Transform and differential evolution-based load output prediction regulation and control method

Abstract

The invention discloses a load output prediction regulation and control method based on Transform and differential evolution, and the method comprises the steps: collecting real-time data through edge equipment, and predicting the load of a power distribution network and the output of distributed energy; the predicted power distribution network load and distributed energy output are sent to a differential evolution module arranged at the cloud end, and an optimal control instruction sequence is solved; a differential evolution module arranged at the cloud sends the optimal control instruction sequence to an edge controller, and meanwhile, the edge device monitors the local state in real time and uploads operation data and execution feedback to the cloud to obtain a cloud and edge cooperative control instruction; cloud and edge cooperative control instructions are sent to the distributed resource execution module, specific actions are executed according to the instructions issued by the cloud, and local measurement data are monitored, controlled and corrected. Compared with the prior art, the method at least has the beneficial effects that the electricity purchasing power is obviously reduced, the regulation and control smoothness is improved, the load tracking capability is enhanced, the distributed output utilization rate is improved, and the real-time performance and the robustness are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a load output prediction and control method based on Transformer and differential evolution. Background Art

[0002] With the influx of large-scale distributed power sources (such as photovoltaic and wind power), energy storage systems, and flexible loads, the load characteristics of distribution networks have become highly volatile, random, and exhibit significant peak-to-valley variations, severely impacting their control capabilities. Traditional load and distributed output forecasting methods primarily rely on statistical models (such as ARIMA) and time series forecasting methods based on recurrent neural network (RNN) frameworks (such as LSTM and GRU). Because ARIMA relies solely on historical time series data, it struggles to handle complex multi-source features and long-term dependencies, resulting in insufficient forecasting accuracy. While LSTM and GRU models have the ability to model time series features, they still suffer from inadequate fitting and low sensitivity to external features when dealing with distributed output and load forecasting involving sudden changes, periodicity, and multi-timescale coupling. Furthermore, due to the long-range dependencies inherent in RNNs, their predictions of load peaks and distributed resource output extremes are often distorted, impacting the effectiveness of subsequent control decisions.

[0003] On the other hand, traditional distribution network optimization and control methods are represented by MPC (model predictive control) and PSO (particle swarm optimization). MPC relies on accurate system modeling and prediction, and it is difficult to adapt to the dynamic complexity of highly penetrated distributed resources. Heuristic optimization methods such as PSO have problems such as being prone to falling into local optimality and slow convergence, making it difficult to converge efficiently in the scenario of real-time control of distribution networks. In addition, existing optimization methods and prediction models generally adopt a serial architecture, and the prediction results do not participate in the optimization inner loop, resulting in prediction-optimization decoupling and a lack of system adaptability. Therefore, there is an urgent need for a load output prediction and control method and system based on Transformer and differential evolution to solve existing technical problems. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a load output forecasting and control method based on Transformer and differential evolution to overcome the above problems or at least partially solve the above problems.

[0005] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention discloses a load output forecasting and control method based on Transformer and differential evolution, comprising:

[0007] S100. The edge device collects real-time data and sends the real-time data to the pre-built load and distributed output prediction model in the cloud to predict the distribution network load and distributed energy output;

[0008] S200. Send the predicted distribution network load and distributed energy output to the differential evolution module deployed in the cloud, and solve the optimal control instruction sequence through the differential evolution module;

[0009] S300. The differential evolution module deployed in the cloud sends the optimal control instruction sequence to the edge controller. At the same time, the edge device monitors the local status in real time, uploads the operation data and execution feedback to the cloud, and obtains the cloud and edge collaborative control instructions;

[0010] S400. Send cloud and edge collaborative control instructions to the distributed resource execution module, which performs specific actions according to the instructions issued by the cloud, and monitors and controls the local measurement data through the edge computing device.

[0011] Furthermore, in S100, a load and distributed output prediction model is constructed, specifically including the design of a multi-objective optimization function, the design of constraint conditions, and the design of a penalty function.

[0012] Furthermore, the multi-objective optimization function satisfies the requirements of minimizing the total cost of purchasing electricity from the main grid while ensuring power supply demand, while minimizing the cost of using energy storage equipment and minimizing the regulation cost. The specific expression of the multi-objective optimization function is:

[0013]

[0014] Among them, P grid,t Indicates the power purchased from the main grid at time t, ΔP dg,t represents the change in the output of the distributed generation during period t, SOC_{dev,t} represents the degree to which the energy storage SOC deviates from the target range, and w1, w2, and w3 are the weight coefficients of each target, which are used to balance the relative importance of economic cost and operation smoothness.

