Source network load storage regulation and control device based on model prediction

Abstract

The invention provides a source network load storage regulation and control device based on model prediction, and the device comprises a data obtaining device which is used for obtaining and preprocessing real-time source network load storage data based on a global data collection system covering a source side power generation unit, a network side power transmission network, a load side power utilization terminal and a storage side energy storage device, and obtaining a prediction data set; the model construction equipment is used for constructing a new energy output prediction model and a load prediction model based on a preset prediction time scale; the model correction equipment is used for acquiring output parameters of the new energy output prediction model and the load prediction model based on the prediction data set, comparing the output parameters with actual operation data and correcting errors of the new energy output prediction model and the load prediction model; and the regulation and control equipment is used for solving the source network load storage multi-objective optimization function by adopting a particle swarm optimization algorithm based on the corrected new energy output prediction model and the load prediction model so as to obtain an optimized source network load storage distribution scheme.

Description

Technical Field

[0001] This invention belongs to the field of smart grid energy dispatching technology, specifically relating to a source-grid-load-storage control device based on model prediction. Background Technology

[0002] The operation and management of energy installations are core areas for ensuring stable power supply and promoting green and low-carbon development in modern society. With the large-scale integration of renewable energy sources such as wind and solar power, the importance of energy installations is increasingly prominent, as they not only concern the efficient use of energy but also directly impact economic operations and environmental protection. However, the volatility and uncertainty of renewable energy pose serious challenges to the scheduling and management of traditional energy installations. How to achieve efficient energy allocation and stable operation in a complex and ever-changing environment has become a critical issue that urgently needs to be addressed in the energy sector.

[0003] Traditional energy dispatching methods have significant limitations in addressing the volatility of renewable energy. Existing dispatching strategies largely rely on historical data analysis or short-term forecasts, making it difficult to adapt to the rapid changes in new energy generation. For example, forecasting models based on fixed time periods often fail to capture sudden weather changes or drastic fluctuations in load demand, leading to a disconnect between dispatching plans and actual operating conditions. This static dispatching approach struggles to achieve coordinated optimization across the power generation, grid, load, and storage systems in the face of complex and ever-changing energy environments, thus impacting the stability and economic efficiency of power installations.

[0004] Against this backdrop, the core technical challenge in energy device regulation lies in how to achieve dynamic adjustments to operating strategies. The volatility of renewable energy generation requires dispatching devices to respond quickly to changes in the external environment, such as sudden changes in wind speed or a sharp drop in power generation due to cloud cover. However, current dispatching devices are often limited by insufficient computing speed and prediction accuracy when processing real-time data. This deficiency directly leads to the device's inability to generate operating strategies adapted to the current situation in a timely manner. For example, when wind power generation suddenly decreases, the device may be unable to promptly mobilize energy storage devices or adjust load distribution due to a lag in response, resulting in insufficient power supply or resource waste.

[0005] Looking further, the challenge of dynamically adjusting operating strategies lies not only in rapid response but also in achieving global optimization under multi-objective constraints. In energy systems, the coordination between power sources, the grid, loads, and energy storage devices requires comprehensive consideration of multiple objectives, including economic efficiency, stability, and environmental friendliness. However, existing technologies often sacrifice real-time performance due to high computational complexity when handling multi-objective optimization. For example, during peak electricity consumption periods, the system needs to ensure stable power supply while minimizing reliance on high-cost backup power sources. Current scheduling algorithms often struggle to find a balance that addresses the needs of all parties in a short time, leading to inefficient resource allocation.

[0006] Therefore, how to achieve efficient coordinated operation of power generation, grid, load, and storage through a combination of rapid response and multi-objective optimization in the context of highly volatile renewable energy sources and complex operating environments has become a key issue in the field of energy device regulation. The core of this issue lies in the fact that, faced with rapidly changing power generation and consumption demands, devices must complete accurate forecasting and dynamic adjustments within a limited timeframe, while simultaneously balancing multiple operational objectives to avoid power outages or resource waste. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a source-grid-load-storage regulation device based on model prediction, aiming to achieve efficient coordinated operation of source, grid, load, and storage.

