Method and apparatus for optimizing power management system

Abstract

The present disclosure provides a method and apparatus for optimizing a power management system. The method includes inputting training data collected from a power grid into a machine learning model, which is a feedforward neural network, identifying a power grid power flow (PF) model and gradient information required for optimal power flow (OPF) calculation, obtaining an initial solution of a decision variable in the optimal power flow model, and proceeding with an iterative calculation process with the initial solution as the current solution of the optimal power flow model, and obtaining necessary control commands and display information through multiple iterative calculations of the OPF. The gradient information (Jacobian matrix) used in the OPF calculation is calculated without requiring multiple iterations, and the gradient information is used to improve the performance and speed of the OPF calculation, reduce the amount of calculation in the OPF calculation, and improve calculation efficiency. The machine learning model is used to model the PF model, and more types of input data can be adopted as training data for machine learning, improving the robustness and accuracy of the PF model.

Description

Incorporation by Reference

[0001] This application claims priority to China Patent Application No. 2021111151583.4, filed in China on September 29, 2021, the entire contents of which are incorporated herein by reference. [Technical field]

[0002] The present disclosure relates to the field of power system simulation and analysis technology, and more particularly to a method and apparatus for optimizing a power management system. [Background technology]

[0003] A small-scale power grid, also known as a micro-grid, is a concept that contrasts with traditional large-scale power grids (main power grids). It refers to a network in which multiple distributed power sources and their associated loads are configured in a fixed topology structure and are related to the main power grid via static switches. The development and expansion of micro-grids can fully promote the large-scale access of distributed power sources and renewable energy, and realize highly reliable supply of diverse energy forms to loads, and is an effective way to realize an active power distribution grid.

[0004] During the operation of a power system, under the action of the power source, current and power flow to the load through each element of the system and are distributed to various parts of the power grid, which is called the power flow (PF) distribution. The power flow calculation of a power system is a basic electrical calculation to study the steady operation status of the power system, and its task is to determine the operation status of the entire system based on the given operating conditions and network configuration, such as the voltage (amplitude value and phase angle) on each bus, the power distribution and power loss in the network. The main contents of the power flow calculation in a power system include the calculation of current and power distribution, the calculation of power loss, the calculation of voltage loss and node voltage, etc. The result of the power flow calculation of a power system is the basis for the stability calculation and fault analysis of the power system.

[0005] The OPF (Optimal Power Flow) problem involves selecting the best solution that satisfies certain index requirements from among all feasible power flow solutions in order to optimize system operation. Optimal power flow refers to an optimization process that adjusts the parameters of various control devices in the system from the perspective of optimized operation of the power system, and minimizes the objective function under constraints that satisfy the normal power balance of nodes and various safety indexes. Optimal power flow is an analysis method that simultaneously considers the safety and economics of the network, and is therefore widely used in the safe operation of power systems, economic scheduling, power grid planning, reliability analysis of complex power systems, and economic control of communication congestion.

[0006] Mathematical modeling of power systems is important for their monitoring, optimization or safe operation. Optimal power flow is a prerequisite for optimal operation of a power grid. Optimal power flow is usually used to calculate optimal control commands of devices or loads in a power system.

[0007] The first power flow calculation method in the prior art uses load power as input to calculate the power flow model. The input parameters of this solution are very limited, resulting in large errors in the power flow model and low accuracy. At the same time, this solution requires multiple iterations during calculation, which is slow.

[0008] The second power flow calculation method of the prior art performs optimal power flow calculation using a traditional power flow model, and similarly has the problem of large errors in the power flow model and low accuracy.

[0009] The third power flow calculation method of the prior art also uses a traditional power flow model to perform optimal power flow calculation, and finally converts the optimal power flow into machine learning to speed up the calculation. This solution requires multiple iterations when calculating the power flow, and multiple iterations are also required when calculating the optimal power flow, so the number of iterations for the final power flow calculation increases exponentially, the calculation speed is slow, and the calculation accuracy is low within a limited time.

