Optimized dispatching system and method based on wind-solar-storage combined power generation

Abstract

The invention relates to an optimal scheduling system and method based on wind-solar-storage combined power generation, and belongs to the technical field of new energy power generation scheduling and energy optimization control. The system comprises a data acquisition module, a computing center and a central control module, and the modules cooperate to realize accurate prediction of wind and light output and load demand. The data acquisition module collects wind and light output, load demand and energy storage state data in real time and carries out data processing. End-to-end optimization of a wind and light output and scheduling strategy is realized in the computing center through a Physics-informed DDPG integrated intelligent decision-making model; the central control module dynamically distributes output of each unit according to an optimization result to realize collaborative balance of each unit; the wind and light storage execution module adopts a power generation and energy storage unit designed by a flexible material, the installation adaptability and the operation toughness of equipment are improved, a scheduling instruction is executed, and closed-loop feedback control is formed. Compared with the prior art, a reliable solution can be provided for supply and demand balance, low-carbon operation and large-scale application of the new energy power system.

Description

Technical Field

[0001] This invention belongs to the field of new energy power generation dispatch and energy optimization control technology, specifically involving an optimized dispatch system and method based on wind, solar and energy storage combined power generation. Background Technology

[0002] In the field of new energy power systems, current renewable energy power generation, such as wind and solar power, is significantly random and volatile due to natural conditions. This makes it difficult to stably match user-side electricity load demand, leading to problems such as wind and solar curtailment and power supply gaps, resulting in energy waste and system instability. Meanwhile, existing wind-solar-storage dispatch systems have limited prediction accuracy, low algorithm efficiency, and are prone to getting trapped in local optima, making it difficult to achieve a multi-objective optimal balance between system operating costs, energy storage charging and discharging losses, and wind and solar curtailment rates. Furthermore, most dispatch schemes lack closed-loop feedback control mechanisms, failing to dynamically adjust the output of each unit based on real-time operating conditions. This further reduces the system's adaptability to wind and solar power output fluctuations and energy utilization efficiency, hindering the low-carbon transformation and large-scale application of new energy power systems. Summary of the Invention

[0003] The purpose of this invention is to address the problems mentioned in the background art by providing an optimized scheduling system and method based on wind, solar, and energy storage combined generation. This system can collect and process data in real time on user-side electricity consumption, real-time wind and solar power output, and energy storage unit status. It outputs predictions of wind and solar power output fluctuations and load changes through a Physics-informed DDPG integrated intelligent decision-making model with embedded physical constraints. With the goal of minimizing total energy loss within the scheduling cycle, it solves for the optimal operating power parameters of each device through an improved optimization algorithm. Furthermore, it uses a dynamic power allocation algorithm to regulate the power output ratio between distributed generation units and energy storage units. Combined with the deformation resistance and fluctuation tolerance characteristics of flexible material components, it avoids equipment overload. Simultaneously, it transmits actual power output and equipment status back at a second-level frequency, forming a closed-loop feedback. This effectively improves the wind and solar energy absorption capacity, reduces energy storage charging and discharging losses and wind and solar curtailment rates, fully balances system operating costs, and effectively ensures the supply-demand balance and operational stability of the new energy power system.

[0004] To achieve the above objectives, the technical solution of this invention is: an optimized dispatching system based on wind, solar, and energy storage combined power generation, comprising a data acquisition module, a computing center, and a central control module. These modules work together to accurately predict wind and solar power output and load demand, effectively reducing energy losses; wherein,

[0005] The data acquisition module collects three types of core data in real time, including user-side electricity load data, real-time wind and solar power output data of distributed generation units, and state of charge and charging / discharging power data of energy storage units, and processes the data.

[0006] The computing center uses data collected by the data acquisition module to solve the problem using the Physics-informed DDPG model with the objective function of "minimizing total energy loss within the scheduling cycle". The Physics-informed DDPG model relies on a deep deterministic policy gradient framework and learns temporal correlations through an Actor-Critic network to output predicted values ​​of wind and solar power output fluctuations and load changes. The predictions embed physical information to conform to actual constraints, thereby directly generating scheduling strategies. Multi-objective optimization is achieved through a reward function, ultimately generating the optimal output parameters for distributed generation and energy storage units, realizing dynamic prediction and end-to-end optimization of wind and solar power output and load demand.

[0007] The central control module, based on the optimized scheduling strategy obtained from the computing center, adjusts the output ratio of the wind, solar and storage execution modules in real time through a dynamic output allocation algorithm. It also combines the physical properties of flexible materials to match the environment and avoid equipment overload. During execution, the actual output and equipment status are transmitted back at a frequency of seconds. After comparing the deviation, the central control module triggers the instruction correction or protection mechanism, forming a closed loop of "optimization-execution-feedback-re-optimization", and outputs real-time control instructions for each device.

[0008] Furthermore, the wind-solar-storage execution module is used to accurately respond to the optimized scheduling instructions of the central control module, and the wind-solar-storage execution module transmits the actual operating data of the equipment back to the central control module, forming a closed-loop control link to ensure that the system operates stably and optimally with the goal of "minimizing the total energy loss within the scheduling cycle".

