Search

CN-122026325-A - Distribution optimization method and device of distributed power grid and electronic equipment

CN122026325ACN 122026325 ACN122026325 ACN 122026325ACN-122026325-A

Abstract

The invention relates to the field of power distribution optimization, and particularly provides a power distribution optimization method, device and electronic equipment of a distributed power grid. The method is applied to a power distribution network comprising a plurality of sub power distribution networks, and comprises the steps of predicting load demand power of the sub power distribution networks in a prediction period based on first date data and first weather data corresponding to the prediction period, predicting output power of the sub power distribution networks in the prediction period based on second date data and second weather data corresponding to the prediction period, wherein the output power comprises at least one or the sum of photovoltaic output power representing photovoltaic and wind speed output power representing wind power generation, and generating a power distribution control instruction for the sub power distribution networks according to the load demand power, the output power, the energy storage charge state of the sub power distribution networks and local node voltage of the sub power distribution networks. The technical scheme provided by the invention can give consideration to the comprehensive benefits of each distributed sub-power distribution network.

Inventors

  • DING LONG
  • LIU CHANGYI

Assignees

  • 国网山西省电力有限公司经济技术研究院

Dates

Publication Date
20260512
Application Date
20260128

Claims (10)

  1. 1. A power distribution optimization method for a distributed power grid, the method being applied to a power distribution network, the power distribution network comprising a plurality of sub-power distribution networks, the method comprising: Predicting load demand power of the sub-distribution network in a prediction period based on first date data and first meteorological data corresponding to the prediction period; predicting the output power of the sub-distribution network in the prediction period based on second date data and second meteorological data corresponding to the prediction period, wherein the output power comprises at least one or the sum of photovoltaic output power representing photovoltaic and wind speed output power representing wind power generation; generating a distribution control instruction for the sub-distribution network according to the load demand power, the output power, the energy storage charge state of the sub-distribution network and the local node voltage of the sub-distribution network, wherein the distribution control instruction comprises energy storage exchange power and grid exchange power, and the distribution control instruction enables the sum of the local running cost, the green rewarding cost and the global out-of-limit punishment cost of the sub-distribution network, which are calculated according to the energy storage exchange power and the grid exchange power, to be maximum.
  2. 2. The method of claim 1, wherein the first date data comprises time data and a date type, the first weather data comprises temperature data and a weather type, and predicting the load demand power of the sub-distribution network during the predicted period based on the first date data and the first weather data corresponding to the predicted period comprises: Constructing a first training data set, wherein the first training data set comprises a first input data set and a first output data set corresponding to the first input data set, the first input data set comprises months corresponding to the time data, the date type, the temperature data and the weather type, and the first output data set comprises historical load power corresponding to unit area; Converting the first input dataset into a first historical condition vector of numerical value type; training the load prediction model by taking the first historical condition vector as an input of the load prediction model and taking historical load power corresponding to the first historical condition vector as an output of the load prediction model; converting the first date data and the first meteorological data into first condition vectors, and inputting the first condition vectors into the trained load prediction model to obtain standard load demand power in unit area; and multiplying the standard load demand power by the area covered by the sub power distribution network to obtain the load demand power of the sub power distribution network in the prediction period.
  3. 3. The method of claim 2, wherein the load prediction model is a modified generation countermeasure network model comprising a generator and a arbiter, wherein training the load prediction model with the first historical condition vector as an input to the load prediction model and the historical load power corresponding to the first historical condition vector as an output of the load prediction model comprises: inputting the first historical condition vector and random noise into the generator to obtain predicted load power; Inputting a plurality of predicted load powers and historical load powers corresponding to the predicted load powers into the discriminator to obtain bulldozer distances between the historical load powers and the predicted load powers; And under the condition that the bulldozer distance is smaller than or equal to the preset distance, taking the current improved generation countermeasure network model as a load prediction model.
  4. 4. The method of claim 1, wherein the second date data comprises time data, the second weather data comprises temperature data, solar radiation data weather type, and wind speed data, and predicting the power output of the sub-distribution network during the predicted period based on the second date data and the second weather data corresponding to the predicted period comprises: Constructing a second training data set, wherein the second training data set comprises a second input data set and a second output data set corresponding to the second input data set, the second input data set comprises months, temperature data, solar radiation data and weather types corresponding to the time data, and the second output data set comprises historical photovoltaic output power corresponding to a unit area; Converting the second input dataset into a second historical condition vector of numerical value; Training the photovoltaic output prediction model by taking the second historical condition vector as an input of the photovoltaic output prediction model and taking the historical photovoltaic output corresponding to the second historical condition vector as an output of the photovoltaic output prediction model; Converting the second date data and the second meteorological data into a second condition vector, and inputting the second condition vector into a trained photovoltaic output prediction model to obtain standard photovoltaic output power of a unit area; multiplying the standard photovoltaic output power by the area of a photovoltaic power generation plate in the sub-power distribution network to obtain the photovoltaic output power of the sub-power distribution network in the prediction period; Constructing a third training data set, wherein the third training data set comprises a third input data set and a third output data set corresponding to the third input data set, the third input data set comprises historical wind speed data, and the third output data set comprises the historical wind speed output power of a single wind driven generator; Training the wind speed output prediction model by taking the historical wind speed data as the input of the wind speed output prediction model and taking the historical wind speed output corresponding to the historical wind speed data as the output of the wind speed output prediction model; inputting the wind speed data into the wind speed output prediction model to obtain standard wind speed output power of a single wind driven generator; And multiplying the standard wind speed output power by the number of wind driven generators in the sub-distribution network to obtain the wind speed output power of the sub-distribution network in the predicted period.
