Search

CN-122022079-A - Campus comprehensive energy supply and demand optimal configuration method based on AI multi-mode prediction

CN122022079ACN 122022079 ACN122022079 ACN 122022079ACN-122022079-A

Abstract

The invention relates to the technical field of park comprehensive energy optimization, in particular to a park comprehensive energy supply and demand optimization configuration method based on AI multi-mode prediction, which comprises the following steps: the method comprises the steps of collecting meteorological time sequence, historical energy load curve, equipment operation condition record and real-time energy market price sequence multi-mode data of a park comprehensive energy system, forming a standardized multi-mode data cube through space-time alignment and missing value complement processing, inputting the standardized multi-mode data cube into an AI multi-mode prediction model to process different data modes in parallel, outputting multi-period comprehensive loads of cold and heat and electricity and predicted values of distributed photovoltaic and wind power generation power, constructing a multi-time scale operation optimization model by combining the real-time energy market price, solving and obtaining a scheduling plan covering the day, day and in real time, and determining power and output instructions of an energy storage, heat storage, gas internal combustion engine and an electric refrigerator. The method can eliminate the space-time deviation of the multi-source data, improve the prediction matching degree and realize the accurate and collaborative configuration of the comprehensive energy supply and demand of the park.

Inventors

  • LI KONGZHENG
  • WANG XIAOMING
  • CHEN XIAOFENG

Assignees

  • 广东百德朗科技有限公司

Dates

Publication Date
20260512
Application Date
20260413

Claims (10)