[0009] Furthermore, the data acquisition module processes the data in the following way:

[0010] (1) Cleaning of abnormal values ​​in user-side power load data: When the load power exceeds the rated capacity of the transformer, the system automatically returns to zero and triggers an overload alarm; if negative power occurs, it is marked as a back-to-grid event; for missing data, single-point missing data is filled by linear interpolation of two points before and after, and continuous missing data for more than 5 minutes is filled by simulating and adding labels using historical data under the same operating conditions; then, the sliding window mean filtering is used to smooth the micro-oscillation within ±5% range and retain the step change of equipment start-up and shutdown; finally, the capacity benchmark is used for normalization, and the actual power is divided by the rated capacity of the transformer to generate a standardized sequence in the [0,1] interval;

[0011] (2) The wind and solar power output data need to be verified by physical logic: when the photovoltaic irradiance exceeds 1500W / m², it is truncated to the upper limit of the range, the wind speed negative value is returned to zero and the sensor maintenance alarm is triggered; when the irradiance is >800W / m² and the photovoltaic output is less than 20% of the rated value, it is determined to be component shading and an operation and maintenance work order is generated; noise reduction adopts wavelet threshold layer processing: the sym5 wavelet basis is selected for 3-level decomposition, and the improved soft threshold function is applied to the high frequency coefficients; the normalization stage implements two-dimensional standardization: the wind turbine / photovoltaic output is divided by the rated power, the environmental parameters are normalized according to the upper limit of the range, and the four-dimensional feature vector is output;

[0012] (3) Energy storage data processing to ensure state of charge: When the SOC jump of adjacent sampling points is greater than 5%, the BMS communication failure is determined and cubic spline interpolation is used for smoothing; if there is a logical conflict in the charging and discharging power, the discharging power is forced to return to zero; the filtering adopts a composite strategy: first, the voltage acquisition spike is eliminated by median filtering with a 5-second window, and then the power change rate is constrained by limiting filtering with ±10% / second; the normalization adopts a bidirectional mapping mechanism: the SOC is directly divided by 100 and linearly compressed to [0,1], and the charging and discharging power is divided by their respective maximum allowable values.

[0013] Furthermore, the computing center constructs the objective function of "minimizing total energy loss within the scheduling cycle," which is specifically as follows:

[0014] 1) Use The total wind and solar power curtailment loss at time t is calculated using the following formula:

[0015]

[0016] Among them, wind curtailment loss , The actual wind power output at time t is the value that must satisfy subsequent wind and solar power output constraints. Contribute the most to wind power technology;

[0017] Waste light loss , The actual power output of photovoltaic power at time t. This is the component occlusion coefficient, which is 1 when occluded and 0 when unoccluded. Actual attenuation 20%, Contribute to the smallest technology in photovoltaics;

[0018] use The total energy storage charge / discharge loss at time t is represented by the following formula:

[0019]

[0020] Among them, charging loss This refers to the energy loss that occurs during the charging process of energy storage due to factors including efficiency. This refers to the charging efficiency under normal energy storage conditions. For BMS fault marking, It is an energy storage system that absorbs electrical energy, only when Loss occurs when the value is greater than 0;

[0021] Discharge loss This refers to the energy loss that occurs during the discharge process of energy storage due to factors including efficiency. This represents the discharge efficiency under normal conditions. It is the energy storage system that releases electrical energy, only when Loss occurs when the value is greater than 0;

[0022] 2) Assuming a scheduling period of 24 hours and a time step of 15 minutes, the total number of scheduling steps T = 24 / 0.25 = 96. The objective function formula is:

[0023]

[0024] The objective function described above aims to minimize the combined costs of the three types of energy storage within a 24-hour scheduling cycle, thereby achieving optimal scheduling of the wind, solar, and energy storage systems. In the formula, F represents the total energy loss within the 24-hour scheduling cycle, minF is a weighted multi-objective optimization function that minimizes the total cost by considering "solar curtailment cost," "energy storage cost," and "start-up and shutdown cost," and α represents the weight of wind and solar curtailment losses. Let t be the power lost due to light discard. Let be the energy storage charging / discharging loss power at time t, and β be the economic weighting coefficient for energy storage loss. Let be the equipment operating cost power at time t, and γ be the economic weighting coefficient of the equipment operating cost. For time step;

[0025] 3) The formula for the power output constraint of wind and solar power is as follows:

[0026]

[0027]

[0028] in, Contribute to the minimum technical effort in wind power Minimum power output for photovoltaic technology; introduction of upper limit for photovoltaic power output ;

[0029] The energy storage operation constraint formula is as follows:

[0030] ,

[0031] in , Rated charge and discharge power for energy storage;

[0032]

[0033] Let t be the state of charge of the energy storage unit. The minimum state of charge allowed for the energy storage unit. This represents the minimum permissible state of charge for the energy storage unit.

[0034] in , Furthermore, the SOC must satisfy temporal continuity;

[0035]

[0036] in This refers to the rated capacity of the energy storage.

[0037] The power balance constraint formula is as follows:

[0038]

[0039] in, Consumes electrical energy for user-side equipment. Power purchased / sold to the power grid.

[0040] Furthermore, the computing center solves the objective function using the Physics-informed DDPG model, specifically as follows:

[0041] 1) Model Input Layer Construction

[0042] Input state vector Integrating three types of features:

[0043] Load characteristics ;

[0044] Landscape features ;

[0045] Energy storage characteristics ;

[0046] in Let t be the standardized power of the user-side electrical load; The status marker of the load at time t; Let t be the standardized power generation capacity of wind power at time t; Let t be the standardized power generation of the photovoltaic system at time t; Let be the standardized wind speed at time t; Let be the standardized irradiance at time t; Let be the standardized value of the energy storage state of charge at time t; The standardized charging power for energy storage at time t; Let be the standardized discharge power of the stored energy at time t;

[0047] The vector dimension is 11, and the time-series input covers 24 hours of data;

[0048] 2) Actor-Critic Network and Physical Constraint Embedding

[0049] Actor network is based on Output scheduling action:

[0050] These correspond to the actual wind power output, actual photovoltaic power output, energy storage charging power, and energy storage discharging power at time t, respectively. The output layer mapping ensures the load constraint of the action.

[0051] Fengguang's output is limited to ,

[0052] ( );

[0053] Energy storage charging and discharging power is limited to , Furthermore, the SOC (State of Charge) is constrained to be between 20% and 80%.