[0008] To achieve the above objectives, the present invention provides the following solution:

[0009] A source-grid-load-storage regulation device based on model prediction, the device comprising: data acquisition equipment, model building equipment, model correction equipment, and regulation equipment;

[0010] Data acquisition equipment is used to acquire and preprocess real-time source-grid-load-storage data based on a comprehensive data acquisition system covering source-side power generation units, grid-side power transmission networks, load-side power consumption terminals, and energy storage devices to obtain a predictive dataset.

[0011] Model building equipment is used to build new energy output prediction models and load prediction models based on preset prediction time scales;

[0012] The model correction device is used to obtain the output parameters of the new energy output prediction model and the load prediction model based on the prediction dataset, and compare the output parameters with the actual operating data to correct the errors of the new energy output prediction model and the load prediction model.

[0013] The control equipment is used to solve the multi-objective optimization function of source-grid-load-storage based on the modified new energy output prediction model and load prediction model, and obtain the optimized source-grid-load-storage allocation scheme.

[0014] Preferably, the real-time source-grid-load-storage data acquired by the data acquisition device includes:

[0015] New energy output data: power output of wind power and photovoltaic power, and weather forecast data;

[0016] Load demand data: historical load curves, real-time electricity consumption data;

[0017] Energy storage device status: SOC, charge / discharge efficiency, capacity;

[0018] Power grid operating parameters: node voltage, line power flow, and reserve capacity.

[0019] Preferably, the new energy output prediction model constructed by the model building equipment includes:

[0020] Constructing a physical-data hybrid wind power output prediction model:

[0021] Physics section: Based on the real-time wind speed, cut-in wind speed, cut-out wind speed and rated wind speed of the wind turbine, the constraints are constructed. At the same time, based on the air density, the swept area of ​​the wind turbine blades, the wind energy utilization coefficient, the tip speed ratio, the pitch angle and the rated power of the wind power, the predicted wind power is obtained.

[0022] Data-driven part: Based on the LSTM error correction function, the real-time wind speed, wind direction and historical power sequence of the wind turbine, the corrected predicted power is obtained;

[0023] In low-wind-speed regions, a Gaussian process is used to replace the deterministic power curve.

[0024] Preferably, the load forecasting model constructed by the model building device includes:

[0025] Constructing a multi-scale prediction framework:

[0026] Minute-level forecasting is performed using time series decomposition.

[0027] Hourly forecasts based on the impact of weather and economic activity;

[0028] Use graph attention networks to model load correlation.

[0029] Preferably, the control device includes:

[0030] A multi-objective optimization function for source-grid-load-storage is constructed based on economic objectives, environmental objectives, power grid stability objectives, and renewable energy consumption objectives.

[0031] A user default penalty term is added to the objective function, and the penalty coefficient is dynamically adjusted according to the grid reserve capacity.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. Improve the renewable energy absorption rate and grid stability: By using physical-data hybrid wind power prediction models and multi-scale load prediction models, the prediction error of wind and solar power can be reduced, and the curtailment of wind and solar power can be reduced; by using a penalty mechanism coupled with dynamic reserve capacity, the renewable energy absorption rate can be improved, while avoiding frequency overruns caused by insufficient reserves.

[0034] 2. Reduce operating costs and carbon emissions: Multi-objective optimization reduces the operating costs of coal-fired power units and reduces carbon emissions; user default penalties can guide demand response, reduce the number of standby unit start-ups and shutdowns during peak hours, and greatly save operating costs.

[0035] 3. Enhanced response speed of power generation, grid, load and storage coordination: Minute-level prediction-rolling optimization closed loop shortens the delay of dispatch command update, which is better than traditional static dispatch; the dynamic adjustment response speed of energy storage and unit output is improved, effectively suppressing voltage fluctuations caused by sudden drops in wind and solar power.

