenly distribute the reactive power to be allocated to the inverters other than the abnormal inverter;

[0028] Step 6. Repeat steps 2-3.

[0029] Further, a method for calculating the temperature corresponding to the reactive power outputted and distributed by the Q inverters at B time points includes:

[0030] Input the temperature at the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to the qth inverter into the trained temperature prediction model to obtain the temperature of the qth inverter at the bth time;

[0031] The training method of the temperature prediction model includes:

[0032] The previously collected temperature at the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to the qth inverter are used as the input of the temperature prediction model, q belongs to Q, and the temperature prediction model takes the temperature corresponding to the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to each group of the qth inverter as the output;

[0033] The temperature at the previous time point corresponding to the qth inverter in each group, the allocated reactive power and the actual temperature corresponding to the time characteristic data of the next unit time are taken as the prediction targets, and minimizing the sum of the prediction accuracies of all predicted temperatures is taken as the training target. The training is stopped when the sum of the prediction accuracies converges. The temperature prediction model is a neural network model.

[0034] Furthermore, the reactive power support interval is the reactive power of the nth power consumption area at the beginning and end of the next unit time; the active power interval is composed of the active power at the beginning and end of the next unit time.

[0035] Furthermore, the load types include industrial, commercial and residential, and the environmental data include temperature and humidity; the average active power is the average of p active powers collected in the current unit time, and the active power standard deviation is the standard deviation of p active powers collected in the current unit time.

[0036] Furthermore, the time feature is the time period value and date value of the current unit time, and the method for determining the time period value and date value includes:

[0037] The time periods include morning peak, evening peak and non-peak. Morning peak, evening peak and non-peak all have corresponding time periods and preset time period values. If the current unit time is in a certain time period, the corresponding preset time period value will be used as the time period value of the current unit time.

[0038] Dates include working days and holidays. The working day or holiday of the current unit time is determined according to the date of the current unit time. Working days or holidays have preset date values, and the corresponding preset date values ​​are used as the date value of the current unit time.

[0039] Further, the training method of the power prediction model includes: pre-collecting h groups of power training data of the nth power grid, where h is an integer greater than 1, the power training data includes power characteristic data and an active power interval corresponding to the power characteristic data for the next unit time; the power characteristic data includes power grid power characteristic data and time characteristic data;

[0040] The power training data is divided into a training set and a test set, a first classifier is constructed, the power feature data in the training set is used as input data, the active power interval in the training set is used as output data, the first classifier is trained to obtain an initial first classifier, the initial first classifier is tested using the test set, and the first classifier that meets the preset accuracy is output as a power prediction model, wherein the power prediction model is preferably a naive Bayes model or a support vector machine model.

[0041] The present invention provides a photovoltaic power generation intelligent adaptive control system that responds to grid flexibility demand. The technical effects and advantages are as follows:

[0042] By collecting the grid power characteristic data of n power consumption areas per unit time and the time characteristic data of the next unit time, the active power range of the nth power consumption area in the next unit time can be effectively predicted. Furthermore, by combining the active power range, load type and the time characteristic data of the next unit time, the reactive power support range of the area in the next unit time can be predicted. This process provides a basis for early prediction and adjustment, thereby reducing lag.

[0043] The reactive power support interval provides clear boundaries for the starting and ending values ​​of the reactive power required by the power grid under specific conditions. Using this interval as the input data of the curve generation model can effectively ensure that the generated reactive power smoothing data always meets the actual needs of the power grid. At the same time, considering the load type and the time characteristic data of the next unit time can help improve the accuracy of the reactive power smoothing data output, reduce power grid transmission losses, and improve energy utilization efficiency. The predicted reactive power smoothing data is then used to adaptively control the reactive power output of the photovoltaic power inverter to ensure the stability and smoothness of the power supply voltage in the nth power consumption area in the next unit time, thereby ensuring the reliability of the power supply of the power grid system and reducing the risk of power supply interruptions caused by severe voltage fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments 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 also be obtained based on these drawings.

[0045] Figure 1 A schematic diagram of a photovoltaic power generation intelligent adaptive control system for responding to grid flexibility demand in Embodiment 1 of the present invention;

[0046] Figure 2 A schematic diagram of a photovoltaic power generation intelligent adaptive control system for responding to grid flexibility demand in Example 1 of the present invention;

[0047] Figure 3 This is a flow chart of a method for calculating reactive power outputted by a photovoltaic power generation device inverter at each time point according to the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element at the same time.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides a photovoltaic power generation intelligent adaptive control system that responds to grid flexibility demand, including a data acquisition module, a first analysis module, a second analysis module, a third analysis module and an adaptive control module, and each module is connected by wire and / or wirelessly.

