Energy supply and demand allocation method under meteorological influence

Abstract

The application discloses a kind of meteorological influence under energy supply and demand allocation method, belong to energy supply and demand allocation technical field, including the following steps: step one: collection multidimensional meteorological data, energy supply and demand historical data and real-time operation data;Meteorology, energy characteristic tensor is constructed;Step two: based on ensemble prediction and quantile regression model, generate multi-scale supply and demand probability prediction interval, output the probability density function of net load;Step three: according to meteorological forecast standard deviation and measured change rate calculation meteorological influence risk index, dispatching period is divided into stable period, fluctuation period and emergency period;The application is by constructing multidimensional meteorological, energy characteristic tensor covering temperature and humidity index, wind cold index, cloud cover coefficient and other interactive characteristics, delay inertia characteristics and spatial correlation characteristics, in combination with quantile regression and probability density output of ensemble prediction, significantly refine the nonlinear influence description of meteorological multi-factor cooperation and lag effect on supply and demand.

Description

Technical Field

[0001] This invention relates to the field of energy supply and demand allocation technology, and more specifically, to a method for energy supply and demand allocation under the influence of meteorological conditions. Background Technology

[0002] Existing energy supply and demand dispatching technologies typically rely on deterministic load forecasting and renewable energy output forecasting, and conduct day-ahead or intraday dispatching based on fixed confidence levels or simple reserve capacity. However, these conventional methods generally suffer from significant drawbacks, such as the coarse-grained modeling of meteorological and energy response relationships. They not only consider a few isolated meteorological factors such as temperature and sunshine, but also completely ignore the nonlinear interactive effects of humidity, wind speed, and cloud cover, as well as the time-delay inertia effect of meteorological changes. Furthermore, the forecasting and dispatching time scales are severely disconnected, lacking the probabilistic dynamic absorption capacity for ensemble forecast biases, thus failing to proactively adjust according to continuously deteriorating meteorological trends. The existing dispatching system is mostly an open-loop control architecture. After the dispatching instructions are issued, there is a lack of online error inversion and closed-loop correction mechanisms based on high-frequency measured meteorological and power grid operation data. This causes the prediction deviation to accumulate continuously during the daily rolling process. Moreover, when extreme weather causes communication congestion or computing resource shortages, conventional centralized optimization models are easily paralyzed. They are completely lacking in the ability to adaptively sample and decentralized autonomously respond to micro-meteorological changes. Therefore, they cannot meet the comprehensive requirements of new power systems for the accuracy, robustness and resilience of supply and demand dispatching in scenarios with a high proportion of new energy access.

[0003] Therefore, we have made improvements to this and proposed a method for energy supply and demand allocation under the influence of meteorological conditions. Summary of the Invention

[0004] In view of the above-mentioned problems in the existing technology, the purpose of this invention is to provide a method for energy supply and demand allocation under the influence of meteorology.

[0005] To solve the above problems, the technical solution adopted by the present invention is as follows: including the following steps: Step 1: Collect multidimensional meteorological data, historical energy supply and demand data, and real-time operational data; construct meteorological and energy characteristic tensors; Step 2: Based on ensemble forecasting and quantile regression models, generate multi-scale supply and demand probability prediction intervals and output the probability density function of net load; Step 3: Calculate the meteorological impact risk index based on the standard deviation of meteorological forecasts and the measured rate of change, and divide the scheduling period into stable period, fluctuating period and emergency period; Step 4: Using a time-varying risk preference robust optimization model, the confidence level of the opportunity constraint is dynamically adjusted according to the time period type to obtain the unit combination and output scheduling instructions; Step 5: After executing the scheduling command, collect real-time meteorological observation data, invert the micro-meteorological field through extended Kalman filter, and correct the prediction model parameters and short-term scheduling command online; feed the correction results back to the feature construction and prediction steps to form a closed loop.

