New energy station energy storage control system and method

Abstract

The present invention provides a new energy station energy storage control system and method. The system includes a high-precision prediction module for new energy output, which is used to integrate local laser radar and numerical weather forecast data to output an output forecast value; an SOC curve initialization module, which initializes the energy storage SOC reference curve based on the output forecast value and the day-ahead market electricity price information; a real-time node electricity price prediction module, which is used to predict real-time electricity price trends; and a rolling optimization control module for the energy storage system, which is connected to other modules and dynamically optimizes the charge and discharge instructions and SOC status based on the output forecast value, the energy storage SOC reference curve, and the electricity price prediction value of the real-time node electricity price prediction module. By integrating multi-source prediction results with real-time market information, the present invention constructs a highly adaptive dynamic optimization and control mechanism for energy storage, while meeting various operating and market constraints. This provides key support for the economic operation of new energy stations in a spot market environment.

Description

Technical Field

[0001] The present invention relates to the field of smart grid and new energy dispatching and control, and specifically to a new energy station energy storage control system and method, in particular to a new energy station energy storage control system and method that combines high-precision output forecasting with rolling optimization of intraday spot electricity prices. Background Art

[0002] With the rapid global development of renewable energy generation, the installed capacity of renewable energy sources such as wind power and photovoltaics continues to grow, making them an essential component of power systems worldwide. However, renewable energy sources exhibit significant volatility and uncertainty, posing significant challenges to the safety and economic efficiency of grid dispatch, particularly in high-penetration scenarios. Currently, renewable energy generation is fully participating in power market transactions, and the electricity spot market, in particular, places higher demands on renewable energy stations' power generation plan submissions, real-time tracking and execution, and deviation assessment. Because renewable energy output is difficult to accurately predict, large forecast deviations can easily lead to deviations in spot market assessments, directly impacting the profitability of renewable energy stations. Furthermore, the electricity spot market has introduced node-specific level-of-marginal prices (LMPs), creating a strong price signal-driven mechanism that further encourages renewable energy stations to improve their output controllability and market responsiveness. Furthermore, energy storage, as a crucial resource for optimizing renewable energy dispatch, determines the overall market competitiveness of renewable energy stations by determining whether its dispatch strategy can effectively respond to price signals.

[0003] However, traditional numerical weather prediction (NWP) models primarily rely on mesoscale numerical model simulations, which have low spatial resolution (10-30 kilometers) and slow update frequency (updated every 1-3 hours). This makes it difficult to capture the rapidly changing local meteorological characteristics of the areas where renewable energy sites are located. For example, meteorological phenomena such as localized cloud movement and sudden changes in wind speed often occur within minutes and have a significant impact on power output. NWP models are unable to promptly reflect these changes in their forecasts, resulting in large errors between ultra-short-term forecasts and actual power output. This forecast error not only affects the accuracy of renewable energy output plans but also causes economic losses.

[0004] On the other hand, while the energy storage systems currently used in new energy stations have the ability to participate in electricity market regulation, most still employ static charging and discharging strategies based on the day-ahead spot electricity price, ignoring the significant fluctuations and deviations between intraday (real-time) spot prices and day-ahead prices. This strategy makes it difficult for energy storage systems to flexibly respond to real-time node prices, failing to fully exploit their potential for "buy low, sell high" regulation, resulting in low returns. Especially in scenarios with frequent node price fluctuations, the execution of a fixed SOC (state of charge) trajectory is more likely to become disconnected from the system's marginal value.

[0005] Furthermore, new energy stations participating in the electricity spot market must be able to predict node electricity prices at various future time periods to accurately bid and dispatch energy storage resources. However, most existing stations lack the ability to predict real-time electricity prices, making it impossible to effectively support decision-making. Furthermore, traditional strategies ignore the bidding risks and economic consequences of output forecast errors and lack optimized dispatch mechanisms for uncertainties.

[0006] Therefore, there is an urgent need for a new control system for new energy stations that can integrate local high-precision meteorological perception and numerical weather forecast information to achieve high-frequency and accurate short-term power forecasting; at the same time, combined with the dual-time scale prediction capabilities of day-ahead and real-time electricity prices, it can dynamically formulate and optimize energy storage charging and discharging strategies, thereby improving the overall profitability and risk resistance of new energy stations. Summary of the Invention

[0007] In view of the defects in the prior art, the purpose of the present invention is to provide a new energy station energy storage control system and method.

[0008] According to the present invention, a new energy station energy storage control system is provided, comprising:

[0009] A high-precision prediction module for renewable energy output, which integrates local lidar and numerical weather forecast data to output predicted output values;

[0010] An SOC curve initialization module initializes an energy storage SOC reference curve based on the output forecast value and the day-ahead market electricity price information;

[0011] The real-time node electricity price prediction module is used to predict the real-time electricity price trend of the node where the new energy station is located during the operation day;

[0012] The rolling optimization control module of the energy storage system is connected to other modules to dynamically optimize the charge and discharge instructions and SOC status based on the output forecast value, the energy storage SOC reference curve and the electricity price forecast value of the real-time node electricity price prediction module.

[0013] Preferably, the new energy output high-precision prediction module periodically collects three-dimensional wind field and cloud optical characteristic data covering the prediction area through lidar equipment deployed at the new energy power generation station, and extracts key time series features through physical modeling and data processing methods.

[0014] Preferably, before the spot market operates, the SOC curve initialization module initializes the energy storage SOC reference curve based on the day-ahead market electricity price information, the predicted new energy output curve, and the declared power generation power curve; the energy storage SOC reference curve provides an initial operating boundary for the subsequent scheduling module, and serves as an auxiliary decision-making basis to guide the optimization of the declaration strategy.

