Search

CN-121813340-B - Optical storage and charge dynamic power distribution control method and system based on AI prediction

CN121813340BCN 121813340 BCN121813340 BCN 121813340BCN-121813340-B

Abstract

The invention discloses an AI prediction-based optical storage and charging dynamic power distribution control method and system, wherein the method collects photovoltaic, energy storage, charging, power grid and environment multidimensional operation data through a sensor network, and a history and real-time data set is obtained through preprocessing; the method comprises the steps of training an AI combined prediction model by using a historical data set, outputting a photovoltaic output and charging load demand prediction result, inputting system operation constraint condition parameters to define an optimization boundary, solving power instruction reference values of each power module through a multi-objective optimization algorithm by combining the prediction result and the constraint condition, realizing global optimal allocation, correcting the reference values based on hardware operation parameters and deviation between the prediction and real-time values, issuing instructions, monitoring operation and triggering fault protection. The method effectively solves the problems of stiff power dispatching, low photovoltaic absorption rate, large energy storage life loss, obvious power grid impact, insufficient prediction precision, high energy transmission loss and weak multi-objective coordination of the traditional system.

Inventors

  • LIU JINZAN
  • WANG HAILIANG
  • LIANG JIANLEI

Assignees

  • 天津安捷物联科技股份有限公司

Dates

Publication Date
20260508
Application Date
20260310

Claims (9)