[0052] The data acquisition module is used to collect the grid power characteristic data of n power consumption areas per unit time and the time characteristic data of the next unit time. The grid power characteristic data includes load type, environmental data, time characteristics, average active power, and standard deviation of active power; the time characteristic data includes environmental data and time characteristics.

[0053] Load types include industrial, commercial and residential. Industrial loads are mainly inductive loads, which usually require a large amount of reactive power. Commercial loads are both capacitive and inductive, and residential loads are mainly resistive loads, with relatively low demand for reactive power. Different load types require different active and reactive powers per unit time, which affects changes in voltage demand. Environmental data include temperature and humidity. Changes in temperature and humidity will affect changes in electricity demand.

[0054] The average value of active power is the mean value of the p active powers collected in the current unit time, and the standard deviation of active power is the standard deviation of the p active powers collected in the current unit time; active power refers to the work or energy actually completed in a period of time. Its average value and standard deviation are two important parameters to describe the stability of power supply in the power grid. When the voltage is stable, the power grid can provide stable power supply capacity, and the average value of active power is relatively stable. Voltage fluctuations will cause the power supply capacity of the power grid to decline, affecting the average value of active power. The standard deviation of active power reflects the degree of fluctuation of the power supply capacity of the power grid. The stability of voltage has a direct impact on the fluctuation of active power. When the voltage is stable, the fluctuation of active power is small and the standard deviation is low. Voltage fluctuations may cause the fluctuation of active power to increase and the standard deviation is high.

[0055] The time feature is the time period value and date value of the current unit time. The method for determining the time period value and date value includes:

[0056] The time periods include morning peak, evening peak and non-peak. Morning peak, evening peak and non-peak all have corresponding time periods and preset time period values. If the current unit time is in a certain time period, the corresponding preset time period value will be used as the time period value of the current unit time.

[0057] Dates include working days and holidays. The working day or holiday of the current unit time is determined based on the date of the current unit time. Working days or holidays have preset date values, and the corresponding preset date values ​​are used as the date value of the current unit time. The time period value and the date value reflect different characteristics of power demand, thereby affecting changes in voltage demand.

[0058] The first analysis module is used to input the power characteristic data of the nth power grid into the pre-trained power prediction model to obtain the active power range of the nth power consumption area in the next unit time.

[0059] The active power interval is composed of the active power at the start time and the active power at the end time of the next unit time.

[0060] The training method of the power prediction model includes: pre-collecting the power training data of the nth power grid of h groups, the power training data includes the power characteristic data and the active power interval corresponding to the power characteristic data for the next unit time; the power characteristic data includes the power characteristic data of the power grid and the time characteristic data, and by considering the time characteristics and environmental factors of the current and next time, the power prediction model is trained. The model can capture the trend changes between the environmental data, time characteristics and active power in the two time periods, and the predicted active power interval can more comprehensively reflect the changes in electricity demand. This method not only improves the accuracy of the prediction, but also helps the power system better cope with demand fluctuations and ensure the stability of power supply.

[0061] The power training data is divided into a training set and a test set, a first classifier is constructed, the power feature data in the training set is used as input data, the active power interval in the training set is used as output data, the first classifier is trained to obtain an initial first classifier, the initial first classifier is tested using the test set, and the first classifier that meets the preset accuracy is output as a power prediction model. The power prediction model is preferably a naive Bayes model or a support vector machine model.

[0062] The second analysis module determines the reactive power support interval of the nth power consumption area in the next unit time according to the active power interval, load type, and time characteristic data of the next unit time. The reactive power support interval is the reactive power of the nth power consumption area at the beginning and end of the next unit time.

[0063] The method for determining the reactive power support interval of the nth power consumption area in the next unit time includes:

[0064] The active power range, load type, and time characteristic data of the next unit time are input into the trained reactive power prediction model, and the reactive power support range in the next unit time is output.