[0015] Preferably, the real-time node electricity price prediction module constructs a real-time node electricity price prediction model based on the power grid operation data; by integrating the operation parameters published by the comprehensive power grid, combining the recently known bidding space and real-time operation data, the real-time electricity price trend of the node where the new energy station is located during the operation day is predicted.

[0016] Preferably, the rolling optimization control module of the energy storage system incorporates the following input information:

[0017] -New energy output high-precision prediction module for new energy high-precision prediction output sequence ;

[0018] -SOC curve initialization module initializes the SOC curve and the timing sequence of the initial operation strategy 、 ;

[0019] -Real-time node electricity price prediction curve of the real-time node electricity price prediction module ;

[0020] Based on the above, with the goal of maximizing benefits, the charge and discharge power sequence is re-solved for each future rolling control window. and the updated SOC state sequence .

[0021] According to the present invention, a new energy station energy storage control method is provided, comprising:

[0022] Step S1: Outputting the predicted value of renewable energy output by fusing the lidar data deployed at the renewable energy station with the numerical weather forecast data;

[0023] Step S2: Based on the predicted output value of new energy and the day-ahead spot electricity price, an optimization model with the goal of maximizing revenue is established to solve the initial charge and discharge power sequence and the corresponding SOC state curve;

[0024] Step S3: Outputting electricity price forecast values ​​based on the operating data and historical node electricity price information released by the power grid;

[0025] Step S4: Dynamically update the charge and discharge power instructions and SOC status according to the output forecast value, the SOC state curve and the electricity price forecast value.

[0026] Preferably, step S1 includes the following sub-steps:

[0027] Step S1.1: Use Doppler lidar to perform stereo scanning, using a combination of PPI and RHI modes to obtain lidar volume scanning data;

[0028] Inversion of horizontal and vertical wind speeds based on the VAD algorithm:

[0029]

[0030] in, is the horizontal wind speed, is the vertical wind speed, is the angle between the radar and the wind speed direction;

[0031] The optical depth is obtained by integrating the backscattering coefficient:

[0032]

[0033] in, is the optical thickness, which indicates the scattering power of the atmosphere; is the backscatter coefficient.

[0034] Construct the feature matrix:

[0035]

[0036] in, is the time window length; is the number of vertical layers;

[0037] Step S1.2: Temporal interpolation, spatial downscaling, and physical quantity normalization are performed on the numerical weather forecast data to fully align them with the lidar data in both temporal and spatial dimensions.

[0038] The original NWP data is updated at a 1-hour interval using cubic spline interpolation to update it to a 5-minute interval consistent with the radar data:

[0039]

[0040] in, is the interpolated meteorological variable, is the original time point, is the interpolation coefficient, which is obtained by fitting the hourly meteorological data values ​​throughout the day;

[0041] Improve spatial resolution by downscaling the spatial dimensions:

[0042] Use bilinear interpolation or high-order interpolation to map NWP data to a high-resolution grid within the radar field of view, ensuring that features at each spatial point are aligned:

[0043]

[0044] in, is the transformed local meteorological variable, is the interpolation weight, is the original grid point of NWP;

[0045] Construct the final fused feature tensor:

[0046] The processed NWP features are aligned with the five-dimensional optical and wind field features extracted by the radar in time and space to construct a unified input tensor:

[0047]

[0048] in, is the number of time steps, is the number of spatial sampling points, is the feature dimension after fusion;

[0049] Step S1.3: Construct a fusion prediction model based on the LSTM long short-term memory network to implement accurate wind power prediction; the fusion prediction model is designed with a multi-channel hybrid structure, inputting the lidar data and the processed numerical weather forecast data into two parallel LSTM network channels respectively to capture the temporal characteristics and evolution trends of different source data.

[0050] Preferably, step S2 includes the following sub-steps:

[0051] Step S2.1: Set and filter input data and constraints; input data includes the day-ahead market electricity price series , predicted new energy output curve , the electricity declaration curve submitted in the day-ahead market and energy storage system operation constraint parameter set;

[0052] The output prediction value and the energy storage adjustment value are added together and do not exceed the limit:

[0053]

[0054] in, is the net output power of the energy storage system at time t, with positive values ​​indicating discharge and negative values ​​indicating charge; is the excess profit recovery deviation coefficient;

[0055] Step S2.2: Define the SOC recursive expression:

[0056]

[0057] in:

[0058] is the state of charge of the energy storage system at time t;

[0059] , are the charging and discharging powers at time t, respectively;

[0060] , are the charging and discharging efficiencies of the energy storage system, respectively;

[0061] is the rated capacity of the energy storage system;

[0062] is the time interval;

[0063] SOC updates meet the following operational boundary conditions:

[0064]

[0065] ,

[0066] The above power variables also satisfy the mutually exclusive relationship:

[0067] ;

[0068] Step S2.3: Establish the profit maximization objective function:

[0069] ;

[0070] Step S2.4: Construct a timing optimization problem with equality and inequality constraints to solve the SOC curve and the corresponding charge and discharge power sequence.

[0071] Preferably, step S3 includes the following sub-steps:

[0072] Step S3.1: By building a data collection interface, regularly obtain various power grid operation data from the power trading center;

[0073] Step S3.2: Establishing an intraday node electricity price prediction model to perform electricity price prediction;

[0074] Step S3.3: Establishing a node electricity price correction model to correct the electricity price prediction result;

[0075] Step S3.4: combining the intraday node electricity price prediction model and the node electricity price correction model to obtain a final prediction model for the real-time node electricity price.