  1. 1. The park comprehensive energy supply and demand optimal configuration method based on AI multi-mode prediction is characterized by comprising the following steps: The method comprises the steps of collecting multi-mode history and real-time data of a park comprehensive energy system, wherein the multi-mode history and real-time data at least comprise a meteorological time sequence, a historical energy load curve, equipment operation condition records and a real-time energy market price sequence; carrying out space-time alignment and missing value complementation on the acquired multi-modal history and real-time data to form a standardized multi-modal data cube; Inputting the standardized multi-mode data cube into a pre-trained AI multi-mode prediction model, wherein the AI multi-mode prediction model processes different types of data modes in parallel and outputs cold, heat and electricity comprehensive load predicted values and power generation power predicted values of distributed photovoltaic and wind power of a plurality of time periods in the future of a park; Receiving a load predicted value and a power generation predicted value which are output by the AI multi-mode predicted model, and constructing a multi-time scale operation optimization model of a park comprehensive energy system by combining a real-time energy market price sequence; and solving the multi-time-scale operation optimization model to obtain an optimization scheduling plan covering a plurality of time periods in the day before, in the day and in real time, wherein the optimization scheduling plan comprises charge and discharge power of the electric energy storage equipment, heat storage and release power of the heat storage device and output instructions of the gas internal combustion engine and the electric refrigerator.
  2. 2. The AI-multimodal prediction-based park comprehensive energy supply and demand optimization configuration method according to claim 1, wherein the performing space-time alignment and missing value complementation on the collected multimodal history and real-time data comprises: identifying respective time stamps and data sampling frequencies of the meteorological time sequence, the historical energy load curve, the equipment operation condition record and the real-time energy market price sequence; resampling and interpolating various data with different time stamps and sampling frequencies according to preset unified time reference and sampling interval, so that all data are aligned in time dimension; detecting missing data points in the data sequence after space-time alignment, generating a complement value to fill the missing data points by adopting a double fitting method based on adjacent time interval type data and associated modal data trend, and finally forming a standardized multi-modal data cube which is continuous in time and associated in a modal.
  3. 3. The AI-multimodal prediction-based park comprehensive energy supply and demand optimization configuration method of claim 2, wherein inputting the standardized multimodal data cube into a pre-trained AI multimodal prediction model comprises: The AI multi-modal prediction model comprises a meteorological feature extraction branch, a load feature extraction branch, an equipment working condition feature extraction branch and a market feature extraction branch, and respectively processes corresponding data in the standardized multi-modal data cube; the meteorological feature extraction branch extracts meteorological feature vectors of future time periods from a meteorological time sequence, the load feature extraction branch extracts load time sequence feature vectors from a historical energy load curve, the equipment working condition feature extraction branch extracts equipment state feature vectors from equipment operation working condition records, and the market feature extraction branch extracts price fluctuation feature vectors from a real-time energy market price sequence; Fusing the meteorological feature vector, the load time sequence feature vector, the equipment state feature vector and the price fluctuation feature vector, and inputting the fused meteorological feature vector, the load time sequence feature vector, the equipment state feature vector and the price fluctuation feature vector into a space-time attention prediction network in the AI multi-mode prediction model; The spatiotemporal attention prediction network outputs a park cold load predicted value, a heat load predicted value, an electric load predicted value, a photovoltaic power predicted value and a wind power predicted value for a plurality of time periods in the future.
  4. 4. The optimal configuration method for the comprehensive energy supply and demand of the campus based on the AI multi-mode prediction according to claim 3, wherein the method for receiving the load predicted value and the power generation predicted value output by the AI multi-mode prediction model and combining the real-time energy market price sequence to construct a multi-time scale operation optimization model of the comprehensive energy system of the campus comprises the following steps: Targeting overall operating costs of the campus integrated energy system, including electricity purchasing costs from the upper grid, gas fuel costs, and equipment operating and maintenance costs; Constructing constraint conditions of the multi-time-scale operation optimization model, wherein the constraint conditions comprise electric power balance constraint, thermal power balance constraint, cold power balance constraint, electric energy storage equipment operation constraint, heat storage device operation constraint, and physical operation upper and lower limits and climbing rate constraint of key energy conversion equipment of a gas internal combustion engine and an electric refrigeration machine; The multi-time-scale running optimization model divides a future day into three time scales of day front, day inner and real time, the decision device starts and stops at the day front scale and transfers energy in a long time scale, the power plan rolling optimization is carried out in the day inner scale for a plurality of hours, and the power deviation balance of the minute level is processed in the real time scale.
  5. 5. The AI-multimodal-prediction-based park comprehensive energy supply and demand optimization configuration method of claim 4, wherein solving the multi-time-scale operation optimization model to obtain an optimized scheduling plan covering a plurality of time periods in the day, the day and the real time comprises: Solving the multi-time-scale operation optimization model by adopting a decomposition coordination algorithm, and decomposing the original problem into a day-ahead main problem and a plurality of day-ahead sub-problems; solving a day-ahead main problem to obtain a planned charge-discharge power curve of the electric energy storage equipment, a planned heat storage and release power curve of the heat storage device and a planned output curve of the gas internal combustion engine and the electric refrigerator which are taken as intervals of one hour in twenty four hours in the future; on the basis of the solving result of the day-ahead main problem, rolling to solve the day-ahead sub-problem covering a plurality of hours in the future, and refining and correcting the planned charge-discharge power curve of the electric energy storage device, the planned heat storage and release power curve of the heat storage device and the planned output curve of the key device to form a day-ahead rolling plan taking fifteen minutes as an interval; And forming an optimized scheduling plan by the daily master problem solving result and the daily rolling plan together.