[0036] 4. Improve robustness in extreme scenarios: In sudden scenarios such as typhoon weather or sudden load surges, the number of times the device voltage exceeds the limit and the risk of power outages can be reduced; through the topology awareness capability of the GAT network, the power distribution of associated nodes can be automatically adjusted when the regional load is unbalanced.

[0037] 5. Promote refined operation of the electricity market: Dynamic penalty coefficients provide a basis for the design of differentiated electricity prices, which can effectively enhance users' enthusiasm for participating in demand response; new energy power plants can increase their revenue by participating in spot market quotations based on forecast results. Attached Figure Description

[0038] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of a source-grid-load-storage regulation device based on model prediction, according to an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of the control device according to an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Example 1

[0044] like Figure 1 As shown, the present invention provides a source-grid-load-storage regulation device based on model prediction. The device includes: a data acquisition device, a model building device, a model correction device, and a regulation device.

[0045] Data acquisition equipment is used to acquire and preprocess real-time source-grid-load-storage data based on a comprehensive data acquisition system covering source-side power generation units, grid-side power transmission networks, load-side power consumption terminals, and energy storage devices to obtain a predictive dataset.

[0046] Model building equipment is used to build new energy output prediction models and load prediction models based on preset prediction time scales;

[0047] The model correction device is used to obtain the output parameters of the new energy output prediction model and the load prediction model based on the prediction dataset, and compare the output parameters with the actual operating data to correct the errors of the new energy output prediction model and the load prediction model.

[0048] The control equipment is used to solve the multi-objective optimization function of source-grid-load-storage based on the modified new energy output prediction model and load prediction model, and obtain the optimized source-grid-load-storage allocation scheme.

[0049] Furthermore, the real-time source-grid-load-storage data acquired by the data acquisition equipment includes:

[0050] New energy output data: power output of wind power and photovoltaic power, and weather forecast data;

[0051] Load demand data: historical load curves, real-time electricity consumption data;

[0052] Energy storage device status: SOC, charge / discharge efficiency, capacity;

[0053] Power grid operating parameters: node voltage, line power flow, and reserve capacity.

[0054] Furthermore, the new energy output prediction model constructed by the model building equipment includes:

[0055] Constructing a physical-data hybrid wind power output prediction model:

[0056] Physics section: Based on the real-time wind speed, cut-in wind speed, cut-out wind speed and rated wind speed of the wind turbine, the constraints are constructed. At the same time, based on the air density, the swept area of ​​the wind turbine blades, the wind energy utilization coefficient, the tip speed ratio, the pitch angle and the rated power of the wind power, the predicted wind power is obtained.

[0057] Data-driven part: Based on the LSTM error correction function, the real-time wind speed, wind direction and historical power sequence of the wind turbine, the corrected predicted power is obtained;

[0058] In low-wind-speed regions, a Gaussian process is used to replace the deterministic power curve.

[0059] Furthermore, the load forecasting model built by the model building equipment includes:

[0060] Constructing a multi-scale prediction framework:

[0061] Minute-level forecasting is performed using time series decomposition.

[0062] Hourly forecasts based on the impact of weather and economic activity;

[0063] Use graph attention networks to model load correlation.

[0064] Furthermore, the control device, such as Figure 2 As shown, it includes:

[0065] A multi-objective optimization function for source-grid-load-storage is constructed based on economic objectives, environmental objectives, power grid stability objectives, and renewable energy consumption objectives.

[0066] A user default penalty term is added to the objective function, and the penalty coefficient is dynamically adjusted according to the grid reserve capacity.