[0010] For the above-mentioned first conventional power flow calculation method, reference may be made to the following documents 1 to 5. For the above-mentioned second conventional power flow calculation method, reference may be made to the following patent documents 6 to 9. For the above-mentioned third conventional power flow calculation method, reference may be made to the following documents 10 to 20.

[0011] [1] Artificial neural network based load flow solution of Saudi national grid, Wael Abdullah Alsulami;R Sreerama Kumar, 2017 Saudi Arabia Smart Grid (SASG); [2] Physics-Guided Deep Neural Networks for Power Flow Analysis, Xinyue Hu;Haoji Hu;Saurabh Verma;Zhi-Li Zhang, IEEE Transactions on Power Systems, Year: 2021; [3] Load flow estimation in electrical systems using artificial neural networks, Elena-Diana Aparaschivei;Ovidiu Ivanov;Mihai Gavrilas, 2012 International Conference and Exposition on Electrical and Power Engineering, Year: 2012; [4] Model-Driven Architecture of Extreme Learning Machine to Extract Power Flow Features, Qian Gao, Zhifang Yang, Juan Yu, Wei Dai, Xingyu Lei, Bo Tang, Kaigui Xie, Wenyuan Li, IEEE Transactions on Neural Networks and Learning Systems, Year: 2020; [5] Fast Calculation of Probabilistic Power Flow: A Model-Based Deep Learning Approach, 1999; Yan Yang;Zhifang Yang;Juan Yu;Baosen Zhang;Youqiang Zhang;Hongxin Yu. IEEE Transactions on Smart Grid. Year: 2020; [6]Patient ID US9507367B2; [7]Photo ID US9953117B2; [8]Patient ID US9705336B2; [9]Frequently Launched Equipment EP2 773 005 B1;

[10] Li G, Lu W, Bian J, Qin F and Wu J (2019) Probabilistic Optimal Power Flow Calculation Method Based on Adaptive Diffusion Kernel Density Estimation. Front. Energy Res. 7:1 doi: 10.3389 / fenrg.2019.00128;

[11] Optimal power flow study using conventional and neural networks methods, Labed Imen;Labed Djamel;Serghine Hassiba;Draidi Abdellah;Fetissi Selwa. 2015 International Conference on Renewable Energy Research and Applications (ICRERA). Year: 2015;

[12] Fast Calculation of Probabilistic Optimal Power Flow: A Deep Learning Approach, Yan Yang;Juan Yu;Zhifang Yang;Mingxu Xiang;Ren Liu. 2019 IEEE Power & Energy Society General Meeting (PESGM). Year: 2019;

[13] Chinese Patent CN104091092B;

[14] Reinforcement Learning Techniques for Optimal Power Control in Grid-Connected Microgrids: A Comprehensive Review. Erick O. Arwa;Komla A. Folly. IEEE Access. Year: 2020;

[15] A Data-driven Method for Fast AC Optimal Power Flow Solutions via Deep Reinforcement Learning, Yuhao Zhou;Bei Zhang;Chunlei Xu;Tu Lan;Ruisheng Diao;Di Shi;Zhiwei Wang;Wei-Jen Lee. Journal of Modern Power Systems and Clean Energy. Year: 2020;

[16] Real-Time Optimal Power Flow: A Lagrangian Based Deep Reinforcement Learning Approach, Ziming Yan;Yan Xu. IEEE Transactions on Power Systems. Year: 2020;

[17] DeepOPF: Deep Neural Network for DC Optimal Power Flow, Xiang Pan;Tianyu Zhao;Minghua Chen. 2019 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm). Year: 2019;

[18] Neural Network Optimal-Power-Flow. Proceedings of the 4th International Conference on Advances in Power System Control, Operation and Management, APSCOM-97, Hong Kong, November 1997.’. T.T. Nguyen;