[0054] The Critic network associates the reward function with the objective function, as shown in the formula:

[0055]

[0056] in, , , , These are the corresponding weighting coefficients used to balance multiple objectives such as "reducing losses," "meeting constraints," "equipment lifespan," and "grid dependence." , , Mark the constraint satisfaction;

[0057] 3) Model training and objective function solution

[0058] An initial experience replay pool with a capacity of 10,000 is set up. Each time step the model runs, actions are calculated based on real-time data. ,award and the next state ;

[0059] In each iteration, 32 samples are randomly drawn from the replay pool, and the target value of the next state is calculated using the target Critic network:

[0060]

[0061] Discount factor;

[0062] The Critic network update aims to "minimize the current Q-value and..." The target network is the mean squared error, and the parameters are adjusted using the Adam optimizer. The Actor network is updated with the goal of maximizing the Q-value of the Critic output, using a policy gradient ascent method to update the parameters, while gradient pruning is used to avoid parameter oscillations. Every 10 iterations, the target network parameters are adjusted. The model integrates the current network parameters proportionally; when the average cumulative reward fluctuation over 10 consecutive iterations is less than 1% and the action constraint satisfaction rate is greater than 99%, the model converges; at this point, the Actor network outputs the optimal scheduling action. Substituting these values ​​into the objective function yields the minimum total energy loss over a 24-hour period. ,and It is used directly as a control instruction.

[0063] Furthermore, the central control module regulates the output ratio of the wind, solar, and energy storage execution modules as follows:

[0064] First, the central control module receives the optimized scheduling strategy from the computing center; after receiving it, it performs a pre-verification of the real-time equipment status using the data acquisition module: if the fan gearbox temperature... Photovoltaic module temperature Based on the characteristics of flexible materials, such as high resistance to deformation and high tolerance to fluctuations, the target output can be finely adjusted using the following formula:

[0065]

[0066] in, This represents the actual corrected output of the wind turbine at time t; The rated operating output of the fan at time t; Let t be the equipment temperature of the fan; The actual corrected output of the photovoltaic system at time t; The rated power output of the photovoltaic system at time t; Let t be the temperature of the photovoltaic module at time t;

[0067] Next, the fine-tuned target parameters are converted into control commands using a dynamic output allocation algorithm:

[0068] Photovoltaic actuators, combined with real-time irradiance Adjusting the DC side voltage of the inverter The formula is

[0069]

[0070] in, Let t be the actual output voltage of the photovoltaic system at time t; This is the reference voltage for photovoltaics; The actual corrected output of the photovoltaic system at time t; This refers to the rated power of the photovoltaic system. This is the ratio of actual irradiance to reference irradiance.

[0071] in The reference voltage is the voltage at a rated irradiance of 1000 W / m².

[0072] Wind power actuators, based on real-time wind speed Adjusting the angle of the flexible pitch blades The formula is

[0073]

[0074] in This is the reference angle at rated wind speed;

[0075] Energy storage execution unit, based on real-time When switching converter topologies, the charging and discharging power must meet the following requirements:

[0076]

[0077]

[0078] in, The actual corrected charging power of the stored energy at time t; The maximum allowable charging power for energy storage; This represents the maximum state of charge allowed for energy storage. The current state of charge of the stored energy at time t; The actual corrected discharge power of the stored energy at time t; This represents the maximum permissible discharge power of the energy storage. This represents the minimum state of charge allowed for energy storage. The current state of charge of the stored energy at time t;

[0079] During execution, each unit transmits actual data back at a frequency of seconds: photovoltaic backhaul. Wind power backhaul Energy storage backhaul and Meanwhile, the central control module calculates the deviation:

[0080] Relative deviation in wind and solar power output: Absolute deviation of energy storage power: ;

[0081] in, The relative deviation of wind output at time t; The actual contribution of the scenery at time t; The output is adjusted to reflect the scenery at time t; Let be the absolute deviation of the energy storage power at time t; The actual power of the stored energy at time t; Let t be the corrected power of the stored energy.

[0082] If the deviation exceeds the threshold, a correction is triggered. After the correction command is issued, the execution unit adjusts and continues to send back data every second. The central control module repeats the deviation comparison and correction, forming a closed loop of "receiving strategy - transferring command - second feedback - correcting deviation - re-execution", ensuring that the output matches the loss reduction target, while using flexible materials to adapt to the environment and avoid overload.

[0083] Furthermore, the wind-solar-storage execution module, as the core of system instruction execution, has the following workflow:

[0084] The module receives the optimized scheduling instructions from the central control module and, in conjunction with its own status monitoring components, then the three main units—the photovoltaic execution unit, the wind power execution unit, and the energy storage execution unit—synchronously adjust their output: the wind power execution unit adjusts its output based on real-time wind speed. By adjusting the angle of the flexible blades using a pitch motor, the output is controlled at... The photovoltaic actuator adjusts the DC-side voltage through an intelligent inverter, while utilizing the flexible array panel's bendability to avoid localized shading, ensuring optimal power output. The energy storage execution unit reduces charging and discharging losses by switching the converter topology based on real-time SOC and the tolerance of the flexible battery pack.

[0085] Subsequently, the wind, solar, and energy storage execution module collects the operating data of each unit at a frequency of seconds and then sends it back to the data acquisition module, providing data basis for the central control module to calculate the deviation. If an equipment abnormality is detected, the wind, solar, and energy storage execution module will automatically trigger a local drop and prioritize sending back the abnormal signal, forming a secondary protection with the central control module to ensure the stable operation of the system and indirectly support the goal of minimizing total energy loss.

[0086] This invention also provides an optimized scheduling method based on wind, solar, and energy storage combined power generation, comprising the following steps:

[0087] Step (1): The data acquisition module collects three types of core data in real time, including user-side power load data, wind and solar power output data of distributed generation units, and state of charge and charging / discharging power data of energy storage units, and processes the data.