[0015] Furthermore, the constraints include power balance constraint, energy storage SOC constraint, voltage constraint, output power constraint, control step constraint, and control budget constraint. The power balance constraint ensures that the power revenue and expenditure of the distribution network are balanced in each time period. The specific expression is:

[0016] P dg，t +P grid,t =P load,t +P loss,t

[0017] P dg,t Indicates the output power of distributed power generation, Pgrid,t Indicates the power purchased by the main grid, P load,t Indicates the power purchased by the main grid, P loss,t Indicates network loss power;

[0018] The energy storage SOC constraint ensures that the state of the energy storage device must be limited to a preset range. The specific expression is:

[0019]

[0020] SOC min Indicates the lowest state of charge of energy storage, SOC max Indicates the maximum state of charge of the energy storage;

[0021] The voltage constraint ensures that the voltage of each node in the distribution network must be kept within the standard range. The specific expression is:

[0022]

[0023] U min Indicates the minimum voltage of each node in the distribution network, U max Indicates the maximum voltage of each node in the distribution network.

[0024] Furthermore, the output power constraint ensures that the output power of distributed power sources and adjustable loads does not exceed their rated capacity. The specific expression is:

[0025]

[0026] P dg,max Indicates the maximum output of distributed power generation, Indicates the maximum amount of controllable load adjustment;

[0027] The control step constraint ensures that the change range of two consecutive control instructions is within the preset range. The specific expression is:

[0028] |Δu t |=|u t -u t-1 |≤Δu max

[0029] U t Indicates the control instruction for the tth period, U t-1 Indicates the control instruction for the t-1 period, △U max The maximum value of the control instruction, which is a set of decision variables for controlling the energy storage charging and discharging power and the controllable load level;

[0030] The control budget constraint is used to control demand response costs and set an upper limit on the total control cost budget, including that the cumulative cost of starting and stopping controllable loads or calling backup resources within a control cycle does not exceed a predetermined threshold.

[0031] Furthermore, the specific design method of the penalty function includes: using the penalty function method to convert the violation degree of each constraint into a penalty term in the objective function; when the solution violates a constraint, the penalty term will significantly increase the value of the objective function, thereby guiding the evolutionary algorithm away from infeasible solutions; for the voltage constraint U_i(t), if a solution causes the voltage U_i to be lower than the lower limit at a certain moment, a penalty λ is generated. U [U min -U i (t)] 2 ; If it is higher than the upper limit, penalty λ U [U i (t)-U max ] 2 ;λ U is the voltage constraint penalty coefficient. The constraint penalty coefficient should be large enough to ensure that the solution with voltage violation is not selected.

[0032] Furthermore, in S200, the optimal control instruction sequence is solved by the differential evolution module, and the specific method includes:

[0033] Encode the model solution into a vector containing all decision variables for all control periods; the vector is represented as X = [u_1, u_2, ..., u_T], where u_t is the control instruction for period t, and the vector dimension is equal to the number of control periods multiplied by the number of decision variables;

[0034] Perform mutation and crossover operations on the solution of the model. For each solution x i , randomly select three different solutions x from the population a ,x b ,x c , generating the difference d = x b -x c , and then amplify the difference with the scaling factor F to obtain the mutation vector v i =x a +F·d, the v i is in x a Based on the new solution obtained by moving along the random difference direction; the mutation vector vi and the solution xi are recombined according to a certain crossover probability to generate the experimental solution ui;

[0035] Determine the update strategy of the solution in the population, for each solution x i , substitute the corresponding test solution ui into the objective function to calculate the fitness; then execute the survivor selection strategy, if u i The objective function value is better than xi , then in the next generation population, u i Replace x i ; Otherwise x i reserve;

[0036] The solution update in the population is iterated, and the population is continuously evolved in an intergenerational iterative manner. When the number of population iterations reaches the preset maximum number of iterations, or the improvement of the optimal solution for several consecutive generations is lower than the threshold, the optimal individual in the current population is regarded as the global approximate optimal solution.