[19] Multi-contingency transient stability-constrained optimal power flow using multilayer feedforward neural networks. Robert T. F. Ah King;Xiaoping Tu;Louis-A. Dessaint;Innocent Kamwa. 2016 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE). Year: 2016;

[20] Optimal power flows with voltage stability constraint based on voltage stability indices and artificial neural networks. Guilherme G. Lage;Ricardo AS Fernandes;Geraldo RM da Costa. 2015 IEEE Eindhoven PowerTech. Year: 2015. Summary of the Invention [Problem to be solved by the invention]

[0012] It can be seen that prior art load flow calculations usually suffer from problems such as low model accuracy, limited input parameters, or slow calculation speed. [Means for solving the problem]

[0013] In one aspect, an embodiment of the present disclosure provides a method for optimizing a power management system, including: Training data collected from a power grid is input into a machine learning model, a feedforward neural network, to identify a power flow model for the power grid. Obtain an initial solution for the decision variables in the optimal power flow model, take the initial solution as the current solution of the optimal power flow model, and proceed to an iterative calculation process, which includes the following steps: Step a: Update the current solution using gradient information for the decision variables of the output result of the objective function of the optimal power flow model obtained in the previous iterative calculation, and obtain an updated solution. Step b: Input the updated solution into the load flow model and obtain the output result of the load flow model. Step c: Using the output result of the power flow model as input, calculate the objective function of the optimal power flow model, obtain the latest solution of the optimal power flow model, and determine whether it satisfies the constraints and convergence criteria of the optimal power flow model. Step d: If the constraints or convergence criteria of the optimal power flow model are not satisfied, the latest solution is set as the current solution, and the process returns to step a for the next iterative calculation. Step e: If the constraints and convergence criteria of the optimal power flow model are satisfied, the iterative calculation process is terminated and the decision variables in the current solution are output as control commands to the power grid.

[0014] Optionally, in step a, if the current iteration is the first iteration in the iteration process, the rated power of the distributed energy devices in the power grid is directly taken as the updated solution, and then proceed to step b.

[0015] Optionally, the gradient information is a Jacobian matrix of each output value of the objective function of the optimal power flow model for each decision variable. In step a, the Jacobian matrix is ​​calculated, and the Jacobian matrix is ​​used to update the current solution according to a descending direction of the output value of the objective function to obtain an updated solution.

[0016] Optionally, the Jacobian matrix is ​​converted into computational equations including the load flow model using the chain rule for partial derivatives of a composite function, and the Jacobian matrix is ​​calculated using the computational equations.

[0017] Optionally, the input data in the training data includes state data collected from a plurality of pre-defined locations in the power grid, the state data including one or more of current, voltage, and power, and the output data in the training data includes at least current or voltage of pre-defined vulnerable locations in the power grid.

[0018] Optionally, the input data in the training data further comprises at least one of environmental information, scheduling commands and grid switching commands from a main power grid, where the environmental information comprises one or more of air temperature, humidity, precipitation, wind speed and solar radiation information.

[0019] Optionally, inputs for the optimal power flow model include pricing information, hardware capacity, grid code and control horizon.

[0020] Optionally, the inputs of the optimal power flow model further include at least one of the forecasted load, the forecasted power of distributed energy devices, scheduling commands from the main power grid and power grid switching commands.

[0021] Optionally, an objective function of the optimal power flow model is configured to add incentives for distributed energy devices to the main grid power purchase cost minus grid customer power sales revenue. Constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits of monitored locations, and controllability limits of distributed energy devices. Decision variables of the optimal power flow model include power of distributed energy devices with controllability in the power grid.

[0022] Optionally, an objective function of the optimal power flow model is configured to add incentives, carbon emission costs, and hardware aging costs for distributed energy devices to the main grid power purchase cost minus grid customer power sales revenue. Constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits of monitored locations, controllable range constraints of distributed energy devices, and main grid scheduling command constraints. Decision variables of the optimal power flow model include power of distributed energy devices with controllability in the power grid.