  1. 1. The optical storage and charging dynamic power distribution control method based on AI prediction is characterized by comprising the following steps: (1) The data acquisition and preprocessing, namely acquiring multidimensional operation data of the optical storage and filling system through a sensor network, and preprocessing the acquired multidimensional operation data to obtain a historical data set and a real-time data set; (2) Training and predicting an AI (advanced technology) model, namely training an AI combined prediction model by using the historical data set, and processing the real-time data set by using the trained AI combined prediction model to obtain a prediction result of photovoltaic output and charging load requirements; (3) Inputting constraint condition parameters in the running process of the optical storage and filling system, and defining a boundary range of power optimization distribution; (4) The global power optimization distribution, namely solving a power instruction reference value of a designated power module by adopting a multi-objective optimization algorithm by combining the prediction result obtained in the step (2) and the constraint condition parameters input in the step (3) so as to realize the global optimal power distribution planning under multiple objectives; (5) Real-time power correction and execution, namely, based on the deviation of the hardware operation parameters and the predicted values and the deviation of the hardware operation parameters and the real-time values, carrying out real-time correction on the power instruction reference value obtained in the step (4), transmitting the corrected power instruction to a designated power module, monitoring the operation state of the designated power module in real time, and triggering a fault protection mechanism; (6) Continuously monitoring the deviation condition of a predicted value and a real-time value, and adaptively updating parameters of the AI joint prediction model according to deviation feedback to enable the AI joint prediction model to adapt to the dynamic changes of environment and load; in the step (4), the multi-objective optimization algorithm is a particle swarm algorithm, and the objective function expression is: ; in the formula, 、 、 As a weight coefficient, adjusting according to scene requirements by a hierarchical analysis method; At a rate of light Fu Xiaona, the light is, , In order for the photovoltaic to actually dissipate power, The photovoltaic output predicted value; the energy storage device is an optimal state of charge for energy storage and is used for considering charge and discharge capacity and service life; for the benefit of the peak-valley electricity price arbitrage, , The electric power is taken from the grid at the valley time, The electricity price is the electricity price at the valley time, For peak-time power grid electricity taking power, Peak electricity price; carrying out real-time correction on the power instruction reference value obtained in the step (4), wherein the correction comprises hardware parameter correction; the hardware parameter correction process is used for collecting the voltage of the direct current bus And module power transmission loss If the DC bus voltage is If the energy storage discharge power is lower than the set lower limit, increasing the energy storage discharge power or the power taking power of the power grid; if the DC bus voltage is And if the voltage is higher than the set upper limit, increasing the stored charge power or reducing the photovoltaic output power.
  2. 2. The AI-prediction-based optical storage-charging dynamic power distribution control method of claim 1, wherein in step (1), the multidimensional operation data includes photovoltaic-side data, energy storage-side data, charging-side data, grid-side data, and environmental-side data; the photovoltaic side data comprise irradiance, component temperature, output power and MPPT working state of the photovoltaic array; the energy storage side data comprise energy storage SOC, charge-discharge current, single voltage and battery temperature; the charging side data comprise a charging pile connection state, charging demand power, charging duration and user charging habit labels; The power grid side data comprise power grid voltage, frequency, peak-valley electricity price time period and grid-connected power limiting instructions; The environmental side data includes real-time weather data and future weather forecast data.
  3. 3. The AI-prediction-based optical storage-charging dynamic power distribution control method of claim 1, wherein in step (1), preprocessing the collected multidimensional operating data comprises: abnormal data caused by sensor faults are removed by adopting 3 sigma criterion, and the 3 sigma criterion meets the following conditions , wherein, As a result of the single data value being, As a mean value of the data set, Standard deviation for the dataset; the missing data is complemented by adopting a linear interpolation method, and the linear interpolation formula is as follows: , in the formula, 、 As the coordinates of the data points are known, 、 For the coordinates of adjacent known data points, In order to miss the abscissa corresponding to the data, Is an interpolation result; normalizing the data to a 0-1 interval, wherein the normalization formula is as follows: , in the formula, As the raw data is to be processed, As a result of the minimum value of the data, As a result of the data maximum value, Is normalized data.
  4. 4. The AI-prediction-based optical storage-charging dynamic power distribution control method of claim 1, wherein in step (2), the AI joint prediction model is a multi-input-multi-output model based on LSTM and an attention mechanism; The input features of the AI joint prediction model comprise historical photovoltaic output data, historical charging load data, real-time environment data, weather forecast data of a future set period, period features and set user charging habit labels; the AI joint prediction model structure comprises an input layer, a 3-layer LSTM hidden layer, an attention layer and an output layer, wherein the input layer converts characteristic parameters into tensors with dimensions of 23 xT, and T is a time step; The number of neurons of each layer of the LSTM hidden layer is 128, the time sequence association relation for extracting the characteristics is extracted, and the LSTM unit state updating formula is as follows: , in the formula, For the current cell state, For the output of the forgetting gate, In order to be in the state of the unit at the previous time, For the output of the input gate, As a result of the candidate cell state, Multiplying the elements; the attention layer distributes weight to the feature vector output by the LSTM layer, and the weight calculation formula is as follows: , in the formula, The attention weight for the t-th feature, An attention score for the T-th feature, T being the feature sequence length; output layer outputs photovoltaic output prediction curve of future set period Prediction curve of charging load demand The prediction error satisfies: , in the formula, As a result of the fact that the value, In order to be able to predict the value, Is the set prediction error deviation rate threshold.
  5. 5. The AI-prediction-based optical storage-charge dynamic power distribution control method of claim 1, wherein in step (3), the constraint condition includes: Energy storage SOC constraints: , in the formula, For the energy storage of the real-time state of charge, In order to store the energy in the minimum state of charge, Is the maximum state of charge of the stored energy; grid-connected power constraint: , in the formula, The power command is interacted with for the power grid, The upper limit of grid-connected power is determined by a grid dispatching instruction; Power balance constraint: , in the formula, For the photovoltaic output power command, For the energy storage charge/discharge power instruction, the charge is positive, the discharge is negative, A power command is required for the charging load.
  6. 6. The AI-prediction-based optical storage-charging dynamic power distribution control method according to claim 1, wherein in step (5), the power instruction reference value obtained in step (4) is corrected in real time, including deviation correction; The deviation correcting process, if Readjusting the energy storage and power command of the power grid according to the real-time photovoltaic output, wherein, For the real-time photovoltaic output to be applied, As a predicted value of the output of the photovoltaic power, Is the set deviation rate threshold.
  7. 7. The method for controlling dynamic power distribution of optical storage and charging based on AI prediction according to claim 1, wherein in step (6), in the process of adaptively updating the parameters of the AI joint prediction model according to deviation feedback, the triggering condition of the adaptive update is that when the deviation rate is established for 3 consecutive time periods, the model parameters are updated by adopting a gradient descent method, and the parameter update formula of the gradient descent method is as follows: , in the formula, As a parameter of the model, it is possible to provide, In order for the rate of learning to be high, Gradient of parameters for the loss function.
  8. 8. An AI-prediction-based optical storage-charging dynamic power distribution control system, employing the AI-prediction-based optical storage-charging dynamic power distribution control method according to any one of claims 1 to 7, characterized by comprising: The data acquisition and preprocessing unit is used for acquiring the multidimensional operation data of the optical storage and filling system through the sensor network, preprocessing the acquired multidimensional operation data and obtaining a historical data set and a real-time data set; The AI model training and predicting unit is used for training an AI combined predicting model by utilizing the historical data set, and processing the real-time data set through the trained AI combined predicting model to obtain a predicting result of photovoltaic output and charging load requirements; The constraint condition input unit is used for inputting constraint condition parameters in the operation process of the optical storage and filling system and defining a boundary range of power optimization distribution; The global power optimization distribution unit is used for solving the power instruction reference value of the designated power module by adopting a multi-objective optimization algorithm in combination with the prediction result obtained by the AI model training and predicting unit and the constraint condition parameters input by the constraint condition input unit so as to realize global optimal power distribution planning under multiple objectives; The real-time power correction and execution unit is used for correcting the power instruction reference value obtained by the global power optimization distribution unit in real time based on the hardware operation parameter and the predicted value and the deviation of the hardware operation parameter and the real-time value, transmitting the corrected power instruction to the appointed power module, monitoring the operation state of the appointed power module in real time and triggering a fault protection mechanism; And the model self-adaptive updating unit is used for continuously monitoring the deviation condition of the predicted value and the real-time value, and self-adaptively updating the parameters of the AI joint prediction model according to deviation feedback so as to enable the AI joint prediction model to adapt to the dynamic change of the environment and the load.
  9. 9. A computer storage medium having stored therein program code of an AI-prediction-based optical storage-charge dynamic power allocation control method, the program code comprising instructions for executing the AI-prediction-based optical storage-charge dynamic power allocation control method of any one of claims 1 to 7.