[0088] Step (2): The computing center uses the data constructed in step (1) to solve the problem through the Physics-informed DDPG model with the objective function of "minimizing total energy loss within the scheduling cycle". The Physics-informed DDPG model relies on the deep deterministic policy gradient framework and learns the temporal correlation through the Actor-Critic network to output the predicted values ​​of wind and solar power output fluctuations and load changes. The prediction embeds physical information to conform to actual constraints, and then directly generates the scheduling strategy. The reward function realizes multi-objective optimization and finally generates the optimal output parameters of distributed generation and energy storage units, realizing the dynamic prediction and end-to-end optimization of wind and solar power output and load demand.

[0089] Step (3): The central control module adjusts the output ratio of the wind, solar and energy storage execution modules in real time through the dynamic output allocation algorithm based on the optimization scheduling strategy of the computing center. It also matches the environment with the physical properties of flexible materials to avoid equipment overload. During execution, the actual output and equipment status are transmitted back at a frequency of seconds. After comparing the deviation, the central control module triggers the instruction correction or protection mechanism to form a closed loop of "optimization-execution-feedback-re-optimization".

[0090] Furthermore, the wind-solar-storage execution module is used to accurately respond to the optimized scheduling instructions of the central control module, and the wind-solar-storage execution module transmits the actual operating data of the equipment back to the central control module, forming a closed-loop control link to ensure that the system operates stably and optimally with the goal of "minimizing the total energy loss within the scheduling cycle".

[0091] The present invention also provides a computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor, wherein when the processor executes the computer program instructions, it can implement the steps of the method described above.

[0092] Compared with the prior art, the present invention has the following beneficial effects:

[0093] 1. This invention employs a multi-source data fusion prediction model with an attention mechanism, combining the ability of temporal deep learning to capture the temporal characteristics of wind and solar power output with the ability of statistical learning to accurately characterize the energy storage status. Compared with a single prediction model, this significantly reduces the prediction error of wind and solar power output and available energy storage capacity, providing a more reliable basis for the joint scheduling of wind, solar and energy storage.

[0094] 2. This invention uses minimizing wind and solar curtailment rates, overall system operating costs, and energy storage charging and discharging losses as multiple objective functions. It employs an improved particle swarm optimization algorithm with an adaptive weighting strategy to quickly determine the optimal output coordination scheme for wind power, photovoltaic, and energy storage units, achieving an optimal balance between energy consumption, economic costs, and equipment losses. This significantly improves the level of new energy consumption and the economic efficiency of system operation. At the same time, by introducing an adaptive weighting strategy and a chaotic perturbation mechanism, this improved algorithm enhances both the breadth of global optimization and the ability to perform fine-grained local searches, effectively avoiding getting trapped in local optima and significantly improving the efficiency and quality of solving scheduling schemes.

[0095] 3. The closed-loop feedback control mechanism constructed in this invention can dynamically adjust the scheduling instructions of each unit according to the real-time fluctuations in wind and solar power output, load changes and the actual status of energy storage, which greatly enhances the system's adaptability to the randomness and volatility of new energy sources, and further ensures the stability and continuous efficiency of the new energy power system operation. Attached Figure Description

[0096] Figure 1 This is a structural framework diagram of the present invention;

[0097] Figure 2 This is a flowchart of the algorithm optimization process of the present invention;

[0098] Figure 3 This is a comparison diagram of wind power dispatch strategies between the present invention and traditional dispatching methods;

[0099] Figure 4 This is a comparison chart of the cumulative energy consumption of the present invention and traditional scheduling. Detailed Implementation

[0100] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0101] This invention provides an optimized scheduling system based on wind, solar and energy storage combined power generation, including a data acquisition module, a computing center and a central control module. The modules work together to achieve accurate prediction of wind and solar power output and load demand, effectively reducing energy loss.

[0102] This invention also provides an optimized scheduling method based on wind, solar, and energy storage combined power generation. This method is implemented based on the above system and specifically includes the following steps:

[0103] First, the data acquisition module performs data acquisition and processing, and the specific steps are as follows:

[0104] (1) User-side power load data anomaly cleaning: when the load power exceeds the transformer's rated capacity, the system automatically resets to zero and triggers an overload alarm; if negative power occurs, it is marked as a back-to-grid event. For missing data, single-point missing data is filled by linear interpolation of two points before and after the missing data. If the missing data is continuously missing for more than 5 minutes, historical data under the same operating conditions is called to simulate and fill the missing data and add labels. Then, a sliding window mean filter is used to smooth the micro-oscillations within ±5% range and retain the step changes of equipment start-up and shutdown. Finally, capacity benchmark normalization is used, and the actual power is divided by the transformer's rated capacity to generate a standardized sequence in the [0,1] interval.

[0105] (2) Wind and solar power output data require physical logic verification: When the photovoltaic irradiance exceeds 1500W / m², it is truncated to the upper limit of the range, the wind speed is set to zero and a sensor maintenance alarm is triggered. When the irradiance is >800W / m² and the photovoltaic output is less than 20% of the rated value, it is determined to be module shading and an operation and maintenance work order is generated. Denoising adopts wavelet threshold layer processing: the sym5 wavelet basis is selected for 3-level decomposition, and an improved soft threshold function is applied to the high-frequency coefficients. The normalization stage implements two-dimensional standardization: the wind turbine / photovoltaic output is divided by the rated power, the environmental parameters are normalized according to the upper limit of the range, and a four-dimensional feature vector is output.

[0106] (3) Energy storage data processing to ensure state of charge: When the SOC jump between adjacent sampling points is greater than 5%, a BMS communication fault is determined and cubic spline interpolation is used for smoothing; if there is a logical conflict in the charging and discharging power, the discharging power is forced to zero. The filtering adopts a composite strategy: first, the median filter with a 5-second window is used to eliminate voltage acquisition spikes, and then the power change rate is constrained by the ±10% / second limiting filter. Normalization adopts a bidirectional mapping mechanism: the SOC is directly divided by 100 and linearly compressed to [0,1], and the charging and discharging power are divided by their respective maximum allowable values.