[0006] Preferably, the construction of meteorological and energy feature tensors specifically includes: The temperature, humidity, wind speed, cloud cover, air pressure, precipitation probability, shortwave radiation, historical load, and historical renewable energy output data in numerical weather forecasts are aligned to a 15-minute time grid, and outlier interpolation is performed.

[0007] Preferably, the generation of multi-scale supply and demand probability prediction intervals specifically includes: A quantile regression forest model is used to establish a load probability predictor and a new energy output probability predictor in parallel. The input is the feature tensor described in claim 2, and the output is the load value and the new energy output value at the 10%, 50%, and 90% quantiles. Multiple disturbance members of numerical weather prediction are obtained, and each member is input into the two forecasters mentioned above to obtain multiple forecast intervals. The mean and standard deviation of net load at each time section are calculated. The probability density function of the net load is expressed by the following mathematical equation: ; in: For time period The net load random variable; The mean is variance is The normal distribution; For time period The net load mean is calculated as the arithmetic mean of the predictions from multiple set members; For time period The net load standard deviation characterizes the forecast uncertainty.

[0008] Preferably, the calculation of the meteorological impact risk index and the division of time periods specifically include: Define time period Meteorological impact risk index Calculated by the following mathematical equation: ; in: For time period The meteorological impact risk index; For time period The standard deviation of net load; This represents the average standard deviation of net load at the same time over the past 7 days. It is the one with the largest absolute value of the rate of change among temperature, wind speed, and humidity; This is an extreme weather warning signal; , , For the preset weighting coefficients, satisfy + + =1; when The period is divided into a stable period, when The time is divided into fluctuation periods, when The time period is divided into emergency periods; the output time period type label sequence and the corresponding risk threshold are output.

[0009] Preferably, the time-varying risk preference robust optimization model specifically includes: The objective function is established as minimizing the sum of total operating costs and expected load shedding and wind / solar curtailment penalties; The constraints include power balance constraints, unit ramping constraints, energy storage SOC constraints, line power flow constraints, and opportunity constraints. The allowable load shedding probability of the opportunity constraint is dynamically adjusted according to the time period type: the allowable load shedding probability is 0.1 during the stable period, 0.05 during the fluctuating period, and 0.01 during the emergency period; Dynamic reserve capacity configuration is represented by the following set of mathematical equations: ; in: , Time periods Upper and lower reserve capacities; , These are the basic and lower reserve capacities, respectively; , Reserve factor: 1.0 for stable periods, 1.5 for fluctuating periods, and 2.5 for emergency periods; The standard deviation of net load forecast; This is for backup response time; An affine adjustable robust optimization strategy is adopted to transform chance constraints into deterministic second-order cone constraints. The column AND constraint generation algorithm is used to solve the problem and output day-ahead, intraday, and real-time three-level scheduling instructions for each power generation unit, energy storage, and demand response.

[0010] Preferably, the inversion of the micro-meteorological field via extended Kalman filtering specifically includes: A meteorological and physical model for attenuation was established to express the attenuation of power line carrier communication signals as a linear combination of relative humidity, rainfall rate, and fog concentration. Using the deviation of the meteorological forecast system as the state variable and the attenuation and the measured deviation of the meteorological station as the observation vector, the state variable is updated in real time through the extended Kalman filter recursive formula to obtain the corrected micro-meteorological field. The corrected microclimate field is fed back to the feature construction step to update the feature tensor.

[0011] Preferably, the parameters of the online correction prediction model specifically include: A lightweight online learning module was built, which only updates the parameters of the last two fully connected layers of the quantile regression forest model; Stochastic gradient descent is used, with quantile loss as the loss function. After each new 15-minute set of measured data is received, a parameter update is performed, and the learning rate decays exponentially with the number of iterations. The updated model parameters are immediately used for the next round of predictions.