[0037] Furthermore, in S400, the distributed resource execution module is located at the distribution network site and includes at least an energy storage converter controller, a photovoltaic inverter controller, and an adjustable load controller device; the distributed resource execution module performs specific actions according to instructions issued by the cloud, and the specific work includes at least energy storage charging, orderly charging of electric vehicles, and controllable load reduction; and monitors and preliminarily controls and corrects the locally measured voltage data, current data, and SOC data through an edge computing device.

[0038] In a second aspect, an embodiment of the present invention discloses an electronic device, including:

[0039] one or more processors;

[0040] a memory for storing one or more programs;

[0041] When the one or more programs are executed by the one or more processors, the one or more processors implement the load output forecasting and control method.

[0042] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0043] The present invention discloses a load output prediction and control method based on Transformer and differential evolution, comprising: edge devices collecting real-time data and sending the real-time data to a load and distributed output prediction model pre-built in the cloud to predict the load of the distribution network and the output of distributed energy; sending the predicted load of the distribution network and the output of distributed energy to a differential evolution module deployed in the cloud, and solving the optimal control instruction sequence through the differential evolution module; the differential evolution module deployed in the cloud sends the optimal control instruction sequence to the edge controller, while the edge device monitors the local status in real time, uploads the operation data and execution feedback to the cloud, and obtains cloud and edge collaborative control instructions; the cloud and edge collaborative control instructions are sent to the distributed resource execution module, and the distributed resource execution module executes specific actions according to the instructions issued by the cloud, and monitors and controls the local measurement data through the edge computing device. Compared with the existing technology, the present invention has at least the following beneficial effects: significantly reducing the purchased power, improving the smoothness of the regulation, enhancing the load tracking capability, improving the utilization rate of the distributed output, and enhancing the real-time and robustness.

[0044] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0046] Figure 1 This is a flow chart of a load output forecasting and control method based on Transformer and differential evolution in Example 1 of the present invention;

[0047] Figure 2 This is a comparison diagram of the power purchasing dispatch curves of the method disclosed in this embodiment and the traditional method in Example 1 of the present invention;

[0048] Figure 3 This is a structural diagram of an electronic device in Example 2 of the present invention. DETAILED DESCRIPTION

[0049] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0050] In order to solve the problems existing in the prior art, an embodiment of the present invention provides a load output prediction and control method based on Transformer and differential evolution.

[0051] Example 1

[0052] The present invention discloses a load output prediction and control method based on Transformer and differential evolution, such as Figure 1 ,include:

[0053] S100. The edge device collects real-time data and sends the real-time data to the pre-built load and distributed output prediction model in the cloud to predict the distribution network load and distributed energy output;

[0054] In S100 of this embodiment, a load and distributed output forecasting model is constructed, specifically including the design of a multi-objective optimization function, the design of constraints, and the design of a penalty function. The multi-objective optimization function satisfies the requirement to minimize the total cost of purchasing electricity from the main grid while minimizing the cost of using energy storage equipment and minimizing the cost of regulation. The specific expression of the multi-objective optimization function is:

[0055]

[0056] Among them, P grid,t Indicates the power purchased from the main grid at time t, ΔP dg,t represents the change in the output of the distributed generation during period t, SOC_{dev,t} represents the degree to which the energy storage SOC deviates from the target range, and w1, w2, and w3 are the weight coefficients of each target, which are used to balance the relative importance of economic cost and operation smoothness.

[0057] Specifically, in order to balance economic benefits and operational stability, the present invention constructs a multi-objective optimization function that includes power purchase cost, energy storage cost, smoothness index, and regulation cost. The sub-objective functions are defined as follows:

[0058] The power purchase cost objective is to minimize the total cost of purchasing power from the main grid. This can be expressed as the product of the power purchased from the main grid and the price. For simplicity, the square of the power purchased can be used to represent this value (higher power purchases mean higher costs).

[0059] Energy storage cost objective: Minimize the cost of using energy storage equipment, including the reduction in battery life due to charge and discharge cycles. This can be achieved indirectly by maintaining the state of charge (SOC) of the energy storage within a reasonable range, minimizing the sum of the squares of SOC deviations from neutral to reduce losses caused by deep charge and discharge.

[0060] Smoothness objective: Minimize fluctuations in controlled power and improve the smoothness of the output curve. By reducing the sum of the squares of the distributed output variations (ΔP) between adjacent time periods, smooth transitions in control actions are achieved, reducing stress caused by frequent starts and stops of equipment.