[0076] Preferably, step S4 includes the following sub-steps:

[0077] Step S4.1: Define the profit maximization objective function:

[0078]

[0079] in, is the charge and discharge power of the energy storage system at time t, charging is negative and discharging is positive; is the predicted node electricity price at time t; is the time interval; Optimize window length for scrolling; The starting time for current optimization;

[0080] Step S4.2: Use the rolling time domain optimization method, each time with the current time As the starting point, scroll forward for a period of time , solve the optimal charging and discharging strategy for this interval;

[0081] Step S4.3: Set key constraints, including:

[0082] Energy storage system power constraints:

[0083]

[0084] SOC status update constraints:

[0085]

[0086] SOC boundary limits:

[0087]

[0088] Excess profit recovery scope constraints: Real-time charging and discharging strategy combined with new energy output forecast Verify together:

[0089] ;

[0090] Step S4.4: After the rolling optimization is completed, the optimal charge and discharge instructions for the current control step are output , and update the SOC state curve of the energy storage system in real time ; The updated SOC and strategy sequence are used as the initial condition input for the next rolling cycle to form a complete closed loop.

[0091] Compared with the prior art, the present invention has the following beneficial effects:

[0092] 1. By integrating multi-source forecast results with real-time market information, this invention constructs a highly adaptive dynamic optimization and control mechanism for energy storage while satisfying various operational and market constraints, providing key support for the economic operation of new energy stations in a spot market environment.

[0093] 2. The present invention proposes a new energy station energy storage control system and method that combines high-precision output forecasting with rolling optimization of electricity prices. It integrates local lidar and numerical weather forecast data to achieve high-precision short-term output forecasting of 15 minutes to 1 hour.

[0094] 3. The present invention can formulate a preliminary energy storage charge and discharge curve based on the day-ahead electricity price, and then make a rolling forecast of the intraday electricity price based on real-time market data, further dynamically optimizing the charge and discharge behavior and SOC path of the energy storage system; the overall architecture realizes a closed-loop linkage of "perception-prediction-optimization-execution", ensuring that new energy stations have stronger profitability and operational stability in a complex spot market environment.

[0095] 4. The control system of the present invention is mainly composed of four core functional modules, namely: a high-precision prediction module for new energy output, an SOC curve initialization module, a real-time node electricity price prediction module and a rolling optimization scheduling module for the energy storage system; among them, the high-precision prediction module for new energy output is the data perception and modeling basis of the system, which improves the output prediction accuracy through the fusion of multi-source meteorological information and AI algorithms, and provides data support for subsequent market participation; the SOC curve initialization module formulates a preliminary charging and discharging strategy based on the prediction results and the day-ahead electricity price, providing a reasonable starting point for energy storage scheduling; the bidding space fitting module extracts the feasible strategy boundary from historical behavior and formulates a quotation path that maximizes the profit for the site; the real-time node electricity price prediction module realizes the fitting of the bidding space through the day-ahead spot electricity price and the day-ahead forecast data released by the power grid, and realizes the accurate prediction of the node electricity price at future time according to the real-time operation data and change trend of the power grid during real-time operation; finally, the rolling optimization control module integrates all the above information, takes profit optimization as the goal, and outputs the optimal charging and discharging operation instructions for the energy storage system. Through the synergistic effect of the four major modules, this system can achieve accurate forecasting, flexible response and economic operation in a spot market environment where renewable energy output is highly uncertain and electricity prices fluctuate rapidly, significantly improving the regulation capacity and market revenue level of new energy stations.

[0096] Other beneficial effects of the present invention will be explained through the introduction of specific technical features and technical solutions in the specific implementation methods. Those skilled in the art should be able to understand the beneficial technical effects brought about by the introduction of these technical features and technical solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0098] Figure 1 This is a system block diagram of the present invention.

[0099] Figure 2 This is an operation flow chart of the high-precision prediction module for new energy output in the present invention.

[0100] Figure 3 This is an operation flow chart of the SOC curve initialization module in the present invention.

[0101] Figure 4 This is an operation flow chart of the real-time node electricity price prediction module in the present invention.

[0102] Figure 5 This is an operation flow chart of the rolling optimization control module of the energy storage system in the present invention.

[0103] Figure 6 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0104] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0105] Reference Figure 1 As shown, a new energy station energy storage control system includes:

[0106] New energy output high-precision prediction module, SOC curve initialization module, real-time node electricity price prediction module and energy storage system rolling optimization scheduling module.

[0107] The high-precision renewable energy output prediction module forms the foundation of the system's data perception and modeling. It improves output prediction accuracy through the integration of multi-source meteorological information and AI algorithms, providing data support for subsequent market participation. The SOC curve initialization module formulates preliminary charging and discharging strategies based on the forecast results and the day-ahead electricity price, providing a reasonable starting point for energy storage scheduling. The bidding space fitting module extracts feasible strategy boundaries from historical behavior and formulates a profit-maximizing bidding path for the station. The real-time node electricity price prediction module fits the bidding space by combining the day-ahead spot electricity price with the day-ahead forecast data released by the power grid. During real-time operation, it accurately predicts future node electricity prices based on the grid's real-time operating data and changing trends. Finally, the rolling optimization control module integrates all of this information and, with the goal of optimizing revenue, outputs the optimal charging and discharging operation instructions for the energy storage system. Through the synergistic effect of these four modules, the system is able to achieve accurate forecasting, flexible response, and economical operation in a spot market environment characterized by high uncertainty in renewable energy output and rapidly fluctuating electricity prices, significantly improving the control capabilities and market revenue levels of renewable energy stations.