[0012] Preferably, it also includes a self-organizing demand response mechanism based on evolutionary game theory during the emergency period: When the determined time period is an emergency period and the communication system load rate exceeds 80%, the self-organizing mode is triggered. Main station broadcast power deficit and emergency pricing coefficient; Each flexible load node acts as an intelligent game player, running an evolutionary game algorithm locally: each game player determines its own interruptibility... The bidding curve is calculated based on quantity and user comfort tolerance coefficient; the game players exchange bidding information through power line carrier or nearby wireless communication, simulating the ant colony pheromone update rule, and converge to Nash equilibrium after multiple rounds of iteration; The main station only performs security checks on the final response plan. Once the check passes, it issues the execution command. After the emergency period ends, the system automatically switches back to the centralized robust optimization mode.

[0013] Preferably, the closed-loop correction further includes the modification of short-time scheduling instructions: If the absolute value of the calculated prediction error exceeds 5% of the rated capacity, a short-term correction scheduling will be triggered. The short-term correction scheduling adopts the model predictive control method, with a prediction time domain of 15 minutes and a control time domain of 5 minutes. The objective function is to minimize the correction cost and tracking deviation. The corrective instructions obtained within 5 minutes are sent to the execution mechanism; these corrective instructions do not change the main scheduling plan given in claim 5, but only make fine adjustments within a local time.

[0014] Preferably, all steps are executed on a rolling basis with a basic cycle of 15 minutes. The day-ahead forecast is executed once a day, the intraday forecast is executed once every 4 hours, the real-time forecast and rolling scheduling are executed once every 15 minutes, and the correction and adaptive mechanism runs continuously at a frequency of minutes.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing a multidimensional meteorological and energy feature tensor that includes interactive features such as temperature and humidity index, wind chill index, and cloud cover coefficient, as well as time delay inertia features and spatial correlation features, and combining quantile regression and probability density output of ensemble forecasts, the nonlinear impact of meteorological multi-factor synergy and lag effects on supply and demand is significantly refined, enabling the forecast interval to dynamically reflect the uncertainty amplitude, thereby fundamentally overcoming the limitations of coarse modeling and deterministic forecasting.

[0016] 2. Based on the dynamic time period division and robust optimization of time-varying confidence level of the meteorological impact risk index (RI), coupled with dynamic backup configuration that is positively correlated with the forecast standard deviation and the square root of the response time, the conservative performance of the scheduling is actively and smoothly adjusted as the weather deteriorates, which completely solves the problems of fixed backup and blind conservatism or recklessness.

[0017] 3. Based on the real-time correction of the micro-meteorological field by fusing extended Kalman filter with power line carrier communication signal attenuation inversion, and the mechanism of adaptively adjusting the sampling frequency and the number of ensemble perturbation members based on chaotic Lyapunov exponent, not only is online dynamic compensation for numerical weather forecast deviations achieved, but also the computational load and correction granularity can be actively adjusted according to the error divergence rate, effectively curbing the accumulation of errors during the intraday rolling process and significantly improving the time-frequency matching of forecast and control.

[0018] 4. Construct a closed-loop architecture for the entire process from data acquisition and decision execution to error inversion, online updating of model parameters, short-term model prediction and control correction, and finally feedback to feature construction and scheduling instructions. When extreme weather causes communication congestion, the emergency period automatically switches to a self-organized demand response mode based on evolutionary game theory and ant colony pheromones. This enables the power grid to spontaneously converge to supply and demand balance through distributed intelligent agents even when central computing resources are limited or communication is degraded, giving the system unprecedented resilience and autonomous survival capabilities. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a method for coordinating energy supply and demand under meteorological influences, provided in this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] A method for coordinating energy supply and demand under the influence of meteorological conditions includes the following steps: Step 1: Collect multidimensional meteorological data, historical energy supply and demand data, and real-time operational data; construct meteorological and energy characteristic tensors; Step 2: Based on ensemble forecasting and quantile regression models, generate multi-scale supply and demand probability prediction intervals and output the probability density function of net load; Step 3: Calculate the meteorological impact risk index based on the standard deviation of meteorological forecasts and the measured rate of change, and divide the scheduling period into stable period, fluctuating period and emergency period; Step 4: Using a time-varying risk preference robust optimization model, the confidence level of the opportunity constraint is dynamically adjusted according to the time period type to obtain the unit combination and output scheduling instructions; Step 5: After executing the scheduling command, collect real-time meteorological observation data, invert the micro-meteorological field through extended Kalman filter, and correct the prediction model parameters and short-term scheduling command online; feed the correction results back to the feature construction and prediction steps to form a closed loop.