[0067] Example 2

[0068] This invention also provides a source-grid-load-storage regulation method based on model prediction, implemented using the apparatus described in the foregoing embodiments, the method comprising:

[0069] Based on a comprehensive data acquisition system covering source-side power generation units, grid-side power transmission networks, load-side power consumption terminals, and energy storage devices, real-time source-grid-load-storage data are acquired and preprocessed to obtain a predictive dataset.

[0070] Based on the preset prediction time scale, a new energy output prediction model and a load prediction model are constructed.

[0071] Based on the prediction dataset, the output parameters of the new energy output prediction model and the load prediction model are obtained, and the output parameters are compared with the actual operating data to correct the errors of the new energy output prediction model and the load prediction model.

[0072] Based on the revised new energy output prediction model and load prediction model, the particle swarm optimization algorithm is used to solve the multi-objective optimization function of source-grid-load-storage, and obtain the optimized source-grid-load-storage allocation scheme.

[0073] Furthermore, the specific implementation process of this invention is as follows:

[0074] Based on a comprehensive data acquisition system covering source-side power generation units, grid-side transmission networks, load-side power consumption terminals, and energy storage devices, real-time source-grid-load-storage data are acquired and preprocessed to obtain a predictive dataset, including:

[0075] Based on distributed sensing nodes and a multi-protocol fusion communication network, measurement terminals are simultaneously deployed on the source, grid, load, and storage sides to perform microsecond-level aligned sampling of multi-source heterogeneous physical quantities such as power generation output, node voltage, device frequency, equipment temperature, and battery SOC. Subsequently, an adaptive filtering algorithm is used to remove noise and outliers, and finally, a high-quality standardized data stream with a unified clock and format is output to obtain a prediction dataset, providing a panoramic, real-time, and reliable digital foundation for subsequent analysis.

[0076] Real-time source-grid-load-storage data includes:

[0077] New energy output data: power output of wind power and photovoltaic power, and weather forecast data;

[0078] Load demand data: historical load curves, real-time electricity consumption data;

[0079] Energy storage device status: SOC, charge / discharge efficiency, capacity;

[0080] Power grid operating parameters: node voltage, line power flow, and reserve capacity.

[0081] Furthermore, based on a pre-defined forecast time scale, a new energy output forecasting model and a load forecasting model are constructed, including:

[0082] Based on the forecast time scale required by the dispatch plan, which covers multiple levels including ultra-short-term (15 minutes to 4 hours), short-term (1 to 3 days), and medium-to-long-term (1 to 4 weeks), output forecast models for wind power and new energy and load forecast models for the entire process of power generation, grid, load, and storage are built respectively.

[0083] Constructing a physical-data hybrid wind power output prediction model:

[0084] Physics section:

[0085]

[0086] Where v is the real-time wind speed, v cut-in To cut off the wind speed, v cut-out To cut off the wind speed, v rated Where ρ is the rated wind speed, A is the air density, and C is the swept area of ​​the fan blades. p λ is the wind energy utilization coefficient, β is the tip speed ratio, and P is the blade pitch angle. wind For wind power forecast, P rated This refers to the rated power of the wind power.

[0087] Data-driven part:

[0088]

[0089] Among them, The corrected predicted power is given by f, where t is time and f is the predicted power. LSTM (·) represents the LSTM error correction function, v(t) is the real-time wind speed at time t, θ(t) is the wind direction at time t, and P hist (tk:t) represents the historical power sequence from time tk to time t, and ∈(t) represents the residual term;

[0090] In low-wind-speed regions, a Gaussian process is used to replace the deterministic power curve:

[0091] P wind ~GP(f(v,TI),k Matern (v,v′));

[0092] Where GP(·) is the Gaussian function, f(·) is the mean function, TI is the turbulence intensity, and k Matern This is a kernel function used to capture the nonlinear relationship between wind speed and power, where v and v′ are both real-time wind speeds.