  6. 6. The AI-multimodal-prediction-based park comprehensive energy supply and demand optimization configuration method as set forth in claim 5, further comprising: Decomposing a day-ahead plan in the optimized scheduling plan into a specific equipment start-stop instruction and a reference power curve, and issuing the specific equipment start-stop instruction and the reference power curve to each energy conversion and storage equipment for execution; In the intra-day execution stage, correcting an intra-day rolling plan in the optimized scheduling plan based on the latest ultra-short-term load prediction and power generation prediction, and generating a device power adjustment instruction in a current time window; In a real-time execution stage, monitoring the actual running power of each energy conversion and storage device, comparing the actual running power with the device power adjustment instruction, and generating a power compensation signal through real-time feedback control; Aggregating the equipment start-stop instruction, the reference power curve, the equipment power adjustment instruction and the power compensation signal to form a final control instruction set of each energy equipment; According to the final control instruction set of each energy device, driving physical devices in the park comprehensive energy system to execute corresponding energy production, conversion, storage and consumption actions, and completing supply and demand optimal configuration; the decomposing the day-ahead plan in the optimized scheduling plan into a specific equipment start-stop instruction and reference power curve comprises the following steps: Analyzing a planned charge-discharge power curve of the electric energy storage equipment at the day-ahead part in the optimized dispatching plan, and generating a charge-discharge state instruction and a reference charge-discharge power value of the electric energy storage equipment at the starting moment of each hour; Analyzing a planned heat storage and release power curve of the heat storage device at the day front part in the optimized dispatching plan, and generating a heat storage or release state instruction and a reference power value of the heat storage device at the initial moment of each hour; And analyzing the planned output curves of the gas internal combustion engine and the electric refrigerator at the day front part in the optimized dispatching plan to generate a start-stop state instruction and a reference running power value of the gas internal combustion engine and the electric refrigerator in each hour.
  7. 7. The method for optimizing configuration of integrated energy supply and demand in a campus based on AI multi-modal prediction according to claim 6, wherein the performing the step in the day, based on the latest ultra-short term load prediction and power generation prediction, corrects the intra-day rolling plan in the optimized scheduling plan, includes: In the daily execution stage, invoking the AI multi-mode prediction model to execute ultra-short-term prediction to obtain cold, heat and electric load predicted values and photovoltaic and wind power generation predicted values updated in a plurality of hours in the future; Updating related parameters in the multi-time scale operation optimization model by using the latest obtained ultra-short term predicted value, and re-solving the intra-day rolling plan within a plurality of hours to obtain an updated equipment power plan curve; Comparing the updated equipment power plan curve with an intra-day rolling plan in the original optimized scheduling plan, and calculating the power adjustment quantity of each equipment; and generating equipment power adjustment instructions of the electric energy storage equipment, the heat storage device, the gas internal combustion engine and the electric refrigerator in the current time window according to the power adjustment quantity of each equipment.
  8. 8. The method for optimizing configuration of integrated energy supply and demand in a campus based on AI multi-modal prediction according to claim 7, wherein monitoring the actual operating power of each energy conversion and storage device and comparing with the device power adjustment command in the real-time execution phase, generating a power compensation signal through real-time feedback control, comprises: Collecting the actual output power or the storage power of the electric energy storage equipment, the heat storage device, the gas internal combustion engine and the electric refrigerator in real time; Comparing the actual power of each device with a power set value required in the device power adjustment instruction corresponding to the current moment, and calculating to obtain a power deviation value; The power deviation value is input to a pid controller, which outputs a power compensation signal for canceling the power deviation.
  9. 9. The method for optimizing configuration of a campus integrated energy supply and demand based on AI multi-modal prediction according to claim 8, wherein aggregating the device start-stop command, the reference power curve, the device power adjustment command, and the power compensation signal to form a final set of energy device control commands comprises: For the electric energy storage equipment, superposing a charge and discharge state instruction in the equipment start-stop instruction, a reference charge and discharge power value in the reference power curve, a power adjustment amount in the equipment power adjustment instruction and a power compensation value in the power compensation signal to obtain a final instantaneous power control instruction of the electric energy storage equipment; For the heat storage device, superposing a heat storage state instruction in the equipment start-stop instruction, a reference power value in the reference power curve, a power adjustment amount in the equipment power adjustment instruction and a power compensation value in the power compensation signal to obtain a final instantaneous power control instruction of the heat storage device; And superposing a start-stop state instruction in the equipment start-stop instruction, a reference running power value in the reference power curve, a power adjustment amount in the equipment power adjustment instruction and a power compensation value in the power compensation signal to obtain a final instantaneous power control instruction of the gas internal combustion engine and the electric refrigerator.
  10. 10. The method for optimizing configuration of supply and demand of integrated energy resources in a campus based on AI multi-modal prediction according to claim 9, wherein driving physical devices in the integrated energy resource system to execute corresponding energy production, conversion, storage and consumption actions according to the final control instruction set of each energy resource device, and completing configuration of supply and demand optimization comprises: converting a final instantaneous power control instruction of the electric energy storage device into a driving signal of a power electronic converter to control the charging or discharging process of an electric energy storage battery; Converting a final instantaneous power control instruction of the heat storage device into a water pump and valve opening adjusting signal to control the heat storage or heat release process of the heat storage tank; Converting the final instantaneous power control instruction of the gas internal combustion engine into a fuel supply adjusting signal and a generator excitation adjusting signal to control the output electric power of the gas internal combustion engine; and converting the final instantaneous power control instruction of the electric refrigerator into a compressor frequency adjusting signal to control the refrigerating power output of the electric refrigerator.