[0022] Furthermore, the construction of meteorological and energy characteristic tensors specifically includes: Align the temperature, humidity, wind speed, cloud cover, air pressure, precipitation probability, shortwave radiation, historical load, and historical renewable energy output data from numerical weather forecasts to a 15-minute time grid and perform outlier interpolation. Construct interactive features: temperature and humidity index, wind chill index, and cloud cover coefficient; Construction of time delay characteristics: The first difference of meteorological elements over the past 3 hours and the weighted average temperature over the past 6 hours show an exponential decay in weight; Constructing spatial characteristics: Divide the regional power grid into 1km×1km geographical grids, calculate the correlation coefficient matrix between meteorological elements and load and power output of new energy power plants in each grid, and retain factors with an absolute value of correlation coefficient greater than 0.3; After standardizing all the above features, a tensor with dimensions of time × spatial grid × feature type is formed.

[0023] Furthermore, generating multi-scale supply and demand probability forecast intervals specifically includes: A quantile regression forest model is used to build a load probability predictor and a new energy output probability predictor in parallel. The input is the feature tensor of claim 2, and the output is the load value and the new energy output value at the 10%, 50%, and 90% quantiles. Multiple disturbance members of numerical weather prediction are obtained, and each member is input into the two forecasters mentioned above to obtain multiple forecast intervals. The mean and standard deviation of net load at each time section are calculated. The probability density function of net load is expressed by the following mathematical equation: ; in: For time period The net load random variable; The mean is variance is The normal distribution; For time period The net load mean is calculated as the arithmetic mean of the predictions from multiple set members; For time period The net load standard deviation characterizes the forecast uncertainty; The forecasts are generated at the following scales: day-ahead (one interval per hour for the next 24 hours), intraday (one interval per 15 minutes for the next 4 hours), and real-time rolling (one interval per minute for the next 15 minutes to 2 hours).

[0024] Furthermore, the calculation of the meteorological impact risk index and the division of time periods specifically include: Define time period Meteorological impact risk index Calculated by the following mathematical equation: ; in: For time period Meteorological impact risk index (dimensionless). For time period The standard deviation of net load; This represents the average standard deviation of net load at the same time over the past 7 days. It is the one with the largest absolute value of change rate among temperature (°C / h), wind speed (m / s / h), and humidity (% / h); This is an extreme weather warning signal (1 for warnings, 0 for no warnings). , , For the preset weighting coefficients, satisfy + + =1; when The period is divided into a stable period, when The time is divided into fluctuation periods, when The time period is divided into emergency periods; the output time period type label sequence and the corresponding risk threshold are output.

[0025] Furthermore, the robust optimization model for time-varying risk preferences specifically includes: The objective function is established as minimizing the sum of total operating costs and expected load shedding and wind / solar curtailment penalties; The constraints include power balance constraints, unit ramping constraints, energy storage SOC constraints, line power flow constraints, and opportunity constraints. The allowable load shedding probability of the opportunity constraint is dynamically adjusted according to the time period type: the allowable load shedding probability is 0.1 during the stable period, 0.05 during the fluctuating period, and 0.01 during the emergency period; Dynamic reserve capacity configuration is represented by the following set of mathematical equations: ; in: , Time periods Upper and lower reserve capacities; , These are the basic and lower reserve capacities, respectively; , Reserve factor: 1.0 for stable periods, 1.5 for fluctuating periods, and 2.5 for emergency periods; The standard deviation of net load forecast; This is the backup response time (taken as 15 minutes, equivalent to 0.25 hours). An affine adjustable robust optimization strategy is adopted to transform chance constraints into deterministic second-order cone constraints. The column AND constraint generation algorithm is used to solve the problem and output day-ahead, intraday, and real-time three-level scheduling instructions for each power generation unit, energy storage, and demand response.