[0093] Furthermore, a load forecasting model is constructed, including:

[0094] Constructing a multi-scale prediction framework:

[0095] Minute-level forecasting using time series decomposition:

[0096] L(t) = T(t) + S(t) + R(t);

[0097] Where L is the minute-level forecast value, T is the trend term, S is the seasonal term, R is the residual term, and t is the time;

[0098] Hourly forecasts based on the impact of weather and economic activity:

[0099]

[0100] in, For hourly forecast values, L base The baseline load is given, ΔT is the temperature deviation, and I is the base load. holiday For holidays, α, β, and γ are dummy variables;

[0101] Meanwhile, this invention utilizes graph neural networks to reshape the entire "source-grid-load-storage" chain into a dynamic topology graph: power sources such as thermal power plants and wind farms are mapped as source-measurement vertices, grid equipment such as substations and overhead lines are mapped as grid-measurement vertices, power-consuming units such as residential communities and industrial parks are mapped as load-measurement vertices, and battery energy storage power stations are mapped as storage-measurement vertices. The connection between any two vertices is no longer just a physical wire, but an abstract relationship carrying power exchange, energy buffering, and regulation signals, thereby elevating the electrical connection to a learnable graph structure.

[0102] By using historical operating data to train the weights of edges between vertices end-to-end, the model can automatically capture the spatial adjacency strength and spatiotemporal coupling patterns. As the weight matrix is ​​continuously updated, the device's operating status is transformed into a high-dimensional graph signal, which can extract key features in the spatial dimension and extrapolate future trends along the temporal dimension, realizing state perception and sequence prediction from a unified perspective of source, grid, load, and storage.

[0103] Modeling load correlation using graph attention networks:

[0104] h i ′=σ(∑ j∈N(i) α ij Wh j );

[0105] Among them, h i Let ' be the updated variable of node i, σ(·) be the non-linear activation function, N(i) be the set of neighboring nodes of node i, and α be the variable of node i. ij Let α be the attention weight of node j to node i. ij ∈[0,1], W is the learnable parameter matrix, h j Let J represent the power consumption characteristics of node j.

[0106] Furthermore, based on the prediction dataset, the output parameters of the new energy output prediction model and the load prediction model are obtained, and the output parameters are compared with the actual operating data to correct the errors of the new energy output prediction model and the load prediction model, including:

[0107] Based on a full-cycle forecast dataset covering "ultra-short term 15 minutes to 4 hours, short term 1 to 3 days, and medium to long term 1 to 4 weeks", the constructed new energy output forecast model and load forecast model are first called to output the key parameters of the corresponding time scale. After the parameters are aligned with a unified clock and standardized in terms of units, they are written into the real-time database to form a "forecast-actual" parallel data pair.

[0108] Then, an online rolling verification and error closed-loop correction mechanism is activated to correct the errors in the new energy output prediction model and the load prediction model:

[0109] ① Using the actual measured active power, bus voltage, frequency, energy storage SOC and aggregated load of the unit by SCADA, AMI and PMU as the true values, calculate multi-dimensional error indicators - including point error (MAE, RMSE, NRMSE), interval error (CRPS, PICP, PINAW) and trend error (direction accuracy DA, ramp error RME).

[0110] ② Based on the error magnitude and trend deviation, a graded correction strategy is triggered: if the ultra-short-term NRMSE > 5% or DA < 85%, a 5-minute sliding window Bayesian linear regression (BLR) is immediately used to incrementally update the weights of the LSTM residual network, and the hyperparameters of the Gaussian process kernel function are simultaneously fine-tuned; if the short-term CRPS is higher than the threshold for 3 consecutive hours, the "meteorological-power" joint recalibration is initiated, and the latest meteorological data + numerical weather forecast are recoupled with the wind turbine physical conversion model to re-estimate the three-dimensional response surface of "tip speed ratio - blade pitch angle - wind energy utilization coefficient"; once the medium- and long-term deviation exceeds 3% of the monthly electricity consumption, the Prophet-SARIMA-X online parameter adaptive algorithm is introduced to perform rolling re-estimation of economic activity indices, etc.