[0023] In a second aspect, at least one embodiment of the present disclosure provides an optimization apparatus for a power management system, including a power flow model identification module and an iterative processing module. The power flow model identification module inputs training data collected from a power grid into a machine learning model, which is a feedforward neural network, to identify a power flow model for the power grid. The iterative processing module obtains an initial solution for decision variables in an optimal power flow model, sets the initial solution as a current solution for the optimal power flow model, and proceeds to an iterative calculation process, which includes the following steps: Step a: Update the current solution using gradient information for the decision variables of the output result of the objective function of the optimal power flow model obtained in the previous iterative calculation, and obtain an updated solution. Step b: Input the updated solution into the load flow model and obtain the output result of the load flow model. Step c: Using the output result of the power flow model as input, calculate the objective function of the optimal power flow model, obtain the latest solution of the optimal power flow model, and determine whether it satisfies the constraints and convergence criteria of the optimal power flow model. Step d: If the constraints or convergence criteria of the optimal power flow model are not satisfied, the latest solution is set as the current solution, and the process returns to step a for the next iterative calculation. Step e: If the constraints and convergence criteria of the optimal power flow model are satisfied, the iterative calculation process is terminated and the decision variables in the current solution are output as control commands to the power grid.

[0024] In a third aspect, at least one embodiment of the present disclosure provides a computer-readable storage medium having stored thereon a program for, when executed by a processor, implementing the steps of the method set forth above. [Brief description of the drawings]

[0025] [Figure 1] FIG. 2 is a schematic diagram of an application scene of a method for optimizing a power management system according to an embodiment of the present disclosure; [Diagram 2] FIG. 2 is a schematic diagram of another application scenario of the optimization method for a power management system according to an embodiment of the present disclosure; [Diagram 3] 2 is a schematic diagram of a functional configuration of an OPF center 120 according to an embodiment of the present disclosure. [Figure 4] FIG. 1 is a schematic diagram of an OPF iterative calculation process according to an embodiment of the present disclosure. [Diagram 5] 1 is a flowchart of a method for optimizing a power management system according to an embodiment of the present disclosure. [Figure 6] FIG. 13 is another schematic diagram of the OPF iterative calculation process of an embodiment of the present disclosure. [Figure 7] 1 is a schematic diagram illustrating a configuration of an optimization device for a power management system according to an embodiment of the present disclosure. [Figure 8] FIG. 13 is another schematic diagram of an optimization device for a power management system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0026] In order to clarify the technical problems, technical solutions and advantages that the present disclosure aims to solve, the following detailed description will be given in conjunction with the drawings and specific embodiments. In the following description, specific details of specific configurations and components are provided only to assist in the complete understanding of the embodiments of the present disclosure. Therefore, it should be understood that those skilled in the art can make various changes and modifications to the embodiments described herein without departing from the scope and spirit of the present disclosure. In addition, the description of known functions and configurations will be omitted for clarity and conciseness.

[0027] It should be understood that the references "in one embodiment" or "in one embodiment" throughout the description mean that a particular feature, structure, or characteristic associated with an embodiment is included in at least one embodiment of the present disclosure. Thus, the appearances of "in one embodiment" or "in one embodiment" in various places throughout the specification are not necessarily all referring to the same embodiment. Moreover, the particular features, structures, or characteristics may be incorporated in any suitable manner in one or more embodiments.

[0028] In various embodiments of the present disclosure, it should be understood that the magnitude of the sequence numbers of each of the following processes does not mean the order of execution, and the order of execution of each process should be determined based on its function and inherent logic, without limiting the implementation process of the embodiments of the present disclosure.