[0061] Control cost objective: Minimize the cost of executing control actions, including the number and magnitude of operations such as on-load tap-changing transformer tap changes and controllable load adjustments. This can be equivalent to setting a penalty for changes in control commands, thereby suppressing overly frequent or drastic control changes.

[0062] In this embodiment, the constraints include power balance constraint, energy storage SOC constraint, voltage constraint, output power constraint, control step constraint, and control budget constraint. The power balance constraint ensures that the power revenue and expenditure of the distribution network are balanced in each time period. The specific expression is:

[0063] P dg，t +P grid,t =P load,t +P loss,t

[0064] P dg,t Indicates the output power of distributed power generation, P grid,t Indicates the power purchased by the main grid, P load,t Indicates the power purchased by the main grid, P loss,t Indicates network loss power;

[0065] The energy storage SOC constraint ensures that the state of the energy storage device must be limited to a preset range. The specific expression is:

[0066]

[0067] SOC min Indicates the lowest state of charge of energy storage, SOC max Indicates the maximum state of charge of the energy storage;

[0068] The voltage constraint ensures that the voltage of each node in the distribution network must be kept within the standard range. The specific expression is:

[0069]

[0070] U min Indicates the minimum voltage of each node in the distribution network, U max Indicates the maximum voltage of each node in the distribution network.

[0071] The output power constraint ensures that the output power of distributed power sources and adjustable loads does not exceed their rated capacity. The specific expression is:

[0072]

[0073] P dg,max Indicates the maximum output of distributed power generation, Indicates the maximum amount of controllable load adjustment;

[0074] The control step constraint ensures that the change range of two consecutive control instructions is within the preset range. The specific expression is:

[0075] |Δu t |=|u t -u t-1 |≤Δu max

[0076] U t Indicates the control instruction for the tth period, U t-1 Indicates the control instruction for the t-1 period, △U max The maximum value of the control instruction, which is a set of decision variables for controlling the energy storage charging and discharging power and the controllable load level;

[0077] The control budget constraint is used to control demand response costs and set an upper limit on the total control cost budget, including that the cumulative cost of starting and stopping controllable loads or calling backup resources within a control cycle does not exceed a predetermined threshold.

[0078] The above constraints ensure the engineering feasibility and safety of the optimization results. For example, power balance constraints ensure that energy does not appear or disappear without reason; SOC and power upper and lower limits ensure that energy storage and power supply equipment operate within physical limits; and voltage and step size constraints ensure power quality and equipment life. Any solution that violates these constraints is infeasible and must be eliminated or penalized during the optimization process.

[0079] In this embodiment, the specific method for designing the penalty function includes: using the penalty function method to convert the violation degree of each constraint into a penalty term in the objective function; when the solution violates a constraint, the penalty term will significantly increase the value of the objective function, thereby guiding the evolutionary algorithm away from infeasible solutions; for the voltage constraint U_i(t), if a solution causes the voltage U_i to be lower than the lower limit at a certain moment, a penalty λ is generated. U [U min -U i (t)] 2 ; If it is higher than the upper limit, penalty λ U [U i (t)-U max ] 2 ;λ U is the voltage constraint penalty coefficient. The constraint penalty coefficient should be large enough to ensure that the solution with voltage violation is not selected.

[0080] Specifically, in order to deal with constraints, incorporating them into the objective function is a common strategy. The present invention adopts a penalty function method to convert the degree of violation of each constraint into a penalty term in the objective function. When a solution violates a constraint, the penalty term will significantly increase the value of the objective function, thereby guiding the evolutionary algorithm away from infeasible solutions. By reasonably setting the penalty coefficient λ of each constraint, the objective function of the solution is effectively solved. k ,, the constraints are internalized into the optimization goal, so that the final comprehensive objective function In this way, the evolutionary algorithm will automatically tend to satisfy all constraints during the search process, thereby improving the feasibility and stability of the optimization solution.

[0081] In this example, load forecasting is performed using a Transformer model to perform short-term load forecasting on the distribution network. The model takes historical load data and related influencing factors (such as weather and date) as input, extracts long-term dependency features through an encoder-decoder architecture, and outputs a load forecast curve for the next 24 hours. This module is deployed in the cloud, leveraging its powerful computing power to train deep models and achieve high-precision forecasts.