[0107] Secondly, the computing center constructs an objective function that minimizes total energy loss within the scheduling cycle, specifically:

[0108] (1) Use The total wind and solar power curtailment loss at time t is calculated using the following formula:

[0109]

[0110] Among them, wind curtailment loss , The actual wind power output at time t is the actual dispatch output of wind power, and this value must meet the subsequent wind and solar power output constraints.

[0111] Waste light loss , The actual power output of photovoltaic power at time t. This is because the component is obstructed. The actual attenuation was 20%.

[0112] The total energy storage charge / discharge loss at time t is represented by the following formula:

[0113]

[0114] Among them, charging loss , This refers to the charging efficiency under normal energy storage conditions. Mark a BMS fault only when Loss occurs when the value is greater than 0;

[0115] Discharge loss , The discharge efficiency under normal conditions is only when Losses occur when the value is greater than 0.

[0116] (2) Assume the scheduling period is 24 hours, the time step is 15 minutes, the total number of scheduling steps is T = 24 / 0.25 = 96, and the objective function formula is:

[0117]

[0118] In the formula, F represents the total energy loss during a 24-hour dispatch cycle, α represents the weight of wind and solar power curtailment losses, and γ represents the penalty coefficient for abnormal operating conditions. For time step.

[0119] (3) The formula for the power output constraint of wind and solar power is as follows:

[0120]

[0121]

[0122] in, Contribute to the minimum technical effort in wind power Minimum power output for photovoltaic technology; introduction of upper limit for photovoltaic power output .

[0123] The energy storage operation constraint formula is as follows:

[0124] ,

[0125] in , Rated charge and discharge power for energy storage

[0126]

[0127] in , Furthermore, the SOC must satisfy temporal continuity;

[0128]

[0129] in Rated capacity of energy storage

[0130] The power balance constraint formula is as follows:

[0131]

[0132] in, Power purchased / sold to the power grid.

[0133] In this example, the objective function is solved using the Physics-informed DDPG at the computing center, specifically as follows:

[0134] (1) Model input layer construction

[0135] Input state vector Integrating three types of features:

[0136] Load characteristics ;

[0137] Landscape features ;

[0138] Energy storage characteristics ;

[0139] The vector dimension is 11, and the time-series input covers 24 hours of data.

[0140] (2) Actor-Critic Network and Physical Constraint Embedding

[0141] Actor network is based on Output scheduling action These correspond to the actual wind power output, actual photovoltaic power output, energy storage charging power, and energy storage discharging power at time t, respectively. The output layer mapping ensures the load constraint of the action.

[0142] Fengguang's output is limited to ,

[0143] ( );

[0144] Energy storage charging and discharging power is limited to , Furthermore, the SOC is constrained to be between 20% and 80%.

[0145] The Critic network associates the reward function with the objective function, as shown in the formula:

[0146]

[0147] in To satisfy the constraint, mark the condition. For grid interaction power, , These parameters ensure that the rewards are aligned with the "loss reduction" objective.

[0148] (3) Model training and objective function solution

[0149] An initial experience replay pool with a capacity of 10,000 is set up. Each time step the model runs, actions are calculated based on real-time data. ,award and the next state .

[0150] In each iteration, 32 samples are randomly drawn from the replay pool, and the target value of the next state is calculated using the target Critic network. ( (Discount factor)

[0151] The Critic network update aims to "minimize the current Q-value and..." The target network aims to improve the accuracy of value assessment by adjusting parameters using the Adam optimizer, with the goal of "maximizing the Q-value of the Critic output". The Actor network update aims to "maximize the Q-value of the Critic output", employing a policy gradient ascent method to update parameters, while using gradient pruning to avoid parameter oscillations. Every 10 iterations, the target network parameters are adjusted to... The current network parameters are proportionally integrated to maintain training stability. The model converges when the following conditions are met: "average cumulative reward fluctuation over 10 consecutive iterations < 1%" and "action constraint satisfaction rate > 99%". At this point, the Actor network outputs the optimal scheduling action. Substituting these values ​​into the objective function yields the minimum total energy loss over a 24-hour period. Furthermore, this action can be directly used as a control command.

[0152] Then, the central control module receives the control strategy from the computing center and adjusts the output ratio of each unit (wind, solar, and energy storage). The specific execution process is as follows:

[0153] The central control module receives the optimized scheduling strategy from the computing center. After receiving the strategy, it performs a pre-verification of the real-time equipment status using the data acquisition module: if the fan gearbox temperature... Photovoltaic module temperature Based on the characteristics of flexible materials, such as high resistance to deformation and high tolerance to fluctuations, the target output can be finely adjusted using the following formula:

[0154]

[0155] The fine-tuned target parameters are converted into control commands through a dynamic power output allocation algorithm.

[0156] Photovoltaic actuators, combined with real-time irradiance Adjusting the DC side voltage of the inverter The formula is

[0157]

[0158] in This is the reference voltage under a rated irradiance of 1000 W / m².

[0159] Wind power actuators, based on real-time wind speed Adjusting the angle of the flexible pitch blades The formula is

[0160]

[0161] in This is the reference angle at rated wind speed.

[0162] Energy storage execution unit, based on real-time When switching converter topologies, the charging and discharging power must meet the following requirements:

[0163]

[0164]

[0165] During execution, each unit transmits actual data back at a frequency of seconds: photovoltaic backhaul. Wind power backhaul Energy storage backhaul and Meanwhile, the central control module calculates the deviation:

[0166] Relative deviation in wind and solar power output: ( (Representing wind / solar power, with a threshold of 3%)

[0167] Absolute deviation of energy storage power:

[0168] If the deviation exceeds the threshold, a correction is triggered. After the correction command is issued, the execution unit adjusts and continues to send back data every second. The module repeats the deviation comparison and correction, forming a closed loop of "receiving strategy - transferring command - second feedback - correcting deviation - re-execution", ensuring that the output matches the loss reduction target, while using flexible materials to adapt to the environment and avoid overload.