[0065] The training method of the reactive power prediction model includes:

[0066] Collect m groups of reactive power training data in advance. The reactive power training data include reactive power characteristic data and reactive power support intervals corresponding to the reactive power characteristic data. The reactive power characteristic data include active power intervals, load types, and time characteristic data of the next unit time. Active power is the power actually consumed by power equipment. The active power interval can provide the model with a range limit for the current load state (i.e., clear boundaries are provided for the start and end of the active power interval), thereby improving the prediction accuracy of the reactive power support interval. If there is no limit, the prediction deviation of the reactive power support interval in the next unit time will be large, which is not in line with reality. The increase in active power often leads to an increase in reactive power demand, especially in power grids dominated by inductive loads (industrial). In order to ensure the stability and efficiency of the power grid, it is crucial to manage the supply of reactive power.

[0067] Time characteristics affect electricity consumption patterns. For example, more reactive power support may be needed during peak hours on weekdays, while demand may be lower during off-peak hours. Temperature and humidity in environmental data also have an important impact on electricity demand. Hot weather usually leads to increased air conditioning load, which in turn increases the demand for reactive power. Humidity also affects the operating efficiency of power equipment.

[0068] Example: The next unit of time is known:

[0069] The active power range is from 150kW to 200kW.

[0070] The load type is "air conditioning".

[0071] The time characteristic is "off-peak hours on weekdays".

[0072] The environmental data is "temperature 35°C, humidity 80%".

[0073] Output: Reactive power support range is 50kVAR to 70kVAR.

[0074] By considering the reactive power demand characteristics under different conditions, the corresponding reactive power support interval is generated to provide guidance for the stable operation of the power grid. This method effectively combines multiple factors and improves the accuracy of the reactive power prediction model.

[0075] Each group of reactive power training data is used as the input of the reactive power prediction model in the form of a vector. The reactive power prediction model uses the reactive power support interval corresponding to a group of reactive power feature data as the output, the reactive power support interval actually corresponding to a group of reactive power feature data as the prediction target, and minimizing the loss function value of the reactive power prediction model as the training target; when the loss function value of the reactive power prediction model is less than or equal to the preset target loss value, the training is stopped to obtain the reactive power prediction model, and the reactive power prediction model is one of the models such as SVM regression, random forest regression or neural network regression.

[0076] The loss function value is mean square error;

[0077] Loss function formula The model is trained with minimization as the goal. In the loss function, MSE is the loss function value of the reactive power prediction model, x is the reactive power feature data group number; m is the number of reactive power feature data groups; y x is the reactive power support interval predicted by the reactive power characteristic data of the xth group, is the reactive power support interval actually corresponding to the x-th group of reactive power characteristic data.

[0078] The third analysis module is used to determine the reactive power smoothing data of the nth power consumption area in the next unit time based on the reactive power support range, load type, and time characteristic data of the next unit time. The reactive power smoothing data includes B continuous reactive powers, that is, the next unit time consists of B time points, and B is an integer greater than 1.

[0079] The method for determining the reactive power smoothing data of the nth power consumption area in the next unit time includes:

[0080] The reactive power support interval, load type, and time characteristic data of the next unit time are input into a pre-trained curve generation model, and the reactive power smoothing data of the nth power consumption area in the next unit time is output.

[0081] The training method of the curve generation model includes:

[0082] The reactive power support interval, load type, and time characteristic data of the next unit time are taken as inputs of a curve generation model, and the curve generation model takes the reactive power smoothed data of the nth power consumption area in the next unit time predicted for each set of reactive power support interval, load type, and time characteristic data of the next unit time as output, and takes the reactive power smoothed data of the actual nth power consumption area in the next unit time corresponding to each set of reactive power support interval, load type, and time characteristic data of the next unit time as a prediction target, and takes minimizing the sum of the prediction accuracies of the reactive power smoothed data of all predicted nth power consumption areas in the next unit time as a training target.

[0083] Among them, the calculation formula for prediction accuracy is: WG r =(WZ r -WY r ) 2 , where r is the number of each group of reactive power support interval, load type, and time characteristic data of the next unit time, WG r For prediction accuracy, WZ r WY is the reactive power smoothing data of the nth power consumption area in the next unit time corresponding to the reactive power support interval, load type, and time characteristic data of the next unit time of the rth group, r The reactive power smoothing data of the actual nth power consumption area in the next unit time corresponding to the rth group of reactive power support interval, load type, and time characteristic data of the next unit time is obtained; the curve generation model is trained until the sum of the prediction accuracies reaches convergence and the training is stopped.

[0084] When collecting reactive power support range, load type, time characteristic data for the next unit time and reactive power smoothing data for the next unit time, staff should ensure that the power grid in the application area achieves voltage stability and optimal loss in the next unit time based on the reactive power provided by the reactive power smoothing data. This process is screened and determined by technical personnel in the field.