[0108] The specific modules are introduced as follows:

[0109] 1. New Energy Output High-Precision Prediction Module

[0110] Reference Figure 2As shown in Figure 1, this module aims to achieve high-precision forecasts of renewable energy output, such as wind power and photovoltaic power, on a minute-by-minute timescale. It proposes a high-precision forecasting method that integrates lidar observation data with numerical weather prediction (NWP) data. This module builds a complete spatiotemporal data processing and deep learning modeling process, covering the entire process from data acquisition, feature processing, model training to forecast output. Its operation steps are as follows:

[0111] Step 1: LiDAR 3D field data acquisition and feature extraction

[0112] This method uses lidar equipment deployed at renewable energy power generation sites to periodically collect three-dimensional wind field and cloud optical property data covering the forecast area. Physical modeling and data processing methods are used to extract key time series features, providing input features with high temporal and spatial resolution and clear physical meaning for subsequent forecast models. The key points of the technical implementation are as follows:

[0113] Data collection method:

[0114] Doppler LiDAR is used to perform stereo scanning, using a combination of PPI (Plan Position Indicator) and RHI (Range Height Indicator) modes to acquire LiDAR volume scan data. Raw echo parameters include but are not limited to:

[0115] Radial wind speed , represents the wind speed along the radar measurement direction;

[0116] Backscatter coefficient , which represents the reflection intensity of radar waves by meteorological particles;

[0117] Signal-to-noise ratio , indicating the clarity of the radar signal.

[0118] Three-dimensional wind field inversion:

[0119] Inversion of horizontal and vertical wind speeds based on the VAD (Velocity Azimuth Display) algorithm:

[0120]

[0121] in, is the horizontal wind speed, is the vertical wind speed, is the angle between the radar and the wind speed direction.

[0122] Optical thickness calculation:

[0123] The optical depth is obtained by integrating the backscattering coefficient:

[0124]

[0125] in, is the optical thickness, which indicates the scattering power of the atmosphere; is the backscatter coefficient.

[0126] Feature matrix construction (time series form):

[0127]

[0128] in, is the time window length; is the number of vertical layers. The matrix contains information on optical depth, wind speed and its spatial gradient.

[0129] Step 2: Spatiotemporal alignment and dynamic fusion of multi-source data

[0130] Numerical Weather Prediction (NWP) data typically has low temporal and spatial resolution, making it difficult to directly meet the needs of short-term nowcasting with minute-level accuracy. To achieve full alignment with LiDAR data in both temporal and spatial dimensions, NWP data requires processing operations such as temporal interpolation, spatial downscaling, and physical quantity standardization. These methods can transform coarse-scale meteorological data into high-quality input features with a unified structure and matching dimensions, providing multi-source fused feature data for subsequent deep learning models. Key technical implementation points are as follows:

[0131] Time dimension alignment (improving time resolution):

[0132] The original NWP data is updated at a 1-hour interval using cubic spline interpolation to update it to a 5-minute interval consistent with the radar data:

[0133]

[0134] in, are interpolated meteorological variables (such as temperature, wind speed, radiation, etc.), is the original time point, is the interpolation coefficient, which is obtained by fitting the hourly meteorological data values ​​throughout the day.

[0135] Downscaling of spatial dimensions (increasing spatial resolution):

[0136] The original NWP grid is generally 3km in scale, while the local area covered by the radar scan data is usually refined to the 100m level. Bilinear interpolation or high-order interpolation is used to map the NWP data to the high-resolution grid within the radar field of view to ensure that the features of each spatial point are aligned:

[0137]

[0138] in, is the transformed local meteorological variable, is the interpolation weight, is the original NWP grid point.

[0139] The final fused feature tensor is constructed:

[0140] The processed NWP features (such as temperature, wind speed, air pressure, radiation, etc.) are aligned with the five-dimensional optical and wind field features extracted by radar in time and space points to construct a unified input tensor:

[0141]

[0142] in, is the number of time steps (e.g., a time window with a 5-minute interval), is the number of spatial sampling points (matching the radar scanning grid), is the fused feature dimension (including radar and NWP variables).

[0143] Through the above processing, the gap between radar data and meteorological forecast data in terms of temporal and spatial scales is bridged, and the low-frequency, low-resolution forecast data is upgraded to a format consistent with radar observations, thereby providing comprehensive and synchronized multi-source meteorological input for the deep learning model.

[0144] Step 3: Hybrid Machine Learning Model Construction and Training (LSTM)

[0145] After aligning the temporal and spatial dimensions of multi-source meteorological data and formatting its features, the model training phase begins. This step builds a fusion prediction model based on an LSTM (Long Short-Term Memory) network, leveraging high-frequency local observation data from lidar and mesoscale trend information from numerical weather forecasts to achieve accurate minute-by-minute wind power forecasts. Specifically, the model design utilizes a multi-channel hybrid structure: lidar data and processed NWP data are fed into two parallel LSTM network channels, respectively, to capture the temporal characteristics and evolutionary trends of the different source data.

[0146] The overall model adopts a parallel structure, mainly including two time series input channels and one feature fusion output channel:

[0147] Radar data channel:

[0148] The input is a sequence of wind speed and turbulence characteristics at different heights, such as:

[0149]

[0150] in, represents the horizontal wind speed at time t and height z, represents the vertical wind speed at time t and height z, represents the optical thickness at time t and height z.

[0151] The LSTM network is used to extract its short-term dynamic features, and the output is:

[0152]

[0153] in, The temporal hidden feature model representation extracted by LSTM for the lidar channel is It is the feature matrix of the lidar input.