[0169] Finally, the wind-solar-storage execution module, as the core of system instruction execution, has the following workflow:

[0170] The module receives the optimized scheduling instructions from the central control module and, in conjunction with its own status monitoring components, the three main units simultaneously adjust their output: the wind power execution unit adjusts its output based on real-time wind speed. By adjusting the angle of the flexible blades using a pitch motor, the output is controlled at... The photovoltaic actuator adjusts the DC-side voltage through an intelligent inverter, while utilizing the flexible array panel's bendability to avoid localized shading, ensuring optimal power output. The energy storage execution unit reduces charging and discharging losses based on the real-time SOC switching converter topology and the tolerance of the flexible battery pack.

[0171] Subsequently, the module collects the operating data of each unit at a frequency of seconds and sends it back to the data acquisition module, providing data basis for the central control module to calculate the deviation. If an equipment abnormality is detected, the execution unit will automatically trigger a local drop and send back the abnormal signal first, forming a two-level protection with the central control module to ensure the stable operation of the system and indirectly support the goal of minimizing total energy consumption.

[0172] Figure 3 This invention visually demonstrates the scheduling advantages of its core technology. Compared to traditional scheduling methods, this invention's scheduling more closely follows the fluctuation curve of available wind power, indicating that the Physics-informed DDPG algorithm significantly reduces wind curtailment through accurate prediction of wind power output and adaptive adjustment of flexible materials. Traditional methods, due to response lag and rigid adjustment, result in the waste of some wind power, while this invention, through intelligent decision-making and physical constraint embedding, maximizes wind power utilization, directly verifying the technical effect of "reducing the curtailment rate to 4.2%" stated in the patent.

[0173] Figure 4 A direct comparison of the cumulative energy consumption performance of the traditional scheduling method and the Physics-informed DDPG algorithm of this invention is presented. The red dashed line in the figure represents the traditional method, with a total loss of 58.0 MWh, while the green solid line shows the method of this invention, with a total loss reduced to 34.1 MWh, a reduction of 41.2%. This result confirms that the algorithm described in the patent, through intelligent decision-making and flexible material adaptation, significantly improves the wind and solar energy absorption rate and achieves the goal of minimizing energy loss.

[0174] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An optimized dispatching system based on wind-solar-storage combined power generation, characterized in that, It includes a data acquisition module, a computing center, and a central control module. These modules work together to achieve accurate forecasting of wind and solar power output and load demand. The data acquisition module collects three types of core data in real time, including user-side electricity load data, real-time wind and solar power output data of distributed generation units, and state of charge and charging / discharging power data of energy storage units, and processes the data. The computing center uses data collected by the data acquisition module to solve the problem using the Physics-informed DDPG model with the objective function of "minimizing total energy loss within the scheduling cycle". The Physics-informed DDPG model relies on a deep deterministic policy gradient framework and learns temporal correlations through an Actor-Critic network to output predicted values ​​of wind and solar power output fluctuations and load changes. The predictions embed physical information to conform to actual constraints, thereby directly generating scheduling strategies. Multi-objective optimization is achieved through a reward function, ultimately generating the optimal output parameters for distributed generation and energy storage units, realizing dynamic prediction and end-to-end optimization of wind and solar power output and load demand. The central control module, based on the optimized scheduling strategy obtained from the computing center, adjusts the output ratio of the wind, solar and storage execution modules in real time through a dynamic output allocation algorithm. It also combines the physical properties of flexible materials to match the environment and avoid equipment overload. During execution, the actual output and equipment status are transmitted back at a frequency of seconds. After comparing the deviation, the central control module triggers the instruction correction or protection mechanism, forming a closed loop of "optimization-execution-feedback-re-optimization", and outputs real-time control instructions for each device.

2. The optimized dispatching system based on wind-solar-storage combined power generation according to claim 1, characterized in that, The wind-solar-storage execution module is used to accurately respond to the optimized scheduling instructions of the central control module, and the wind-solar-storage execution module transmits the actual operating data of the equipment back to the central control module, forming a closed-loop control link to ensure that the system operates stably and optimally with the goal of "minimizing the total energy loss within the scheduling cycle".

3. The optimized dispatching system based on wind-solar-storage combined power generation according to claim 1, characterized in that, The data acquisition module processes data in the following way: (1) Cleaning of abnormal values ​​in user-side power load data: When the load power exceeds the rated capacity of the transformer, the system automatically returns to zero and triggers an overload alarm; if negative power occurs, it is marked as a back-to-grid event; for missing data, single-point missing data is filled by linear interpolation of two points before and after, and continuous missing data for more than 5 minutes is filled by simulating and adding labels using historical data under the same operating conditions; then, the sliding window mean filtering is used to smooth the micro-oscillation within ±5% range and retain the step change of equipment start-up and shutdown; finally, the capacity benchmark is used for normalization, and the actual power is divided by the rated capacity of the transformer to generate a standardized sequence in the [0,1] interval; (2) The power output data of wind and solar power need to be verified by physical logic: when the photovoltaic irradiance exceeds 1500W / m², it is cut off to the upper limit of the range, the wind speed negative value is returned to zero and the sensor maintenance alarm is triggered; when the irradiance is >800W / m² and the photovoltaic power output is less than 20% of the rated value, it is determined that the module is blocked and an operation and maintenance work order is generated. Denoising employs wavelet threshold layering: the sym5 wavelet basis is used for 3-level decomposition, and an improved soft threshold function is applied to the high-frequency coefficients; the normalization stage implements two-dimensional standardization: the wind turbine / photovoltaic output is divided by the rated power, environmental parameters are normalized according to the upper limit of the range, and a four-dimensional feature vector is output. (3) Energy storage data processing to ensure state of charge: When the SOC jump of adjacent sampling points is greater than 5%, the BMS communication failure is determined and cubic spline interpolation is used for smoothing; if there is a logical conflict in the charging and discharging power, the discharging power is forced to return to zero; the filtering adopts a composite strategy: first, the voltage acquisition spike is eliminated by median filtering with a 5-second window, and then the power change rate is constrained by limiting filtering with ±10% / second; the normalization adopts a bidirectional mapping mechanism: the SOC is directly divided by 100 and linearly compressed to [0,1], and the charging and discharging power is divided by their respective maximum allowable values.