[0026] Furthermore, the inversion of microclimate fields through extended Kalman filtering specifically includes: Power line carrier communication signal strength monitoring modules are installed at each node of the power grid to record the received signal levels between adjacent nodes. A meteorological and physical model for attenuation was established to express the attenuation of power line carrier communication signals as a linear combination of relative humidity, rainfall rate, and fog concentration. Using the deviation of the meteorological forecast system as the state variable and the attenuation and the measured deviation of the meteorological station as the observation vector, the state variable is updated in real time through the extended Kalman filter recursive formula to obtain the corrected micro-meteorological field. The corrected microclimate field is fed back to the feature construction step to update the feature tensor; Power line carrier communication signal attenuation data is also used to construct the spatial distribution of micro-meteorological fields: the inversion results of each line segment are mapped to the meteorological correction amount of the corresponding geographic grid point, and the high-resolution micro-meteorological field of the entire region is generated by Kriging interpolation, with the interpolation weight determined by the semi-variogram function.

[0027] Furthermore, the online calibration of prediction model parameters specifically includes: A lightweight online learning module was built, which only updates the parameters of the last two fully connected layers of the quantile regression forest model; Stochastic gradient descent is used, with quantile loss as the loss function. After each new 15-minute set of measured data is received, a parameter update is performed, and the learning rate decays exponentially with the number of iterations. The updated model parameters are immediately used for the next round of predictions; It also includes an adaptive sampling and correction frequency adjustment mechanism based on the chaotic Lyapunov exponent: maintaining a sliding window to store the prediction error sequence of the past hour, and using a small data volume method to calculate the maximum Lyapunov exponent of the sequence. The sampling period of a Kalman filter is determined by the following mathematical equation: ; in: For adaptive sampling period; =5 minutes, which is the basic sampling period; This is a proportionality coefficient, with a value range of [0.5, 2]. The maximum Lyapunov exponent (dimensionless) characterizes the divergence rate of the error trajectory; when If the value exceeds 0.2 for three consecutive time periods, the current time period type will be forcibly upgraded to the emergency period. The maximum Lyapunov exponent is also used to dynamically adjust the number of perturbation members in the ensemble forecast of claim 3. The adaptive adjustment rule is expressed by the following mathematical equation: ; in: The number of perturbation members at the next moment; =10 is the baseline number of disturbance members; The maximum Lyapunov exponent currently calculated; This is the rounding function. and Used to limit the number of members to between 5 and 20.

[0028] Furthermore, it also includes self-organizing demand response mechanisms based on evolutionary game theory during emergency periods: When the determined time period is an emergency period and the communication system load rate exceeds 80%, the self-organizing mode is triggered. Main station broadcast power deficit and emergency pricing coefficient; Each flexible load node (electric vehicle charging pile, temperature-controlled load, small-scale energy storage) acts as an intelligent game player, running an evolutionary game algorithm locally: each game player determines its interruptibility based on its own... The bidding curve is calculated based on quantity and user comfort tolerance coefficient; the game players exchange bidding information through power line carrier or nearby wireless communication, simulating the ant colony pheromone update rule, and converge to Nash equilibrium after multiple rounds of iteration; The main station only performs security checks on the final response plan. Once the check passes, it issues the execution command. After the emergency period ends, the system automatically switches back to the centralized robust optimization mode.