[0085] At the same time, the reactive power support interval provides clear boundaries for the starting and ending values ​​of the reactive power required by the power grid under specific conditions. Using this interval as the input data of the curve generation model can effectively ensure that the generated reactive power smoothing data always meets the actual needs of the power grid. This restriction condition helps to improve the accuracy of the reactive power smoothing data output, thereby maintaining the stability of the power grid voltage. In this way, the power grid can respond to load changes more efficiently and ensure the reliability and stability of power supply.

[0086] The adaptive control module controls the inverter of the photovoltaic power generation equipment corresponding to the nth power consumption area according to the reactive power smoothing data, and outputs reactive power in the next unit time to ensure the stability of the power supply voltage of the nth power consumption area in the next unit time, thereby ensuring the stability and smoothness of the power supply of the power grid system.

[0087] Example 2

[0088] See also Figure 2 As shown, this embodiment provides a photovoltaic power generation intelligent adaptive control system that responds to grid flexibility needs, and also includes: a constraint selection module for calculating the reactive power output by Q inverters at each time point in the next unit time based on B continuous reactive powers and the operating temperature limits of Q photovoltaic power generation equipment inverters in the nth power consumption area, where Q is an integer greater than 1.

[0089] See also Figure 3 , the method for calculating the reactive power output by Q inverters at each time point in the next unit time includes:

[0090] Step 1: Evenly distribute the reactive power corresponding to the b-th time point to Q inverter control systems, b∈B, and calculate the temperature corresponding to the reactive power output distributed by the Q inverters at the b-th time point.

[0091] The method for calculating the temperature corresponding to the reactive power outputted and distributed by Q inverters at B time points includes:

[0092] The temperature at the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to the qth inverter are input into the trained temperature prediction model to obtain the temperature of the qth inverter at the bth time.

[0093] The temperature at the previous time point affects the starting point of the temperature at the next time point. The allocated reactive power directly affects the temperature rise range of the inverter. The time characteristic data of the next unit time affects the inverter output reactive power output frequency and heat dissipation efficiency. The interaction of these factors determines the temperature of each inverter under the allocated reactive power.

[0094] The training method of the temperature prediction model includes:

[0095] The pre-collected temperature at the previous time point corresponding to the qth inverter, the allocated reactive power and the time characteristic data of the next unit time are used as the input of the temperature prediction model. q belongs to Q. The temperature prediction model outputs the temperature corresponding to the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to each group of the qth inverter.

[0096] The temperature at the previous time point corresponding to the qth inverter in each group, the allocated reactive power and the actual temperature corresponding to the time characteristic data of the next unit time are taken as the prediction targets, and minimizing the sum of the prediction accuracies of all predicted temperatures is taken as the training target. The training is stopped when the sum of the prediction accuracies converges. The temperature prediction model is a neural network model.

[0097] Step 2: If the B temperatures are all less than or equal to the corresponding inverter operating temperature, allocate the average reactive power according to the b-th time point, control the Q inverters to output the allocated average reactive power at the b-th time point, and execute step 3; if any of the B temperatures is greater than the corresponding inverter operating temperature, execute step 4.

[0098] Step 3: Let b=b+1, continue processing at the next time point, repeat steps 1-2, and end when b is greater than B.

[0099] Step 4: Mark B inverters with temperatures greater than the corresponding inverter operating temperature as abnormal inverters; input the operating temperature corresponding to the abnormal inverter, the temperature at the previous time point, and the time characteristic data of the next unit time into the pre-built reactive power reconstruction model, and output the reconstructed reactive power corresponding to the abnormal inverter.

[0100] Taking the operating temperature corresponding to the abnormal inverter as the limit and considering the factors affecting the inverter temperature, for example, the temperature at the previous time point and the time characteristic data of the next unit time, the reconstructed reactive power under the operating temperature limit is predicted. The reconstructed reactive power is the reactive power output by the inverter at the bth time point. At the same time, ensure that the temperature of the inverter is less than or equal to its operating temperature to reduce the impact of high temperature on the service life of the inverter.

[0101] The reactive power reconstruction model training method includes:

[0102] The operating temperature, temperature at the previous time point, and time characteristic data of the next unit time corresponding to the abnormal inverters collected in advance are used as the input of the reactive power reconstruction model. The reactive power reconstruction model outputs the reconstructed reactive power corresponding to the operating temperature, temperature at the previous time point, and time characteristic data of the next unit time corresponding to each group of abnormal inverters.