  5. 5. The method of claim 1, wherein prior to the step of generating distribution control instructions for the sub-distribution network based on the load demand power, the output power, the stored state of charge of the sub-distribution network, and the local node voltage of the sub-distribution network, the method further comprises: The method comprises the steps of constructing a sample set, wherein the sample set comprises date condition data, meteorological data corresponding to the date condition data, energy storage charge states of a sub-power distribution network and node voltages of all nodes, and the date condition data comprises first date data and second date data; Generating electricity consumption data sets under various different conditions based on the sample set, wherein the electricity consumption data sets comprise load demand power, output power, energy storage charge state of each sub-distribution network and node voltage of each node; The method comprises the steps of constructing a deep reinforcement learning model, wherein the deep reinforcement learning model comprises a strategy network and an evaluation network, the strategy network is used for generating a power distribution strategy, the power distribution strategy comprises energy storage exchange power and power grid exchange power, and the evaluation network is used for generating an evaluation result according to the power distribution strategy; Generating a reward score for the power distribution strategy according to the power distribution strategy, wherein the reward score comprises the sum of the local running cost of the sub power distribution network, the green reward cost and the weight of the global out-of-limit penalty cost of the power distribution network; and updating network parameters in the strategy network and the evaluation network based on the reward score until the sum of the strategy loss and the evaluation loss of the deep reinforcement learning model is less than or equal to a preset loss.
  6. 6. The method of claim 5, wherein generating a reward score for the power distribution strategy according to the power distribution strategy comprises: alternating current power flow calculation is carried out on the power distribution network aiming at the power distribution strategy of each sub power distribution network to obtain voltage values and power tide values of all nodes in the power distribution network; accumulating the squares of the out-of-limit voltage and/or the squares of the out-of-limit power of the voltage value and the power tide value of each node, and taking the opposite number of the accumulated result as the global out-of-limit penalty cost; Multiplying the power grid exchange power with the local unit operation cost of the sub-distribution network to obtain the local operation cost; multiplying the output power and the energy storage exchange power under the condition that the power grid exchange power is positive, subtracting the result of the power grid exchange power, and multiplying the result by the unit green rewarding cost of the sub-distribution network to obtain green rewarding cost, or multiplying the sum of the output power and the energy storage exchange power under the condition that the power grid exchange power is positive, and multiplying the result of the power grid exchange power by the unit green rewarding cost of the sub-distribution network to obtain green rewarding cost; And accumulating the local running cost, the green rewarding cost and the global out-of-limit punishment cost according to a preset weight coefficient to obtain rewarding scores.
  7. 7. The method according to claim 6, wherein accumulating the square of the out-of-limit voltage and/or the square of the out-of-limit power for the voltage value and the power tidal current value of each node and taking the opposite number of the accumulated result as the global out-of-limit penalty cost comprises: When the voltage value of any node is smaller than the lower limit value of the out-of-limit voltage, square sum operation is carried out on the difference between the voltage value and the lower limit value of the out-of-limit voltage, and a first out-of-limit value is obtained; when the voltage value of any node is larger than the upper limit value of the out-of-limit voltage, square sum operation is carried out on the difference between the voltage value and the upper limit value of the out-of-limit voltage, so that a second out-of-limit value is obtained; under the condition that the power tide value of any node is smaller than the lower limit value of the power flow, square sum operation is carried out on the difference between the power tide value and the lower limit value of the power flow, and a third exceeding limit value is obtained; Under the condition that the power tide value of any node is larger than the upper limit value of the power flow, carrying out square sum operation on the difference between the power tide value and the upper limit value of the power flow to obtain a fourth exceeding limit value; And accumulating one or more of the first threshold value, the second threshold value, the third threshold value and the fourth threshold value, and taking the opposite number of the accumulated result as a global threshold crossing penalty cost.
  8. 8. The method of any of claims 5 to 7, wherein generating power distribution control instructions for the sub-distribution network based on the load demand power, the output power, the stored state of charge of the sub-distribution network, and the local node voltage of the sub-distribution network, comprises: Inputting the load demand power, the output power and the energy storage charge state of the sub-distribution network to a strategy network of the deep reinforcement learning model to obtain a target distribution strategy; and generating a power distribution control instruction for the sub power distribution network according to the target power distribution strategy.
  9. 9. A power distribution optimizing apparatus of a distributed power grid, wherein the power distribution optimizing apparatus of the distributed power grid is applied to a power distribution network including a plurality of sub-power distribution networks, the power distribution optimizing apparatus of the distributed power grid comprising: The load prediction module is used for predicting the load demand power of the sub-distribution network in the prediction period based on first date data and first meteorological data corresponding to the prediction period; The power output prediction module is used for predicting the power output of the sub power distribution network in the prediction period based on second date data and second weather data corresponding to the prediction period, wherein the power output comprises at least one or the sum of photovoltaic power output representing photovoltaic and wind speed power output representing wind power generation; The power distribution control instruction comprises energy storage exchange power and power grid exchange power, and the power distribution control instruction enables the sum of the calculated weights of the local running cost, green rewarding cost and global out-of-limit penalty cost of the power distribution network to be maximum according to the energy storage exchange power and the power grid exchange power.
  10. 10. An electronic device comprising a memory and a processor, the memory storing a computer program, the processor implementing the method of distribution optimization of a distributed power network according to any one of claims 1 to 8 when the computer program is executed.