[0154] NWP data channel:

[0155] The input is the NWP meteorological element sequence after interpolation and downscaling:

[0156]

[0157] in, Respectively represent NWP characteristics such as wind speed, radiation intensity, temperature, humidity, air pressure, and cloud cover

[0158] Similarly, input the LSTM network to extract its temporal evolution features:

[0159]

[0160] in, The temporal hidden feature model representation extracted by LSTM for the meteorological data channel is: The characteristic matrix of meteorological data input.

[0161] Feature fusion and prediction output:

[0162] By fusing the output features of the two channels, we can build an ultra-short-term, high-precision new energy output prediction model through concatenation or attention mechanism:

[0163]

[0164] or:

[0165]

[0166] in, is the representation of the fused feature model.

[0167] The wind power forecast value at a certain moment in the future is output through full-connected network regression:

[0168]

[0169] in, The predicted future wind power value, the prediction step is .

[0170] The model uses the conventional mean square error (MSE) as the loss function to measure the difference between the predicted value and the actual value; the model parameters are continuously updated through backpropagation and gradient optimization algorithms (such as Adam) to improve the prediction accuracy.

[0171] By building an LSTM hybrid model that integrates lidar and numerical weather forecast information, the ability of short-term wind power forecasts to respond to sudden meteorological disturbances can be significantly enhanced, providing more accurate data support for improving the efficiency of new energy grid connection and consumption and power market trading decisions.

[0172] 2. SOC curve initialization module

[0173] Reference Figure 3 As shown, when new energy power plants participate in spot electricity market transactions, the energy storage system's charging and discharging behavior and its state of charge (SOC) control strategy have a decisive impact on economic benefits. The current spot market implements a two-tiered electricity pricing mechanism: the day-ahead market price and the real-time (intraday) price coexist. The day-ahead price is published one day in advance, reflecting the overall power supply and demand balance on the day of operation and providing a strong price guidance function. Furthermore, extensive historical data demonstrates that real-time electricity price trends are typically highly correlated with the day-ahead price, exhibiting a certain degree of synchronization. Therefore, the energy storage system can determine the next day's charging and discharging plan and corresponding SOC curve based on the day-ahead price before operation.

[0174] Before the spot market operates, this module aims to initialize a reference SOC curve for energy storage that complies with market rules, has reasonable operating logic, and maximizes arbitrage potential based on day-ahead market electricity price information, predicted renewable energy output curves, and declared power generation curves. This SOC curve not only provides initial operating boundaries for subsequent scheduling modules but also serves as a decision-making aid to guide the optimization of declaration strategies. The module's workflow consists of the following four steps:

[0175] Step 1: Prepare input data and constraints:

[0176] The main inputs of the SOC initialization module include:

[0177] : Day-ahead market electricity price series (15 minutes as a cycle);

[0178] : Predicted new energy output curve;

[0179] : The electricity declaration curve submitted in the day-ahead market;

[0180] Energy storage system operation constraint parameter set (charge and discharge power upper limit, SOC upper and lower bounds, energy efficiency, initial state, etc.).

[0181] In the spot market, to avoid "excessive profits," trading institutions typically set profit recovery rules: the positive deviation between the reported power curve and the actual output + energy storage discharge must not exceed a certain percentage, otherwise the corresponding profit of the excess will be recovered; however, if the deviation is negative (under-forecast), no compensation will be given. Therefore, during the SOC initialization process, it should be ensured that the predicted output and the energy storage adjustment amount do not exceed the limit after being added together:

[0182]

[0183] in, is the net output power of the energy storage system at time t, with positive values ​​indicating discharge and negative values ​​indicating charge; It is the excess profit recovery deviation coefficient, usually between 10% and 30%, and is set according to market rules.

[0184] Step 2: SOC curve recursive relationship modeling

[0185] When planning the SOC curve, the energy conservation principle must be followed and the following SOC recursive expression is defined:

[0186]

[0187] in:

[0188] is the state of charge of the energy storage system at time t;

[0189] , are the charging and discharging powers at time t, respectively;

[0190] , are the charging and discharging efficiencies of the energy storage system (0~1);

[0191] is the rated capacity of the energy storage system;

[0192] The time interval is 15 minutes in the spot market.

[0193] SOC updates must also meet the following operational boundary conditions:

[0194]

[0195] ,

[0196] The above power variables must also satisfy a mutually exclusive relationship (only charging or discharging can occur at the same time):

[0197]

[0198] Step 3: Establish the objective function of maximizing revenue

[0199] The goal of SOC curve derivation is to maximize the net benefits of charging and discharging energy storage in different time periods while ensuring compliance with market constraints and the physical operation constraints of the energy storage system. The objective function can be expressed as:

[0200]

[0201] This target value represents the total revenue that the energy storage system can obtain by charging during low electricity price periods and discharging during high electricity price periods on spot operation days.

[0202] Step 4: Initialize the solution of the SOC curve

[0203] To solve the aforementioned SOC curve and the corresponding charge-discharge power sequence, a time-series optimization problem with equality and inequality constraints must be constructed. This problem is a typical mixed-integer nonlinear programming (MINLP) or linear programming (LP) problem. In this embodiment of the present invention, linearization and relaxation strategies are preferred for modeling to reduce computational complexity.

[0204] The solution process includes the following steps:

[0205] Time domain discretization: Divide the operating day into several fixed-interval time periods (such as 15 minutes), set the total number of time periods to T (corresponding to 96 points), and convert all variables into time series vector form.