[0103] The prediction target is the operating temperature corresponding to each group of abnormal inverters, the temperature at the previous time point, and the reconstructed reactive power actually corresponding to the time characteristic data of the next unit time. The training target is to minimize the sum of the prediction accuracies of all predicted reconstructed reactive powers. The training is stopped when the sum of the prediction accuracies converges. The reactive power reconstruction model is a neural network model.

[0104] Step 5: Subtract the reconstructed reactive power corresponding to the abnormal inverter from the reactive power corresponding to the b-th time point to obtain the reactive power to be allocated, and evenly distribute the reactive power to be allocated to the inverters except the abnormal inverter. This not only ensures that the allocated reactive power temperature of the Q inverters at each time point does not exceed their operating temperature, but also meets the reactive power demand of the power grid at the b-th time point, while reducing the impact on the service life of the inverter.

[0105] Step 6: Repeat steps 2-3 to continuously monitor and dynamically adjust the distribution of reactive power.

[0106] This embodiment provides a photovoltaic power generation intelligent adaptive control system that responds to grid flexibility needs, fully considers the reactive power distribution to each inverter at each time point in the next unit time, and monitors the temperature in real time, ensuring that the grid's demand for reactive power at each time point in the next unit time is met, maintaining the stability of the power supply voltage, and effectively extending the service life of the inverter.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A photovoltaic power generation intelligent adaptive control system for grid flexibility demand response, characterized in that: include: The data acquisition module is used to collect the power characteristic data of the power grid in n power consumption areas per unit time and the time characteristic data of the next unit time. The power characteristic data of the power grid includes load type, environmental data, time characteristics, average active power, and standard deviation of active power; the time characteristic data includes environmental data and time characteristics; The first analysis module is used to input the nth power grid power characteristic data into the pre-trained power prediction model to obtain the active power interval of the nth power consumption area in the next unit time; The second analysis module determines the reactive power support interval of the nth power consumption area in the next unit time according to the active power interval, the load type, and the time characteristic data of the next unit time; The third analysis module is used to determine the reactive power smoothing data of the nth power consumption area in the next unit time according to the reactive power support interval, the load type, and the time characteristic data of the next unit time, where the reactive power smoothing data includes B continuous reactive powers, where B is an integer greater than 1; The adaptive control module controls the inverter of the photovoltaic power generation equipment corresponding to the nth power consumption area according to the reactive power smoothing data, and outputs the reactive power in the next unit time.

2. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 1 is characterized in that: The method for determining the reactive power support interval of the nth power consumption area in the next unit time includes: Input the active power range, load type, and time characteristic data of the next unit time into the trained reactive power prediction model, and output the reactive power support range in the next unit time; The training method of the reactive power prediction model includes: Pre-collecting m groups of reactive power training data, where m is an integer greater than 1, the reactive power training data includes reactive power characteristic data and a reactive power support interval corresponding to the reactive power characteristic data, the reactive power characteristic data includes an active power interval, a load type, and time characteristic data of a next unit time; Each group of reactive power training data is used as the input of the reactive power prediction model in the form of a vector. The reactive power prediction model uses the reactive power support interval corresponding to a group of reactive power feature data as the output, the reactive power support interval actually corresponding to a group of reactive power feature data as the prediction target, and minimizing the loss function value of the reactive power prediction model as the training target; when the loss function value of the reactive power prediction model is less than or equal to the preset target loss value, the training is stopped to obtain the reactive power prediction model, and the reactive power prediction model is a SVM regression, random forest regression or neural network regression model.

3. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 2 is characterized in that: The method for determining the reactive power smoothing data of the nth power consumption area in the next unit time includes: Input reactive power support interval, load type, and time characteristic data of the next unit time into a pre-trained curve generation model, and output reactive power smoothing data of the nth power consumption area in the next unit time; The training method of the curve generation model includes: The reactive power support interval, load type, and time characteristic data of the next unit time are taken as inputs of a curve generation model, and the curve generation model takes the reactive power smoothing data of the nth power consumption area in the next unit time predicted for each group of reactive power support interval, load type, and time characteristic data of the next unit time as output, and takes the reactive power smoothing data of the actual nth power consumption area in the next unit time corresponding to each group of reactive power support interval, load type, and time characteristic data of the next unit time as a prediction target, and takes minimizing the sum of the prediction accuracies of the reactive power smoothing data of all predicted nth power consumption areas in the next unit time as a training target; the curve generation model is trained until the sum of the prediction accuracies reaches convergence, and the training is stopped; the curve generation model is a random forest regression model or a neural network regression model.

4. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 2, characterized in that: A constraint module is selected to calculate the reactive power output by Q inverters at each time point in the next unit time according to B continuous reactive powers and operating temperature limits of Q photovoltaic power generation equipment inverters in the nth power consumption area, where Q is an integer greater than 1.

5. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 4 is characterized in that: The method for calculating the reactive power output by the Q inverters at each time point in the next unit time includes: Step 1: Evenly distribute the reactive power corresponding to the b-th time point to Q inverter control systems, b∈B, and calculate the temperature corresponding to the reactive power outputted by the Q inverters at the b-th time point; Step 2: If the B temperatures are all less than or equal to the corresponding inverter operating temperature, the average reactive power is allocated according to the b-th time point, and the Q inverters are controlled to output the average reactive power at the b-th time point, and step 3 is executed; if the B temperatures are greater than the corresponding inverter operating temperature, step 4 is executed; Step 3: Let b=b+1, and repeat steps 1-2 until b is greater than B. Step 4: Mark the B inverters whose temperatures are greater than the corresponding inverter operating temperature as abnormal inverters; input the operating temperature, temperature at the previous time point, and time characteristic data of the next unit time corresponding to the abnormal inverter into the pre-built reactive power reconstruction model, and output the reconstructed reactive power corresponding to the abnormal inverter; Step 5: subtract the reconstructed reactive power corresponding to the abnormal inverter from the reactive power corresponding to the b-th time point to obtain the reactive power to be allocated, and evenly distribute the reactive power to be allocated to the inverters other than the abnormal inverter; Step 6. Repeat steps 2-3.

6. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 5, characterized in that: The method for calculating the temperature corresponding to the reactive power outputted by Q inverters at B time points includes: Input the temperature at the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to the qth inverter into the trained temperature prediction model to obtain the temperature of the qth inverter at the bth time; The training method of the temperature prediction model includes: The previously collected temperature at the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to the qth inverter are used as the input of the temperature prediction model, q belongs to Q, and the temperature prediction model takes the temperature corresponding to the previous time point, the allocated reactive power and the time characteristic data of the next unit time corresponding to each group of the qth inverter as the output; The temperature at the previous time point corresponding to the qth inverter in each group, the allocated reactive power and the actual temperature corresponding to the time characteristic data of the next unit time are taken as the prediction targets, and minimizing the sum of the prediction accuracies of all predicted temperatures is taken as the training target. The training is stopped when the sum of the prediction accuracies converges. The temperature prediction model is a neural network model.

7. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 1, characterized in that: The reactive power support interval is the reactive power of the nth power consumption area at the beginning and end of the next unit time; the active power interval is composed of the active power at the beginning and end of the next unit time.

8. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 1, characterized in that: The load types include industrial, commercial and residential, and the environmental data include temperature and humidity; the average active power is the average of p active powers collected in the current unit time, and the standard deviation of active power is the standard deviation of p active powers collected in the current unit time.

9. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 1, characterized in that: The time feature is the time period value and date value of the current unit time. The time period value and date value determination method includes: The time periods include morning peak, evening peak and non-peak. Morning peak, evening peak and non-peak all have corresponding time periods and preset time period values. If the current unit time is in a certain time period, the corresponding preset time period value will be used as the time period value of the current unit time. Dates include working days and holidays. The working day or holiday of the current unit time is determined according to the date of the current unit time. Working days or holidays have preset date values, and the corresponding preset date values ​​are used as the date value of the current unit time.

10. The photovoltaic power generation intelligent adaptive control system for grid flexibility demand response according to claim 1, characterized in that: The training method of the power prediction model includes: pre-collecting h groups of power training data of the nth power grid, where h is an integer greater than 1, the power training data includes power characteristic data and an active power interval corresponding to the power characteristic data for the next unit time; the power characteristic data includes power characteristic data of the power grid and time characteristic data; The power training data is divided into a training set and a test set, a first classifier is constructed, the power feature data in the training set is used as input data, the active power interval in the training set is used as output data, the first classifier is trained to obtain an initial first classifier, the initial first classifier is tested using the test set, and the first classifier that meets the preset accuracy is output as a power prediction model, wherein the power prediction model is preferably a naive Bayes model or a support vector machine model.

Images