[0082] In this embodiment, distributed power output is predicted using a Transformer architecture similar to load forecasting to predict the power of distributed resources such as distributed photovoltaics, electric vehicle energy storage, and controllable loads, resulting in a 24-hour output forecast for each distributed energy source. The prediction module integrates multi-source data in the cloud, improving the predictability of power generation and consumption behavior across various distributed resources.

[0083] S200. The predicted distribution network load and distributed energy output are sent to a differential evolution module deployed in the cloud. The module then solves for the optimal control instruction sequence. Differential evolution (DE) is a highly efficient swarm intelligence evolutionary algorithm suitable for global optimization in continuous space. This patent uses the DE algorithm to solve the optimal control strategy for the aforementioned optimization model. The core concept of the DE algorithm is to iteratively optimize candidate solutions through a genetic operation called "mutation-crossover-selection."

[0084] In S200 of this embodiment, the optimal control instruction sequence is solved by the differential evolution module. The specific method includes:

[0085] Encode the model solution into a vector containing all decision variables for all control periods; the vector is represented as X = [u_1, u_2, ..., u_T], where u_t is the control instruction for period t, and the vector dimension is equal to the number of control periods multiplied by the number of decision variables;

[0086] Perform mutation and crossover operations on the solution of the model. For each solution xi , randomly select three different solutions x from the population a ,x b ,x c , generating the difference d = x b -x c , and then amplify the difference with the scaling factor F to obtain the mutation vector v i =x a +F·d, the v i is in x a Based on the new solution obtained by moving along the random difference direction; the mutation vector vi and the solution xi are recombined according to a certain crossover probability to generate the experimental solution ui;

[0087] Determine the update strategy of the solution in the population, for each solution x i , substitute the corresponding test solution ui into the objective function to calculate the fitness; then execute the survivor selection strategy, if u i The objective function value is better than x i , then in the next generation population, u i Replace x i ; Otherwise x i reserve;

[0088] The solution update in the population is iterated, and the population is continuously evolved in an intergenerational iterative manner. When the number of population iterations reaches the preset maximum number of iterations, or the improvement of the optimal solution for several consecutive generations is lower than the threshold, the optimal individual in the current population is regarded as the global approximate optimal solution.

[0089] After obtaining the prediction information provided by the cloud-based Transformer, the differential evolution optimization module executes the aforementioned algorithmic process to determine the optimal control action for the current period. After each iteration, the edge execution unit feeds back the actual operation to the cloud to update the optimization calculations in the next iteration, thus achieving continuous optimization. The entire DE algorithmic process works in conjunction with the prediction and control stages to ensure optimal (or suboptimal) implementation of multiple objective functions while satisfying constraints.

[0090] S300. The differential evolution module deployed in the cloud sends the optimal control instruction sequence to the edge controller. At the same time, the edge device monitors the local status in real time, uploads the operation data and execution feedback to the cloud, and obtains the cloud and edge collaborative control instructions;

[0091] Specifically, the cloud sends optimized control instructions to the edge controller. Simultaneously, the edge monitors local status in real time and uploads key operational data and execution feedback to the cloud. This cloud-edge communication and collaboration mechanism ensures that global optimization decisions made on the cloud are promptly implemented on-site. Rapid disturbances at the edge are also fed back to the cloud to adjust subsequent strategies, forming a closed loop. Cloud-edge collaboration can be achieved through the Industrial Internet or power distribution IoT platforms, ensuring low-latency and highly reliable data transmission.

[0092] S400. Send cloud and edge collaborative control instructions to the distributed resource execution module, which performs specific actions according to the instructions issued by the cloud, and monitors and controls the local measurement data through the edge computing device.

[0093] In S400 of this embodiment, the distributed resource execution module is located at the distribution network site and includes at least an energy storage converter controller, a photovoltaic inverter controller, and an adjustable load controller device; the distributed resource execution module performs specific actions according to instructions issued by the cloud, and the specific work includes at least energy storage charging, orderly charging of electric vehicles, and controllable load reduction; and monitors and preliminarily controls and corrects locally measured voltage data, current data, and SOC data through an edge computing device.