[0206] State variable expansion: Expand the decision variables of each period into a vector form, including:

[0207]

[0208]

[0209]

[0210] Optimization modeling: A mixed integer programming (MILP) model is constructed based on the profit function maximization objective, SOC recursive formula, charge and discharge power mutual exclusion constraints, SOC boundary conditions, reported deviation constraints, and whether to charge / discharge.

[0211] Optimization solution: Call the solver (such as CPLEX, SCIP) to solve and output the optimal initial SOC curve and preliminary operation strategy 、

[0212] Feasibility verification and smoothing: For jumps, spikes or discontinuous segments in the solution results, necessary post-processing (such as moving average and interpolation correction) is performed to ensure actual feasibility.

[0213] Ultimately, the derived SOC curve not only meets the market transaction boundary requirements, but also maximizes economic benefits on the basis of satisfying the physical constraints of operation, ensuring the efficient operation of the energy storage system in the spot market environment.

[0214] 3. Real-time node electricity price prediction module

[0215] Reference Figure 4 As shown in the figure, this module aims to build a refined real-time node electricity price forecast model based on grid operation data, providing accurate price signals and references for new energy stations to participate in spot market transactions and optimize the scheduling of energy storage systems. Considering that node electricity prices in the spot market are affected by multiple factors such as the supply and demand status of the node's internal grid, the grid congestion status, and the bidding behavior of thermal power units, this module predicts the real-time electricity price trend of the node where the new energy station is located during the operation day by integrating the operating parameters published by the grid, combining the bidding space known in the past and real-time operation data. The module's workflow mainly consists of the following four steps:

[0216] Step 1: Market operation data collection and bidding space calculation

[0217] In the electricity spot market mechanism, grid operators need to publish a large amount of operation-related data every day. This module builds a data collection interface to regularly obtain various types of grid operation data from the power trading center, including:

[0218] Unified load curve ;

[0219] Non-market unit output ;

[0220] New energy unit output ;

[0221] Output of hydropower and other types of units ;

[0222] Outbound channel load ;

[0223] Channel flow and constraint information;

[0224] Day-ahead spot electricity prices and real-time spot electricity prices;

[0225] Day-ahead node electricity price and real-time node electricity price;

[0226] Node congestion price;

[0227] Structural information such as power generation and transmission maintenance plans;

[0228] The data acquisition module supports interface docking with the data publishing platform of the power trading center, automatically pulling and updating data according to a preset cycle (such as every 5 minutes or 15 minutes) to ensure the timeliness and integrity of the information.

[0229] Based on the above operating data and combined with the market mechanism, the bidding space on the operating day can be deduced. The specific calculation is as follows:

[0230]

[0231] Bidding space This represents the available capacity range of thermal power units eligible for market dispatch, effectively defining the dominant bid range for electricity prices. By feeding this bidding space with real-time grid operation data, with bids already determined before the start date, it is possible to derive price trends at the time of operation, providing high-quality structured input for downstream models.

[0232] Step 2: Intraday node electricity price prediction model

[0233] Real-time node electricity price The essence of is to solve the marginal price of a given bidding space based on the current system supply and demand status, network flow, congestion constraints and other factors. Considering that the node price is determined by the system marginal price and node congestion price Superposition, there are:

[0234]

[0235] in, is the real-time spot electricity price of node n; Depends on the real-time net load demand and thermal power quotation space; It is closely related to the grid topology, channel load level and node location.

[0236] Based on the above principles, this module uses the fixed quotation space of the day before and real-time input volume on the operating day Establish a node electricity price prediction model:

[0237]

[0238] in:

[0239] Real-time operation indicators of the power grid;

[0240] is the historical marginal electricity price data of node n;

[0241] It is a fitting function established by deep learning or machine learning model.

[0242] Step 3: Congestion risk assessment and node electricity price correction model

[0243] Since node electricity prices are significantly affected by channel constraints, this module will As an important input variable, it is used to identify potential congestion risks. By analyzing the correlation between the channel load rate and the historical node congestion price, a congestion sensitivity index is constructed. , and correct the system electricity price forecast results:

[0244]

[0245] in, A blocking model function built for deep learning or machine learning models.

[0246] Step 4: Model training and real-time node electricity price prediction mechanism

[0247] Based on the above results, the final prediction model of real-time node electricity price is:

[0248]

[0249] Among them, the input features are:

[0250] Unified bidding parameters for the day ahead: including the network-wide unified load forecast, non-market output forecast, transmission plan, and predicted renewable energy output, forming an estimate of the bidding space for the operation day;

[0251] Actual load and output data for each hour of the operating day: including real-time centralized load, output of various power sources, channel load, and operating status of important units;

[0252] Node historical price data: including the day-ahead electricity price, real-time node electricity price, congestion price, etc. of each node, capturing the historical evolution of the spatial price structure.

[0253] Target variable:

[0254] The node real-time electricity price prediction target is , where n represents the node number and t represents the time point (15-minute granularity).

[0255] Model structure:

[0256] The model adopts a dual-flow modeling structure to take into account both the system marginal electricity price fitting and the characterization of inter-node congestion risk:

[0257] Main Model : Used to fit the evolution of marginal electricity prices at the system level. It uses indicators such as the grid-wide or regional level coordinated load and the size of the bidding space as primary inputs to predict the basic trend of system prices.

[0258] Revised model : Used to predict node congestion premiums (or discounts) caused by factors such as grid topology and channel load changes. The congestion offset term is fitted by inputting the real-time load of each channel, historical node congestion price series, and the channel-node correlation matrix.