[0029] The embodiment of the present disclosure provides an optimization method for a power management system, and in the optimal power flow calculation, an artificial neural network is used to model a power flow model. Specifically, the embodiment of the present disclosure uses a feedforward neural network to identify a power flow model. Input data in the training data of the feedforward neural network includes measured power grid conditions (which may be one or more of power, voltage, and current), environmental information, and power grid switching commands. Output data in the training data includes at least the voltage or power of a location under severe conditions in the power grid. The location under severe conditions may be selected in advance based on a vulnerable location point in the power grid. The vulnerable location point may be a performance bottleneck point, a potential fault point, or a potential danger point of the power grid. For example, if a transformer is a performance bottleneck point of the power grid, the transformer can be a vulnerable location point. The power and voltage in the training data can be obtained from different locations of the power grid. In addition, the embodiment of the present disclosure can use the power flow model obtained in the training to calculate a Jacobian matrix, thereby improving the efficiency of the iterative calculation in the optimal power flow.

[0030] FIG. 1 is a schematic diagram of an application scene of an optimization method for a power management system according to an embodiment of the present disclosure. The scene includes an OPF center 120, which is an optimization device for a power management system according to an embodiment of the present disclosure. The OPF center 120 is connected to a power grid 101 and may also be connected to a main power grid 119. The power grid 101 includes multiple distributed energy DER (Distributed Energy Resource) devices. The power grid 101 accesses the main power grid 119 through an access point 102. Each unit of the scene will be introduced in detail below.

[0031] 101 denotes a power grid under the supervision of the OPF center 120. This power grid may be a distribution grid, a microgrid, or an industrial estate grid, etc., and is typically smaller in size than the main power grid 119.

[0032] 102 shows the interconnection points of the power grid 101 and the main power grid 119. In a real scene, there may be multiple interconnection points. All the status data of these interconnection points are monitored by 120, including voltage, voltage phase angle, active power, reactive power, etc.

[0033] 103 and 104 are transformers in the power grid 101 .

[0034] 105, 110, 112, 114, 116, 118 are loads, energy storage, and power generation devices in the power grid 101, and are also called distributed energy (DER) devices or controllable devices if their power is adjustable, i.e., the power is flexible or controllable. 107 indicates a power control command of 105, and 129 indicates a controllable range of 105 (e.g., the power has a controllable range), and the controllable range information may be uploaded to the OPF center 120. If other devices in 101 also have controllability and their controllable range is one of the decision variables in the optimization calculation process by 120, this device needs to receive a control command similar to 107 and upload controllable range information to 120 similar to 129.

[0035] 106 is a power or voltage monitoring device for 105. Devices 109, 111, 113, 115, and 117 are similar devices. Note that, like 109, the monitoring device may be located in the monitored device 110. Note that, in the embodiment of the present disclosure, the state (power or voltage) of the power grid 101 under severe conditions such as heavy load or near overvoltage should be monitored by device 106. Furthermore, device 106 can also monitor the power or voltage of the interconnection point 102.

[0036] Reference numeral 108 denotes a monitoring device for one power line.

[0037] There may be other types of devices in the power grid 101, and if any device is generating or consuming significant power, the OPF center 120 needs to monitor its status via devices such as 106.

[0038] 119 denotes the main power grid. 120 denotes the OPF center, which calculates the optimal or most economical scheduling commands for the controllable devices (or DER devices) in the power grid 101.

[0039] 121 is a communication channel between 101 and 120. Data from 101 to 120 is specifically measurement data of monitoring devices such as 106 or 109, and data from 120 to 101 is specifically optimal or most economical scheduling or control commands for a particular device.

[0040] 122 is a grid switch command for grid 101. This command comes from the normal scheduling center of 101 or a similar function.

[0041] 123 indicates environmental information, which may specifically include information such as air temperature, humidity, precipitation, wind speed, solar radiation, etc. When the scale of the power grid 101 is large, the power grid 122 may include multiple environmental information of different geographical areas.

[0042] Reference numeral 124 denotes price information, and specifically, may include electricity rates, power generation prices of DER devices, incentives given to DER devices, and the like.

[0043] 125 indicates hardware capacity in the power grid 101, for example, the power, current or voltage capacity or limit range of a hardware device.

[0044] 126 indicates a grid code, for example a voltage limit condition. According to related standards, the grid code is usually static information.