[0094] Through the division of labor and cooperation of the above modules, a combination of centralized prediction, global optimization and local rapid response is achieved. Its control strategy is as follows: the edge side continuously collects real-time data and reports it to the cloud; the cloud prediction module periodically generates rolling predictions; the cloud optimization module solves the optimal control plan based on the latest prediction; the edge controls the execution plan and makes limited adjustments at a higher rate (such as fine-tuning the charging power based on the measured SOC), and at the same time feeds the execution results back to the cloud for reference in the next calculation. This strategy makes full use of the global vision and computing power of the cloud, as well as the real-time and on-site perception capabilities of the edge side, to achieve active optimization and control of distributed resources in the distribution network while ensuring safety constraints.

[0095] To verify the effectiveness of the proposed method, a simulation platform based on an IEEE 33-node distribution system was constructed. It was assumed that the distribution network was connected to a high penetration rate of distributed photovoltaic power generation and a certain scale of energy storage devices, with a diurnal load variation. Simulation Platform: Using tools such as Matlab / Simulink, a 33-node system model was constructed, connecting a total distributed photovoltaic capacity of approximately X MW (with a penetration rate of approximately Y%) and an energy storage capacity of Z MWh. Typical load and photovoltaic output curve data for a single day (24 hours) was generated. Comparison: The proposed "Transformer + DE joint optimization framework" was compared with the traditional "LSTM + PSO prediction and optimization framework." In the traditional approach, LSTM was used to predict load and photovoltaic power generation, followed by a particle swarm optimization algorithm for scheduling optimization; in this patented approach, Transformer prediction and DE optimization were used. Both approaches share the same control objectives to ensure comparability.

[0096] Experimental results show that the proposed solution is significantly superior to traditional solutions in terms of reducing purchased power, improving control smoothness, and tracking load fluctuations. The specific comparative analysis is as follows:

[0097] 1. Significant reduction in power purchase: In most periods, the power purchased from the main grid by this patent solution is lower than that of traditional solutions. Figure 2 As shown, the dashed line (traditional solution) is generally higher than the solid line (this solution) when comparing the purchased power dispatch curves of the two schemes. This demonstrates that this scheme makes better use of local resources such as distributed photovoltaics and energy storage, reducing reliance on the external grid. Statistics show that the average daily purchased power of this scheme is approximately 12.3% lower than that of the traditional scheme. This means that both the amount of purchased power and the cost are reduced, achieving more economical operation.

[0098] 2. Improved Control Smoothness: This solution provides a smoother power purchase curve and significantly reduces volatility. Under traditional solutions, power purchases fluctuate dramatically during the morning and evening peak hours, with sharp rises and falls. However, this solution optimizes energy storage scheduling and pre-charges and discharges energy before peak hours, resulting in a smoother rise in the power purchase curve and lower peaks. Regarding volatility, the standard deviation of power purchase changes between adjacent time periods is calculated, and this solution is 28.7% lower than the traditional solution. This demonstrates that the frequency and magnitude of control actions are reduced, reducing equipment fatigue and ride-through losses.

[0099] 3. Enhanced load tracking capability: During periods of drastic load fluctuations (such as changes in sunlight from the 5th to the 18th hour and midday load fluctuations), this solution can better match actual load demand. Thanks to Transformer's more accurate load and output forecasts, combined with DE optimization's robust handling of prediction errors, this solution promptly increases output when the load rises and appropriately reduces power purchases when the load drops, maintaining a good balance between supply and demand. In contrast, due to limitations in prediction accuracy and response speed, traditional solutions experience load tracking lags, manifested as insufficient power purchases when the load is high (requiring urgent compensation) and excess power purchases when the load is low (causing overflow and waste). Measured by quantitative indicators, the supply and demand deviation of the traditional solution fluctuates significantly, while this solution reduces this deviation, making the supply and demand matching more accurate.

[0100] 4. Improved Distributed Power Utilization: This solution maximizes the utilization of clean energy sources, including local distributed photovoltaics. During peak photovoltaic power periods, traditional solutions, due to inadequate forecasting, may fail to charge and store energy in a timely manner, resulting in some photovoltaic power curtailment. This solution effectively utilizes excess photovoltaic power by accurately predicting photovoltaic redundancy and preemptively freeing up storage capacity. Statistics show that this solution utilizes approximately 15.6% more distributed power than traditional solutions, significantly reducing reliance on purchased electricity during peak periods. This not only increases renewable energy utilization but also further reduces system operating costs.