[0259] The model structure can be implemented using a variety of technical approaches, including but not limited to traditional machine learning models such as gradient boosted tree (GBDT) and random forest (RF), or time series modeling methods based on deep learning structures such as LSTM and Transformer. In actual deployment, flexible model selection and parameter adjustment can be made based on data integrity, operational efficiency, and prediction accuracy requirements.

[0260] The final output node electricity price prediction curve It will serve as an important input feature for the subsequent revenue evaluation and energy storage optimization scheduling modules of new energy power stations, providing basic support for accurate participation in the real-time spot market.

[0261] 4. Rolling Optimization Control Module of Energy Storage System

[0262] Reference Figure 5 As shown in Figure 1, based on the previous calculation results, this module further constructs an energy storage system optimization and control mechanism for the rolling period of the operating day, realizing the dynamic update of the charging and discharging strategy and improving the benefits. Specifically, this module integrates the following input information:

[0263] New energy high-precision prediction output sequence from the new energy output high-precision prediction module ;

[0264] The timing sequence of the initialization SOC curve and the preliminary operation strategy from the SOC curve initialization module 、 ;

[0265] Node real-time electricity price prediction curve from the real-time node electricity price prediction module .

[0266] Based on the above, this module aims to maximize the benefits and re-solves the charge and discharge power sequence for each future rolling control window. and the updated SOC state sequence .

[0267] Step 1: Objective function definition (maximizing revenue)

[0268] The core goal of rolling optimization control is to maximize the benefits of the energy storage system. The objective function can be qualitatively described as follows:

[0269]

[0270] in:

[0271] : The charge and discharge power of the energy storage system at time t, charging is negative and discharging is positive;

[0272] : predicted node electricity price at time t;

[0273] : Time interval (usually 15 minutes);

[0274] : Rolling optimization window length;

[0275] : The current optimization start time.

[0276] Step 2: Create a rolling window mechanism

[0277] This module uses rolling time domain optimization, each time with the current time As the starting point, scroll forward for a period of time , solving the optimal charging and discharging strategy for that interval. After each optimization is completed, the previous short period (such as 1-2 time steps) is used as the execution control instruction. The optimization is repeated at the next moment based on the latest forecast data to ensure the dynamic adaptability of the strategy.

[0278] Step 3: Key Constraints

[0279] The following operating constraints must be met during rolling optimization:

[0280] Energy storage system power constraints:

[0281]

[0282] SOC status update constraints:

[0283]

[0284] SOC boundary limits:

[0285]

[0286] Excess profit recovery scope constraints: Real-time charging and discharging strategies need to be combined with new energy output forecasts Verify together to ensure that the excess profit recovery range does not exceed the output limit declared on the previous day, that is:

[0287]

[0288] Step 4: Control strategy output and SOC update

[0289] After the rolling optimization is completed, the optimal charge and discharge instructions for the current control step are output , and update the SOC state curve of the energy storage system in real time The updated SOC and strategy sequence will serve as the initial condition input for the next rolling cycle, forming a complete closed loop.

[0290] By integrating multi-source forecast results with real-time market information, this invention builds a highly adaptive dynamic optimization and control mechanism for energy storage, while satisfying various operational and market constraints, providing key support for the economic operation of new energy stations in a spot market environment.

[0291] Reference Figure 6 As shown, a new energy station energy storage control method includes:

[0292] Step S1: Outputting the predicted value of renewable energy output by fusing the lidar data deployed at the renewable energy station with the numerical weather forecast data;

[0293] Step S2: Based on the predicted output value of new energy and the day-ahead spot electricity price, an optimization model with the goal of maximizing revenue is established to solve the initial charge and discharge power sequence and the corresponding SOC state curve;

[0294] Step S3: Outputting electricity price forecast values ​​based on the operating data and historical node electricity price information released by the power grid;

[0295] Step S4: Dynamically update the charge and discharge power instructions and SOC status according to the output forecast value, the SOC state curve and the electricity price forecast value.

[0296] Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, and units provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same functions of the system and its various devices, modules, and units provided by the present invention in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; the devices, modules, and units for implementing various functions can also be considered as both software modules implementing the method and structures within the hardware component.