4. The optimized dispatching system based on wind-solar-storage combined power generation according to claim 1, characterized in that, The computing center has constructed an objective function to "minimize total energy loss within a scheduling cycle," which is specifically as follows: 1) Use The total wind and solar power curtailment loss at time t is calculated using the following formula: Among them, wind curtailment loss , The actual wind power output at time t is the value that must satisfy subsequent wind and solar power output constraints. Contribute the most to wind power technology; Waste light loss , The actual power output of photovoltaic power at time t. This is the component occlusion coefficient, which is 1 when occluded and 0 when unoccluded. Actual attenuation 20%, Contribute to the smallest technology in photovoltaics; use The total energy storage charge / discharge loss at time t is represented by the following formula: Among them, charging loss This refers to the energy loss that occurs during the charging process of energy storage due to factors including efficiency. This refers to the charging efficiency under normal energy storage conditions. For BMS fault marking, It is an energy storage system that absorbs electrical energy, only when Loss occurs when the value is greater than 0; Discharge loss This refers to the energy loss that occurs during the discharge process of energy storage due to factors including efficiency. This represents the discharge efficiency under normal conditions. It is the energy storage system that releases electrical energy, only when Loss occurs when the value is greater than 0; 2) Assuming a scheduling period of 24 hours and a time step of 15 minutes, the total number of scheduling steps T = 24 / 0.25 = 96. The objective function formula is: The objective function described above aims to minimize the combined costs of the three types of energy storage within a 24-hour scheduling cycle, thereby achieving optimal scheduling of the wind, solar, and energy storage systems. In the formula, F represents the total energy loss within the 24-hour scheduling cycle, minF is a weighted multi-objective optimization function that minimizes the total cost by optimizing the "curtailment cost," "energy storage cost," and "start-up / shutdown cost," and α represents the weight of wind and solar curtailment costs. Let t be the power lost due to light discard. Let be the energy storage charging / discharging loss power at time t, and β be the economic weighting coefficient for energy storage loss. Let be the equipment operating cost power at time t, and γ be the economic weighting coefficient of the equipment operating cost. For time step; 3) The formula for the power output constraint of wind and solar power is as follows: in, Contribute to the minimum technical requirements of wind power. Minimum power output for photovoltaic technology; introduction of upper limit for photovoltaic power output ; The energy storage operation constraint formula is as follows: ， in , Rated charge and discharge power for energy storage; Let t be the state of charge of the energy storage unit. The minimum state of charge allowed for the energy storage unit. This represents the minimum permissible state of charge for the energy storage unit. in , Furthermore, the SOC must satisfy temporal continuity; in This refers to the rated capacity of the energy storage. The power balance constraint formula is as follows: in, Consumes electrical energy for user-side equipment. Power purchased / sold to the power grid.

5. An optimized dispatching system based on wind-solar-storage combined power generation according to claim 4, characterized in that, The computing center solves the objective function using the Physics-informed DDPG model, specifically as follows: 1) Model Input Layer Construction Input state vector Integrating three types of features: Load characteristics ; Landscape features ; Energy storage characteristics ; in Let t be the standardized power of the user-side electrical load; The status marker of the load at time t; Let t be the standardized power generation capacity of wind power at time t; Let t be the standardized power generation of the photovoltaic system at time t; Let be the standardized wind speed at time t; Let be the standardized irradiance at time t; Let be the standardized value of the energy storage state of charge at time t; The standardized charging power for energy storage at time t; Let be the standardized discharge power of the stored energy at time t; The vector dimension is 11, and the time-series input covers 24 hours of data; 2) Actor-Critic Network and Physical Constraint Embedding Actor network is based on Output scheduling action: These correspond to the actual wind power output, actual photovoltaic power output, energy storage charging power, and energy storage discharging power at time t, respectively. The output layer mapping ensures the load constraint of the action. Fengguang's output is limited to , （ ）； Energy storage charging and discharging power is limited to , Furthermore, the SOC (State of Charge) is constrained to be between 20% and 80%. The Critic network associates the reward function with the objective function, as shown in the formula: in, , , , For the corresponding weighting coefficients, Mark the constraint satisfaction; 3) Model training and objective function solution An initial experience replay pool with a capacity of 10,000 is set up. Each time step the model runs, actions are calculated based on real-time data. ,award and the next state ; In each iteration, 32 samples are randomly drawn from the replay pool, and the target value of the next state is calculated using the target Critic network: Discount factor; The Critic network update aims to "minimize the current Q-value and..." The target network aims to minimize the mean squared error of the network by adjusting parameters using the Adam optimizer. The Actor network updates parameters with the goal of maximizing the Q-value of the Critic output, employing a policy gradient ascent method while using gradient pruning to prevent parameter oscillations. Every 10 iterations, the target network parameters are adjusted to... The model integrates the current network parameters proportionally; when the average cumulative reward fluctuation over 10 consecutive iterations is less than 1% and the action constraint satisfaction rate is greater than 99%, the model converges; at this point, the Actor network outputs the optimal scheduling action. Substituting these values ​​into the objective function yields the minimum total energy loss over a 24-hour period. ,and It is used directly as a control instruction.