[0101] 5. Enhanced real-time performance and robustness: Thanks to cloud-edge collaborative control, this solution exhibits faster response to sudden load changes. Rapid compensation by the edge controller stabilizes key indicators such as voltage, reducing the average control command response time by 22.4% compared to traditional centralized control. Furthermore, thanks to the closed-loop prediction-optimization system, this solution is less sensitive to prediction errors. Deviations can be corrected through optimization in the next cycle, resulting in more stable system operation.

[0102] Example 2

[0103] Based on the same inventive concept, an embodiment of the present disclosure further provides an electronic device. Figure 3 FIG. 1 is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Figure 3 As shown, an embodiment of the present disclosure provides an electronic device comprising: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement any of the optimization methods described in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory and are configured to implement information exchange between the processor and the memory.

[0104] Among them, the processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically such as SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus), etc.

[0105] In some embodiments, the processor 101 , the memory 102 , and the I / O interface 103 are connected to each other via a bus 104 , and further connected to other components of the computing device.

[0106] In some embodiments, the one or more processors 101 include a field programmable gate array.

[0107] According to an embodiment of the present disclosure, a computer-readable medium is further provided, wherein a computer program is stored on the computer-readable medium, wherein when the program is executed by a processor, the steps of any optimization method in the above-mentioned embodiment are implemented.

[0108] It should be understood that the specific order or hierarchy of steps in the disclosed processes is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0109] In the foregoing detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that embodiments of the claimed subject matter require more features than are expressly recited in each claim. On the contrary, as reflected in the appended claims, the invention comprises less than all the features of any individual disclosed embodiment. The appended claims are therefore hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0110] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein may be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described around their functions. Whether such functions are implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. A skilled person may implement the described functions in an adaptable manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of this disclosure.

[0111] The steps of the methods or algorithms described in conjunction with the embodiments herein may be directly embodied as hardware, software modules executed by a processor, or a combination thereof. The software module may be located in a RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and storage medium may also be present in a user terminal as discrete components.

[0112] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or external to the processor. In the latter case, it is communicatively coupled to the processor via various means, which are well known in the art.

[0113] The foregoing description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purposes of describing the above embodiments, but one of ordinary skill in the art will recognize that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to encompass all such changes, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent the term "comprising" is used in the specification or claims, the term is intended to be encompassed in a manner similar to the term "including," as explained in terms of "including," used as a transitional word in the claims. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or."

Claims

1. A load output forecasting and control method based on Transformer and differential evolution, characterized in that: include: S100. The edge device collects real-time data and sends the real-time data to the pre-built load and distributed output prediction model in the cloud to predict the distribution network load and distributed energy output; S200. Send the predicted distribution network load and distributed energy output to the differential evolution module deployed in the cloud, and solve the optimal control instruction sequence through the differential evolution module; S300. The differential evolution module deployed in the cloud sends the optimal control instruction sequence to the edge controller. At the same time, the edge device monitors the local status in real time, uploads the operation data and execution feedback to the cloud, and obtains the cloud and edge collaborative control instructions; S400. Send cloud and edge collaborative control instructions to the distributed resource execution module, which performs specific actions according to the instructions issued by the cloud, and monitors and controls the local measurement data through the edge computing device.

2. The load output forecasting and control method based on Transformer and differential evolution according to claim 1, characterized in that: In S100, a load and distributed output prediction model is constructed, specifically including the design of a multi-objective optimization function, the design of constraints, and the design of a penalty function.

3. The load output forecasting and control method based on Transformer and differential evolution according to claim 2, characterized in that: The multi-objective optimization function satisfies the following requirements: minimizing the total cost of purchasing electricity from the main grid while ensuring power supply demand, minimizing the cost of using energy storage equipment, and minimizing the control cost. The specific expression of the multi-objective optimization function is: Among them, P grid,t Indicates the power purchased from the main grid at time t, ΔP dg,t represents the change in the output of the distributed generation during period t, SOC_{dev,t} represents the degree to which the energy storage SOC deviates from the target range, and w1, w2, and w3 are the weight coefficients of each target, which are used to balance the relative importance of economic cost and operation smoothness.