[0297] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A new energy station energy storage control method, characterized in that: The control steps are completed based on the following modules: A high-precision prediction module for renewable energy output, which integrates local lidar and numerical weather forecast data to output predicted output values; An SOC curve initialization module initializes an energy storage SOC reference curve based on the output forecast value and the day-ahead market electricity price information; The real-time node electricity price prediction module is used to predict the real-time electricity price trend of the node where the new energy station is located during the operation day; The rolling optimization control module of the energy storage system is connected to other modules to dynamically optimize the charge and discharge instructions and SOC status based on the output forecast value, the energy storage SOC reference curve, and the electricity price forecast value of the real-time node electricity price forecast module; The high-precision prediction module for renewable energy output uses lidar equipment deployed at renewable energy power generation sites to periodically collect three-dimensional wind field and cloud optical characteristic data covering the prediction area, and extracts key time series features through physical modeling and data processing methods. Before the spot market operates, the SOC curve initialization module initializes the energy storage SOC reference curve based on the day-ahead market electricity price information, the predicted renewable energy output curve, and the declared power generation curve. The energy storage SOC reference curve provides the initial operating boundary for the subsequent scheduling module and serves as an auxiliary decision-making basis to guide the optimization of the declared strategy. The real-time node electricity price prediction module builds a real-time node electricity price prediction model based on grid operation data. By integrating the operating parameters published by the grid, combining the known bidding space and real-time operation data, it predicts the real-time electricity price trend of the node where the new energy station is located during the operation day. The rolling optimization control module of the energy storage system incorporates the following input information: -New energy output high-precision prediction module for new energy high-precision prediction output sequence ; -SOC curve initialization module initializes the SOC curve and the timing sequence of the initial operation strategy 、 ; -Real-time node electricity price prediction curve of the real-time node electricity price prediction module ; Based on the above, with the goal of maximizing benefits, the charge and discharge power sequence is re-solved for each future rolling control window. and the updated SOC state sequence ; The control steps include: Step S1: Outputting the predicted value of renewable energy output by fusing the lidar data deployed at the renewable energy station with the numerical weather forecast data; The step S1 includes the following sub-steps: Step S1.1: Use Doppler lidar to perform stereo scanning, using a combination of PPI and RHI modes to obtain lidar volume scanning data; Inversion of horizontal and vertical wind speeds based on the VAD algorithm: in, is the horizontal wind speed, is the vertical wind speed, is the angle between the radar and the wind speed direction; The optical depth is obtained by integrating the backscattering coefficient: in, is the optical thickness, which indicates the scattering power of the atmosphere; is the backscattering coefficient; Construct the feature matrix: in, is the time window length; is the number of vertical layers; Step S1.2: Temporal interpolation, spatial downscaling, and physical quantity normalization are performed on the numerical weather forecast data to fully align them with the lidar data in both temporal and spatial dimensions. The original NWP data is updated at a 1-hour interval using cubic spline interpolation to update it to a 5-minute interval consistent with the radar data: in, is the interpolated meteorological variable, is the original time point, is the interpolation coefficient, which is obtained by fitting the hourly meteorological data values ​​throughout the day; Improve spatial resolution by downscaling the spatial dimensions: Use bilinear interpolation or high-order interpolation to map NWP data to a high-resolution grid within the radar field of view, ensuring that features at each spatial point are aligned: in, is the transformed local meteorological variable, is the interpolation weight, is the original grid point of NWP; Construct the final fused feature tensor: The processed NWP features are aligned with the five-dimensional optical and wind field features extracted by the radar in time and space to construct a unified input tensor: in, is the number of time steps, is the number of spatial sampling points, is the feature dimension after fusion; Step S1.3: Construct a fusion prediction model based on the LSTM long short-term memory network to implement accurate wind power forecasting. The fusion prediction model is designed using a multi-channel hybrid structure, inputting the lidar data and processed numerical weather forecast data into two parallel LSTM network channels to capture the temporal characteristics and evolution trends of the different source data. Step S2: Based on the predicted output value of new energy and the day-ahead spot electricity price, an optimization model with the goal of maximizing revenue is established to solve the initial charge and discharge power sequence and the corresponding SOC state curve; Step S3: Outputting electricity price forecast values ​​based on the operating data and historical node electricity price information released by the power grid; Step S4: Dynamically update the charge and discharge power instructions and SOC status according to the output forecast value, the SOC state curve and the electricity price forecast value.

2. The energy storage control method of a new energy station according to claim 1, characterized in that: The step S2 includes the following sub-steps: Step S2.1: Set and filter input data and constraints; input data includes the day-ahead market electricity price series , predicted new energy output curve , the electricity declaration curve submitted in the day-ahead market and energy storage system operation constraint parameter set; The output prediction value and the energy storage adjustment value are added together and do not exceed the limit: in, is the net output power of the energy storage system at time t, with positive values ​​indicating discharge and negative values ​​indicating charge; is the excess profit recovery deviation coefficient; Step S2.2: Define the SOC recursive expression: in: is the state of charge of the energy storage system at time t; , are the charging and discharging powers at time t, respectively; , are the charging and discharging efficiencies of the energy storage system, respectively; is the rated capacity of the energy storage system; is the time interval; SOC updates meet the following operational boundary conditions: ， The above power variables also satisfy the mutually exclusive relationship: ； Step S2.3: Establish the profit maximization objective function: ； Step S2.4: Construct a timing optimization problem with equality and inequality constraints to solve the SOC curve and the corresponding charge and discharge power sequence.

3. The energy storage control method of a new energy station according to claim 2, characterized in that: The step S3 includes the following sub-steps: Step S3.1: By building a data collection interface, regularly obtain various power grid operation data from the power trading center; Step S3.2: Establishing an intraday node electricity price prediction model to perform electricity price prediction; Step S3.3: Establishing a node electricity price correction model to correct the electricity price prediction result; Step S3.4: combining the intraday node electricity price prediction model and the node electricity price correction model to obtain a final prediction model for the real-time node electricity price.

4. The energy storage control method of a new energy station according to claim 3, characterized in that: The step S4 includes the following sub-steps: Step S4.1: Define the profit maximization objective function: in, is the charge and discharge power of the energy storage system at time t, charging is negative and discharging is positive; is the predicted node electricity price at time t; is the time interval; Optimize window length for scrolling; The starting time for current optimization; Step S4.2: Use the rolling time domain optimization method, each time with the current time As the starting point, scroll forward for a period of time , solve the optimal charging and discharging strategy for this interval; Step S4.3: Set key constraints, including: Energy storage system power constraints: SOC status update constraints: SOC boundary limits: Excess profit recovery scope constraints: Real-time charging and discharging strategy combined with new energy output forecast Verify together: ； Step S4.4: After the rolling optimization is completed, the optimal charge and discharge instructions for the current control step are output , and update the SOC state curve of the energy storage system in real time ; The updated SOC and strategy sequence are used as the initial condition input for the next rolling cycle to form a complete closed loop.

Images