Description

Campus comprehensive energy supply and demand optimal configuration method based on AI multi-mode prediction Technical Field The invention relates to the technical field of park comprehensive energy optimization, in particular to a park comprehensive energy supply and demand optimal configuration method based on AI multi-mode prediction. Background The supply and demand optimization configuration of the existing park comprehensive energy system is realized by carrying out energy load and distributed new energy power generation prediction by relying on single-mode data, the adopted data only comprises a single type of energy load curve or basic meteorological parameters, the energy system operation optimization model is usually constructed by adopting a single time scale, the scheduling scheme is only formulated for a single time period, and the output regulation and control of various energy devices and the scheduling of energy storage devices are carried out by relying on single prediction data and fixed price parameters. In the prior art, the related data of different sources of energy sources are not subjected to space-time alignment treatment, missing values in the data cannot be effectively complemented, a standardized multi-mode data carrier is not formed, a prediction model can only process a single type of data mode, multi-dimensional prediction of cold, heat, electric loads and wind-solar power generation power cannot be synchronously completed, and a prediction result has deviation with the actual energy supply and demand conditions of a park. The optimization model with single time scale can not be matched with the real-time dynamic operation demands of the energy system in the park before, during the day, and the scheduling logic is constructed without combining the real-time energy market price sequence, the power output instructions of the electric energy storage and heat storage device, the gas internal combustion engine and the electric refrigerator can not form a multi-period coordinated scheduling scheme, and the rationality and the suitability of the energy supply and demand configuration are insufficient. According to the scheme, space-time alignment, missing value completion and standardization processing of the multi-mode energy data of the park are required to be completed, multi-dimensional energy parameter prediction is achieved through parallel processing of multi-type data modes by the AI multi-mode prediction model, a multi-time scale operation optimization model fusing a prediction result and a real-time energy market price is required to be constructed, and a refinement equipment scheduling instruction covering multiple time periods is output. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a park comprehensive energy supply and demand optimal configuration method based on AI multi-mode prediction. In order to achieve the purpose, the invention adopts the following technical scheme that the park comprehensive energy supply and demand optimizing configuration method based on AI multi-mode prediction comprises the following steps: The method comprises the steps of collecting multi-mode history and real-time data of a park comprehensive energy system, wherein the multi-mode history and real-time data at least comprise a meteorological time sequence, a historical energy load curve, equipment operation condition records and a real-time energy market price sequence; carrying out space-time alignment and missing value complementation on the acquired multi-modal history and real-time data to form a standardized multi-modal data cube; Inputting the standardized multi-mode data cube into a pre-trained AI multi-mode prediction model, wherein the AI multi-mode prediction model processes different types of data modes in parallel and outputs cold, heat and electricity comprehensive load predicted values and power generation power predicted values of distributed photovoltaic and wind power of a plurality of time periods in the future of a park; Receiving a load predicted value and a power generation predicted value which are output by the AI multi-mode predicted model, and constructing a multi-time scale operation optimization model of a park comprehensive energy system by combining a real-time energy market price sequence; and solving the multi-time-scale operation optimization model to obtain an optimization scheduling plan covering a plurality of time periods in the day before, in the day and in real time, wherein the optimization scheduling plan comprises charge and discharge power of the electric energy storage equipment, heat storage and release power of the heat storage device and output instructions of the gas internal combustion engine and the electric refrigerator. As a further aspect of the present invention, the performing space-time alignment and missing value complementation on the collected multi-modal history and real-time data in