Description

Optical storage and charge dynamic power distribution control method and system based on AI prediction Technical Field The invention belongs to the technical field of new energy power system control and artificial intelligence application, and particularly relates to an optical storage and charging dynamic power distribution control method and system based on AI prediction. Background Photovoltaic power generation is an important form of renewable energy utilization, is widely applied by virtue of the advantages of cleanness and environmental protection, and the rapid increase of the charging demand of electric automobiles promotes the large-scale deployment of an optical storage and charging integrated system. The system realizes the on-site production and the absorption of energy sources by integrating the functions of photovoltaic power generation, energy storage and electric vehicle charging, reduces the energy transmission loss, relieves the concentrated impact of charging load on a power grid, and becomes an important development direction of the energy utilization field. However, the current optical storage and filling system still faces a plurality of technical challenges in the operation process, and the operation efficiency and reliability of the optical storage and filling system are limited. Firstly, photovoltaic power generation is obviously influenced by natural factors such as solar irradiance, ambient temperature, cloud layer movement and the like, output power shows strong intermittence, volatility and randomness, large dip can occur in a short time, and an electric vehicle charging load has the characteristics of uneven time distribution and uncertain requirements, so that the problem of space-time mismatch on two sides of a source-charge is outstanding, the real-time power balance of a system is difficult to maintain, photovoltaic light rejection occurs, and the energy utilization efficiency is low. Secondly, the traditional power distribution control mostly adopts a fixed threshold value or a simple regularization strategy, lacks the capacity of prospective prediction of photovoltaic output and charging load, and cannot effectively coordinate multi-objective requirements among photovoltaic consumption, energy storage life protection and operation economy. The strategy has poor adaptability to working condition change, and when photovoltaic power fluctuation or charging load suddenly changes, the strategy for energy storage charging and discharging and power grid interaction is difficult to adjust quickly, so that the problems of direct current bus voltage fluctuation, power grid frequency deviation and the like can be caused, and the stable operation of the system is influenced. Moreover, the existing prediction model depends on single-dimensional data, and multi-source heterogeneous data such as historical operation data, meteorological information, user charging habits and the like are not fully fused, so that the prediction accuracy is insufficient, and the short-term prediction error often exceeds 15%. Meanwhile, the model lacks of an adaptive updating mechanism, and prediction deviation is continuously enlarged easily caused by environmental change and load characteristic change after long-term operation, so that the rationality of a power distribution decision is affected. In addition, the charge and discharge control of the energy storage system does not carry out refined optimization on the charge state, the service life loss of a battery is accelerated by frequent deep charge and discharge, the constraint control of grid-connected power of the power grid is inaccurate, and the counter-current risk can be caused, so that the protection action of the power grid is triggered. In addition, each device of the optical storage and filling system is often from different manufacturers, communication protocols are not uniform, data splitting phenomenon exists, and a unified energy management core is lacked. The traditional control method is difficult to integrate multidimensional operation data of a photovoltaic side, an energy storage side, a charging side and a power grid side, and an optimization algorithm faces contradiction between decision accuracy and real-time response speed in engineering application, so that the real-time regulation and control requirement of dynamic power distribution cannot be met. Disclosure of Invention Therefore, the invention provides an AI prediction-based optical storage and charging dynamic power distribution control method and system, which solve the problems of stiff power dispatching, low photovoltaic absorption rate, large energy storage life loss, obvious power grid impact, insufficient prediction precision, high energy transmission loss and weak multi-objective coordination of the traditional optical storage and charging system. In order to achieve the above purpose, the invention provides a technical scheme that the optical storage and charging d