[0111] ③ All correction results are written back into the new energy output prediction model and load prediction model via the digital twin platform, forming a closed loop of "prediction-comparison-correction-re-prediction". At the same time, the version number and error traceability log are recorded for subsequent interpretability analysis and model performance audit, thereby ensuring that the new energy output prediction model and load prediction model continue to converge and the accuracy steadily improves across the entire time scale. This provides highly reliable, probabilistic and traceable quantitative decision-making basis for source-grid-load-storage coordinated optimization scheduling, spot market clearing and reserve capacity assessment.

[0112] Furthermore, based on the revised new energy output prediction model and load prediction model, the particle swarm optimization algorithm is used to solve the multi-objective optimization function of source-grid-load-storage, obtaining an optimized source-grid-load-storage allocation scheme, including:

[0113] Based on the previous round of closed-loop correction, the new energy output prediction model and load prediction model have significantly reduced point errors and interval errors, and have entered the source-grid-load-storage allocation optimization stage.

[0114] Construct a multi-objective optimization function for source-grid-load-storage:

[0115] min(w1F1+w2F2+w3F3+w4F4);

[0116] Where w1, w2, w3, and w4 are the weight coefficients of F1, F2, F3, and F4, respectively, and F1, F2, F3, and F4 are the economic target, environmental protection target, power grid stability target, and new energy consumption target, respectively.

[0117] Add a user default penalty term to the objective function:

[0118] min(C grid +λ∑max(0,L actual -L contract ) 2 );

[0119] Among them, C gridLet L be the total operating cost of the power grid, λ be the penalty coefficient, and L be the total operating cost of the power grid. actual L represents the actual load power of the user. contract The maximum load limit for contracts signed with users.

[0120] The penalty coefficient λ is dynamically adjusted according to the grid reserve capacity.

[0121] λ=λ0(1-C reserve / C total );

[0122] Where λ0 is the basic penalty coefficient, λ0≥0, C reserve C represents the current available reserve capacity of the power grid. total This represents the upper limit of the total reserve capacity of the power grid.

[0123] To solve this high-dimensional, non-convex, multi-constraint, and multi-stochastic scenario problem, an improved adaptive hybrid discrete particle swarm optimization algorithm (HPSO-TVAC) is adopted:

[0124] ① Particle encoding uses a hybrid string of "real numbers + integers". The real number segment corresponds to the unit output and energy storage power, while the integer segment corresponds to the start-stop status and reactive power equipment level. ② Time-varying acceleration coefficients and chaotic disturbances are introduced into the particle velocity update to avoid premature convergence. ③ For stochastic scenarios, a hybrid Monte Carlo sampling + scenario reduction is embedded. The joint probability distribution of new energy and load is considered when evaluating each particle to achieve a dual measurement of "expected cost + conditional value at risk (CVaR)". ④ Network constraints are processed in two stages: "penalty function + feasible region projection". First, a fast power flow scan is performed on voltage and thermal stability. Then, a pull-back mapping is performed on individuals that exceed the limits to ensure that particles are always within the safe operating domain. ⑤ Energy storage lifetime loss is converted to the objective function using the rain-flow counting method to achieve a unified measurement of "energy-lifetime".

[0125] The algorithm uses a 15-minute rolling step and a 24-hour day-ahead optimization window. After 2000 iterations, it converges to obtain the Pareto front. Then, through fuzzy membership decision, it selects the unique solution with the highest overall satisfaction from the set of non-dominated solutions and outputs the optimized solution for source-grid-load-storage. After safety verification, the solution is automatically distributed to EMS, energy storage EMS, and demand response platform to achieve optimized allocation of source-grid-load-storage.