6. An optimized dispatching system based on wind-solar-storage combined power generation according to claim 4, characterized in that, The central control module regulates the output ratio of the wind, solar, and energy storage execution modules as follows: First, the central control module receives the optimized scheduling strategy from the computing center; after receiving it, it performs a pre-verification of the real-time equipment status using the data acquisition module: if the fan gearbox temperature... Photovoltaic module temperature Based on the characteristics of flexible materials, such as high resistance to deformation and high tolerance to fluctuations, the target output can be finely adjusted using the following formula: in, This represents the actual corrected output of the wind turbine at time t; The rated operating output of the fan at time t; Let t be the equipment temperature of the fan; The actual corrected output of the photovoltaic system at time t; The rated power output of the photovoltaic system at time t; Let t be the temperature of the photovoltaic module at time t; Next, the fine-tuned target parameters are converted into control commands using a dynamic output allocation algorithm: Photovoltaic actuators, combined with real-time irradiance Adjusting the DC side voltage of the inverter The formula is in, Let t be the actual output voltage of the photovoltaic system at time t; This is the reference voltage for photovoltaics, i.e., the reference voltage under a rated irradiance of 1000W / m². This refers to the rated power of the photovoltaic system. This is the ratio of actual irradiance to reference irradiance. Wind power execution unit, based on real-time wind speed Adjusting the angle of the flexible pitch blades The formula is in This is the reference angle at rated wind speed; Energy storage execution unit, based on real-time When switching converter topologies, the charging and discharging power must meet the following requirements: in, The actual corrected charging power for energy storage at time t; The maximum allowable charging power for energy storage; This represents the maximum state of charge allowed for energy storage. The actual corrected discharge power of the stored energy at time t; This represents the maximum permissible discharge power of the energy storage. During execution, each unit transmits actual data back at a frequency of seconds: photovoltaic backhaul. Wind power backhaul Energy storage backhaul and Meanwhile, the central control module calculates the deviation: Relative deviation in wind and solar power output: Absolute deviation of energy storage power: ; in, The relative deviation of wind output at time t; The actual contribution to the scenery at time t; The output is adjusted to reflect the scenery at time t; Let be the absolute deviation of the energy storage power at time t; The actual power of the stored energy at time t; Let t be the corrected power of the stored energy. If the deviation exceeds the threshold, a correction is triggered. After the correction command is issued, the execution unit adjusts and continues to send back data every second. The central control module repeats the deviation comparison and correction, forming a closed loop of "receiving strategy - transferring command - second feedback - correcting deviation - re-execution", ensuring that the output matches the loss reduction target, while using flexible materials to adapt to the environment and avoid overload.

7. An optimized dispatching system based on wind-solar-storage combined power generation according to claim 6, characterized in that, The wind-solar-storage execution module, as the core of system instruction execution, has the following workflow: The module receives the optimized scheduling instructions from the central control module and, in conjunction with its own status monitoring components, then the three main units—the photovoltaic execution unit, the wind power execution unit, and the energy storage execution unit—synchronously adjust their output: the wind power execution unit adjusts its output based on real-time wind speed. The output is controlled by adjusting the angle of the flexible blades using a pitch motor. The photovoltaic actuator adjusts the DC-side voltage through an intelligent inverter, while utilizing the flexible array panel's bendability to avoid localized shading, ensuring optimal power output. The energy storage execution unit reduces charging and discharging losses by switching the converter topology based on real-time SOC and the tolerance of the flexible battery pack. Subsequently, the wind, solar, and energy storage execution module collects the operating data of each unit at a frequency of seconds and then sends it back to the data acquisition module, providing data basis for the central control module to calculate the deviation. If an equipment abnormality is detected, the wind, solar, and energy storage execution module will automatically trigger a local drop and prioritize sending back the abnormal signal, forming a secondary protection with the central control module to ensure the stable operation of the system and indirectly support the goal of minimizing total energy loss.

8. An optimized scheduling method based on wind-solar-storage combined power generation, characterized in that, Includes the following steps: Step (1): The data acquisition module collects three types of core data in real time, including user-side power load data, wind and solar power output data of distributed generation units, and state of charge and charging / discharging power data of energy storage units, and processes the data. Step (2): The computing center uses the data constructed in step (1) to solve the problem through the Physics-informed DDPG model with the objective function of "minimizing total energy loss within the scheduling cycle". The Physics-informed DDPG model relies on the deep deterministic policy gradient framework and learns the temporal correlation through the Actor-Critic network to output the predicted values ​​of wind and solar power output fluctuations and load changes. The prediction embeds physical information to conform to actual constraints, and then directly generates the scheduling strategy. The reward function realizes multi-objective optimization and finally generates the optimal output parameters of distributed generation and energy storage units, realizing the dynamic prediction and end-to-end optimization of wind and solar power output and load demand. Step (3): The central control module adjusts the output ratio of the wind, solar and energy storage execution modules in real time through the dynamic output allocation algorithm based on the optimization scheduling strategy of the computing center. It also matches the environment with the physical properties of flexible materials to avoid equipment overload. During execution, the actual output and equipment status are transmitted back at a frequency of seconds. After comparing the deviation, the central control module triggers the instruction correction or protection mechanism to form a closed loop of "optimization-execution-feedback-re-optimization".

9. The optimized scheduling method based on wind-solar-storage combined power generation according to claim 8, characterized in that, The wind-solar-storage execution module is used to accurately respond to the optimized scheduling instructions of the central control module, and the wind-solar-storage execution module transmits the actual operating data of the equipment back to the central control module, forming a closed-loop control link to ensure that the system operates stably and optimally with the goal of "minimizing the total energy loss within the scheduling cycle".

10. A computer-readable storage medium having stored thereon computer program instructions executable by a processor, wherein when the processor executes the computer program instructions, it is able to implement the steps of the method as described in claim 8 or 9.

Images