Description

Distribution optimization method and device of distributed power grid and electronic equipment Technical Field The invention relates to the technical field of power distribution optimization, in particular to a power distribution optimization method, a power distribution optimization device and electronic equipment of a distributed power grid. Background At present, along with large-scale access of new energy sources such as photovoltaic, wind power and the like, the structure of a distributed power grid is increasingly complex, and the contradiction among the power supply stability, the economy and the environmental protection performance is increasingly outstanding. In the related art, the power distribution optimization method mainly adopts a central controller to collect information of each subordinate power distribution network, and a set of scheduling plans for coping with uncertainty is generated through a random planning or robust optimization method. However, this method is difficult to compromise the integrated benefit game of the local running cost, green energy excitation and global safety constraint of the power grid for each distributed sub-power distribution network, so that the sub-power distribution network is not matched and cannot execute corresponding power distribution instructions. Disclosure of Invention The invention provides a distribution optimization method, a distribution optimization device, electronic equipment, a storage medium and a computer program product of a distributed power grid, so that comprehensive benefits of each distributed sub-distribution network are considered. In a first aspect, the present invention provides a method of power distribution optimization for a distributed power grid, the method being applied to a power distribution network comprising a plurality of sub-power distribution networks, the method comprising: Predicting load demand power of the sub-distribution network in a prediction period based on first date data and first meteorological data corresponding to the prediction period; predicting the output power of the sub-distribution network in the prediction period based on second date data and second meteorological data corresponding to the prediction period, wherein the output power comprises at least one or the sum of photovoltaic output power representing photovoltaic and wind speed output power representing wind power generation; generating a distribution control instruction for the sub-distribution network according to the load demand power, the output power, the energy storage charge state of the sub-distribution network and the local node voltage of the sub-distribution network, wherein the distribution control instruction comprises energy storage exchange power and grid exchange power, and the distribution control instruction enables the sum of the local running cost, the green rewarding cost and the global out-of-limit punishment cost of the sub-distribution network, which are calculated according to the energy storage exchange power and the grid exchange power, to be maximum. In one embodiment of the present invention, the first day data includes time data and a date type, the first weather data includes temperature data and a weather type, and predicting load demand power of the sub-distribution network in a prediction period based on the first day data and the first weather data corresponding to the prediction period includes: Constructing a first training data set, wherein the first training data set comprises a first input data set and a first output data set corresponding to the first input data set, the first input data set comprises months corresponding to the time data, the date type, the temperature data and the weather type, and the first output data set comprises historical load power corresponding to unit area; Converting the first input dataset into a first historical condition vector of numerical value type; training the load prediction model by taking the first historical condition vector as an input of the load prediction model and taking historical load power corresponding to the first historical condition vector as an output of the load prediction model; converting the first date data and the first meteorological data into first condition vectors, and inputting the first condition vectors into the trained load prediction model to obtain standard load demand power in unit area; and multiplying the standard load demand power by the area covered by the sub power distribution network to obtain the load demand power of the sub power distribution network in the prediction period. In one embodiment of the present invention, the load prediction model is an improved generation countermeasure network model, the improved generation countermeasure network model including a generator and a arbiter, wherein training the load prediction model with the first historical condition vector as an input of the load prediction model and the h