[0126] In summary, this invention provides a model-based prediction-driven source-grid-load-storage regulation method. By employing a physical-data hybrid wind power prediction model and a multi-scale load prediction model, it can reduce wind and solar power prediction errors and decrease wind and solar curtailment. A dynamic reserve capacity coupling penalty mechanism can improve the renewable energy absorption rate while avoiding frequency overruns caused by insufficient reserves. Multi-objective optimization reduces the operating costs of coal-fired power units and decreases carbon emissions. User default penalties can guide demand response, reducing the number of standby unit start-ups and shutdowns during peak hours, significantly saving operating costs. A minute-level prediction-rolling optimization closed loop shortens the delay in dispatch command updates, outperforming traditional static dispatch. The dynamic adjustment response speed of energy storage and unit output is improved, effectively suppressing voltage fluctuations caused by sudden drops in wind and solar power. In sudden scenarios such as typhoons or load surges, it can reduce the number of voltage overruns and the risk of power outages. Through the topology awareness capability of the GAT network, it can automatically adjust the power allocation of associated nodes when there is regional load imbalance. The dynamic penalty coefficient provides a basis for differentiated electricity pricing design, effectively increasing users' enthusiasm for participating in demand response. Renewable energy power plants can participate in spot market bidding based on prediction results to increase revenue.

[0127] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A source-grid-load-storage control device based on model prediction, characterized in that, The device includes: a data acquisition device, a model building device, a model correction device, and a control device; Data acquisition equipment is used to acquire and preprocess real-time source-grid-load-storage data based on a comprehensive data acquisition system covering source-side power generation units, grid-side power transmission networks, load-side power consumption terminals, and energy storage devices to obtain a predictive dataset. Model building equipment is used to build new energy output prediction models and load prediction models based on preset prediction time scales; The model correction device is used to obtain the output parameters of the new energy output prediction model and the load prediction model based on the prediction dataset, and compare the output parameters with the actual operating data to correct the errors of the new energy output prediction model and the load prediction model. The control equipment is used to solve the multi-objective optimization function of source-grid-load-storage based on the modified new energy output prediction model and load prediction model, and obtain the optimized source-grid-load-storage allocation scheme.

2. The source-grid-load-storage regulation device based on model prediction according to claim 1, characterized in that, The real-time source-grid-load-storage data acquired by the data acquisition equipment includes: New energy output data: power output of wind power and photovoltaic power, and weather forecast data; Load demand data: historical load curves, real-time electricity consumption data; Energy storage device status: SOC, charge / discharge efficiency, capacity; Power grid operating parameters: node voltage, line power flow, and reserve capacity.

3. The source-grid-load-storage regulation device based on model prediction according to claim 1, characterized in that, The new energy output prediction model constructed by the model building equipment includes: Constructing a physical-data hybrid wind power output prediction model: Physics section: Based on the real-time wind speed, cut-in wind speed, cut-out wind speed and rated wind speed of the wind turbine, the constraints are constructed. At the same time, based on the air density, the swept area of ​​the wind turbine blades, the wind energy utilization coefficient, the tip speed ratio, the pitch angle and the rated power of the wind power, the predicted wind power is obtained. Data-driven part: Based on the LSTM error correction function, the real-time wind speed, wind direction and historical power sequence of the wind turbine, the corrected predicted power is obtained; In low-wind-speed regions, a Gaussian process is used to replace the deterministic power curve.

4. The source-grid-load-storage regulation device based on model prediction according to claim 1, characterized in that, The load forecasting model built by the model building equipment includes: Constructing a multi-scale prediction framework: Minute-level forecasting is performed using time series decomposition. Hourly forecasts based on the impact of weather and economic activity; Use graph attention networks to model load correlation.

5. The source-grid-load-storage regulation device based on model prediction according to claim 1, characterized in that, The control device includes: A multi-objective optimization function for source-grid-load-storage is constructed based on economic objectives, environmental objectives, power grid stability objectives, and renewable energy consumption objectives. A user default penalty term is added to the objective function, and the penalty coefficient is dynamically adjusted according to the grid reserve capacity.

Images