4. The load output forecasting and control method based on Transformer and differential evolution according to claim 2, characterized in that: The constraints include power balance constraint, energy storage SOC constraint, voltage constraint, output power constraint, control step constraint, and control budget constraint. The power balance constraint ensures that the power revenue and expenditure of the distribution network are balanced in each time period. The specific expression is: P dg,t +P grid,t =P load,t +P loss,t P dg,t Indicates the output power of distributed power generation, P grid,t Indicates the power purchased by the main grid, P load,t Indicates the power purchased by the main grid, P loss,t Indicates network loss power; The energy storage SOC constraint ensures that the state of the energy storage device must be limited to a preset range. The specific expression is: SOC min Indicates the minimum state of charge of energy storage, SOC max Indicates the maximum state of charge of the energy storage; The voltage constraint ensures that the voltage of each node in the distribution network must be kept within the standard range. The specific expression is: U min Indicates the minimum voltage of each node in the distribution network, U max Indicates the maximum voltage of each node in the distribution network.

5. The load output forecasting and control method based on Transformer and differential evolution according to claim 4, characterized in that: The output power constraint ensures that the output power of distributed power sources and adjustable loads does not exceed their rated capacity. The specific expression is: P dg,max Indicates the maximum output of distributed power generation, Indicates the maximum amount of controllable load adjustment; The control step constraint ensures that the change range of two consecutive control instructions is within the preset range. The specific expression is: |Δu t |=|u t -u t-1 |≤Δu max U t Indicates the control instruction for the tth period, U t-1 Indicates the control instruction for the t-1 period, △U max The maximum value of the control instruction, which is a set of decision variables for controlling the energy storage charging and discharging power and the controllable load level; The control budget constraint is used to control demand response costs and set an upper limit on the total control cost budget, including that the cumulative cost of starting and stopping controllable loads or calling backup resources within a control cycle does not exceed a predetermined threshold.

6. The load output forecasting and control method based on Transformer and differential evolution according to claim 2, characterized in that: The specific design method of the penalty function includes: using the penalty function method to convert the violation degree of each constraint into a penalty term in the objective function; when the solution violates a constraint, the penalty term will significantly increase the value of the objective function, thereby guiding the evolutionary algorithm away from infeasible solutions; for the voltage constraint U_i(t), if a solution causes the voltage U_i to be lower than the lower limit at a certain moment, a penalty λ is generated. U [U min -U i (t)] 2 ; If it is higher than the upper limit, penalty λ U [U i (t)-U max ] 2 ;λ U is the voltage constraint penalty coefficient. The constraint penalty coefficient should be large enough to ensure that the solution with voltage violation is not selected.

7. The load output forecasting and control method based on Transformer and differential evolution according to claim 1, characterized in that: In S200, the optimal control instruction sequence is solved by the differential evolution module. The specific method includes: Encode the model solution into a vector containing all decision variables for all control periods; the vector is represented as X = [u_1, u_2, ..., u_T], where u_t is the control instruction for period t, and the vector dimension is equal to the number of control periods multiplied by the number of decision variables; Perform mutation and crossover operations on the solution of the model. For each solution x i , randomly select three different solutions x from the population a ,x b ,x c , generate the difference d = x b -x c , and then amplify the difference with the scaling factor F to obtain the mutation vector v i =x a +F·d, the v i is in x a Based on the new solution obtained by moving along the random difference direction; the mutation vector vi and the solution xi are recombined according to a certain crossover probability to generate the experimental solution ui; Determine the update strategy of the solution in the population, for each solution x i , substitute the corresponding test solution ui into the objective function to calculate the fitness; then execute the survivor selection strategy, if u i The objective function value is better than x i , then in the next generation population, u i Replace x i ; Otherwise x i reserve; The solution update in the population is iterated, and the population is continuously evolved in an intergenerational iterative manner. When the number of population iterations reaches the preset maximum number of iterations, or the improvement of the optimal solution for several consecutive generations is lower than the threshold, the optimal individual in the current population is regarded as the global approximate optimal solution.

8. The load output forecasting and control method based on Transformer and differential evolution according to claim 1, characterized in that: In S400, the distributed resource execution module is located at the distribution network site and includes at least an energy storage converter controller, a photovoltaic inverter controller, and an adjustable load controller device; the distributed resource execution module performs specific actions according to instructions issued by the cloud, and the specific work includes at least energy storage charging, orderly charging of electric vehicles, and controllable load reduction; and monitors and preliminarily controls and corrects the locally measured voltage data, current data, and SOC data through an edge computing device.

9. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the load output forecasting and control method described in any one of claims 1-8.

Images