[0029] Furthermore, closed-loop correction also includes modifications to short-time scheduling instructions: If the absolute value of the calculated prediction error exceeds 5% of the rated capacity, a short-term correction scheduling will be triggered. The short-term correction scheduling adopts the model predictive control method, with a prediction time domain of 15 minutes and a control time domain of 5 minutes. The objective function is to minimize the correction cost and tracking deviation. The corrective instructions within 5 minutes are obtained and sent to the actuators; these corrective instructions do not change the main scheduling plan given in claim 5, but only make fine adjustments within a local time. It also includes a knowledge transfer mechanism for historical data: the actual supply and demand deviation and scheduling instructions under each meteorological event are stored in the case library; when a new meteorological feature is encountered, a similarity-based case retrieval is adopted to select the multiple cases with the highest similarity, and the average correction amount is taken as the bias term of the initial prediction deviation and superimposed on the prediction result of claim 3; the bias term gradually decays to zero after the first set of measured data is received.

[0030] Furthermore, all steps are executed on a rolling basis with a basic cycle of 15 minutes. The day-ahead forecast is executed once a day, the intraday forecast is executed once every 4 hours, the real-time forecast and rolling scheduling are executed once every 15 minutes, and the correction and adaptive mechanism runs continuously at a frequency of minutes. The entire method forms a complete closed loop from data acquisition, feature extraction, probability prediction, risk classification, robust optimization, instruction execution, error inversion to online updating of model parameters, and automatically adjusts the calculation frequency and complexity of each sub-module according to the characteristics of meteorological chaos.

[0031] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0032] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for coordinating energy supply and demand under meteorological influences, characterized in that, Includes the following steps: Step 1: Collect multidimensional meteorological data, historical energy supply and demand data, and real-time operational data; construct meteorological and energy characteristic tensors; Step 2: Based on ensemble forecasting and quantile regression models, generate multi-scale supply and demand probability prediction intervals and output the probability density function of net load; Step 3: Calculate the meteorological impact risk index based on the standard deviation of meteorological forecasts and the measured rate of change, and divide the scheduling period into stable period, fluctuating period and emergency period; Step 4: Using a time-varying risk preference robust optimization model, the confidence level of the opportunity constraint is dynamically adjusted according to the time period type to obtain the unit combination and output scheduling instructions; Step 5: After executing the scheduling command, collect real-time meteorological observation data, invert the micro-meteorological field through extended Kalman filter, and correct the prediction model parameters and short-term scheduling command online; feed the correction results back to the feature construction and prediction steps to form a closed loop.

2. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, The construction of meteorological and energy feature tensors specifically includes: The temperature, humidity, wind speed, cloud cover, air pressure, precipitation probability, shortwave radiation, historical load, and historical renewable energy output data in numerical weather forecasts are aligned to a 15-minute time grid, and outlier interpolation is performed.

3. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, The specific steps involved in generating the multi-scale supply and demand probability prediction interval include: A quantile regression forest model is used to establish a load probability predictor and a new energy output probability predictor in parallel. The input is the feature tensor described in claim 2, and the output is the load value and the new energy output value at the 10%, 50%, and 90% quantiles. Multiple disturbance members of numerical weather prediction are obtained, and each member is input into the two forecasters mentioned above to obtain multiple forecast intervals. The mean and standard deviation of net load at each time section are calculated. The probability density function of the net load is expressed by the following mathematical equation: ； in: For time period The net load random variable; The mean is variance is The normal distribution; For time period The net load mean is calculated as the arithmetic mean of the predictions from multiple set members; For time period The net load standard deviation characterizes the forecast uncertainty.

4. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, The calculation of the meteorological impact risk index and the division of time periods specifically include: Define time period Meteorological impact risk index Calculated by the following mathematical equation: ； in: For time period The meteorological impact risk index; For time period The standard deviation of net load; This represents the average standard deviation of net load at the same time over the past 7 days. It is the one with the largest absolute value of the rate of change among temperature, wind speed, and humidity; This is an extreme weather warning signal; , , For the preset weighting coefficients, satisfy + + =1; when The period is divided into a stable period, when The time is divided into fluctuation periods, when The time period is divided into emergency periods; the output time period type label sequence and the corresponding risk threshold are output.

5. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, The time-varying risk preference robust optimization model specifically includes: The objective function is established as minimizing the sum of total operating costs and expected load shedding and wind / solar curtailment penalties; The constraints include power balance constraints, unit ramping constraints, energy storage SOC constraints, line power flow constraints, and opportunity constraints. The allowable load shedding probability of the opportunity constraint is dynamically adjusted according to the time period type: the allowable load shedding probability is 0.1 during the stable period, 0.05 during the fluctuating period, and 0.01 during the emergency period; Dynamic reserve capacity configuration is represented by the following set of mathematical equations: ； in: , Time periods Upper and lower reserve capacities; , These are the basic and lower reserve capacities, respectively; , Reserve factor: 1.0 for stable periods, 1.5 for fluctuating periods, and 2.5 for emergency periods; The standard deviation of net load forecast; This is for backup response time; An affine adjustable robust optimization strategy is adopted to transform chance constraints into deterministic second-order cone constraints. The column AND constraint generation algorithm is used to solve the problem and output day-ahead, intraday, and real-time three-level scheduling instructions for each power generation unit, energy storage, and demand response.

6. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, The process of inverting the micro-meteorological field using extended Kalman filtering specifically includes: A meteorological and physical model for attenuation was established to express the attenuation of power line carrier communication signals as a linear combination of relative humidity, rainfall rate, and fog concentration. Using the deviation of the meteorological forecast system as the state variable and the attenuation and the measured deviation of the meteorological station as the observation vector, the state variable is updated in real time through the extended Kalman filter recursive formula to obtain the corrected micro-meteorological field. The corrected microclimate field is fed back to the feature construction step to update the feature tensor.

7. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, The parameters of the online correction prediction model specifically include: A lightweight online learning module was built, which only updates the parameters of the last two fully connected layers of the quantile regression forest model; The stochastic gradient descent method is used, and the loss function is quantile loss; Each time new 15 minutes of measured data is received, a parameter update is performed, and the learning rate decays exponentially with the number of iterations. The updated model parameters are immediately used for the next round of predictions.

8. The energy supply and demand allocation method under meteorological influence as described in claim 1, characterized in that, It also includes self-organizing demand response mechanisms based on evolutionary game theory during emergency periods: When the determined time period is an emergency period and the communication system load rate exceeds 80%, the self-organizing mode is triggered. Main station broadcast power deficit and emergency pricing coefficient; Each flexible load node acts as an intelligent game player, running an evolutionary game algorithm locally: each game player determines its own interruptibility... Calculate the quotation curve based on quantity and user comfort tolerance coefficient; The players exchange bid information via power line carrier or nearby wireless communication, simulating the pheromone update rule of ant colony, and converge to Nash equilibrium after multiple rounds of iteration. The main station only performs security checks on the final response plan. Once the check passes, it issues the execution command. After the emergency period ends, the system automatically switches back to the centralized robust optimization mode.

9. The energy supply and demand allocation method under meteorological influence according to claim 1, characterized in that, The closed-loop correction also includes the modification of short-time scheduling instructions: If the absolute value of the calculated prediction error exceeds 5% of the rated capacity, a short-term correction scheduling will be triggered. The short-term correction scheduling adopts the model predictive control method, with a prediction time domain of 15 minutes and a control time domain of 5 minutes. The objective function is to minimize the correction cost and tracking deviation. The corrective instructions obtained within 5 minutes are sent to the execution mechanism; these corrective instructions do not change the main scheduling plan given in claim 5, but only make fine adjustments within a local time.

10. The energy supply and demand allocation method under meteorological influence according to claim 1, characterized in that, All steps are executed on a rolling basis with a basic cycle of 15 minutes. The day-ahead forecast is executed once a day, the intraday forecast is executed once every 4 hours, the real-time forecast and rolling scheduling are executed once every 15 minutes, and the correction and adaptive mechanism runs continuously at a frequency of minutes.

Images