[0045] 127 indicates the forecast load or forecast power of distributed energy devices. Here, the forecast load refers to the forecasted load of the power grid 101. The forecast power of DERs is optional and the time period can be 1 hour or 24 hours. 127 is an input of this disclosure.

[0046] 128 indicates a scheduling command input from the main power grid to the power grid 101. This command is optional.

[0047] 2 is a schematic diagram of a simplified hardware configuration of a machine learning-based power management system according to an embodiment of the present disclosure. Compared with FIG. 1, some data is removed in FIG. 2. For example, the predicted load or DER power 127 can be replaced with the same kind of data in the historical time range of the same time length as the OPF optimization calculation 210.

[0048] Additionally, the scheduling command 128 can be replaced with a static average value of the historical power over the same time period as the OPF optimization calculation 210. This historical power is measured at 102.

[0049] 3 provides a schematic diagram of the functional configuration of the OPF center 120 according to an embodiment of the present disclosure, where 201 is a part of the data input. In the embodiment shown in FIG. 2, 201 can include measurement data of the monitoring devices 106, 109, 111, 113, 115, 117. In the embodiment shown in FIG. 1, 201 can include the power grid switching command 122, the environmental information 123, and the measurement data of the monitoring devices 106, 109, 111, 113, 115, 117.

[0050] It should be noted that the measurement data, such as from a monitoring device 106, can include power and voltage. If the input of the machine learning data (ML data) is power, then the output of the machine learning data at 202 is voltage corresponding to a location on the power grid, and vice versa. The voltage and power may be measured from the same device, or from different devices or different locations.

[0051] It is noted that the output of the machine learning data generally includes at least the conditions of specific locations within the power grid 101 that are at risk of violating any of the constraints defined in 210.

[0052] Additionally, measurement data from monitoring devices 106 and the like can include real-time and historical data.

[0053] 202 indicates data pre-processing for machine learning (ML), which includes quantizing the power grid switching commands (mainly some discrete data) into numerical values ​​and reorganizing the measurement data of devices such as the monitoring devices 106 according to a pre-set fixed order. The basic rule of quantization is that the closer the switches are to each other in the power grid, the closer the quantized values ​​of their switching commands will be.

[0054] All of the 201 data are converted to a similar range, and one specific form is to normalize all the machine learning data.

[0055] 203 denotes preprocessed machine learning data for training, i.e., machine learning training data.

[0056] 204 indicates the tidal current (PF) modeling using a machine learning model, and the training data of the machine learning model is 203. The maximum likelihood model used in the machine learning model is a feedforward neural network. The specific configuration of the artificial neural network can be determined by experience and practical testing. Note that the feedforward neural network here does not have an internal feedback circuit, so the calculation (not training) of this feedforward neural network does not need to be repeated multiple times. The training method is the same as the conventional ML training method.

[0057] As defined at 202, the machine learning data can include multiple outputs with an error function that is the sum or weighted sum of all the outputs.

[0058] 205 and 206 are both tidal current (PF) models obtained by training based on ML.

[0059] 207 is the PF calculation of the ML-based PF model. 207 simply executes the trained ML model, whose inputs are 210 to 211, whose data structure is the same as 201, and whose output is the voltage or power of multiple devices or locations in the power grid 101. In this document, this PF model is represented as a function: PF(x).

[0060] 208 calculates the Jacobian matrix of the objective function of 210 for the decision variables of 210. Details will be described after 210 below.

[0061] 209 is a part of the data input, which in the embodiment shown in Fig. 2 includes price information 124, hardware capacity 125, grid code 126, and controllable range 129. In the embodiment shown in Fig. 1, the data input includes price information 124, hardware capacity 125, grid code 126, forecasted load and DER power 127, scheduling command 128, and controllable range 129.

[0062] 210 shows the optimal power flow OPF optimization calculation. The OPF model is a typical optimization problem. In the embodiment of the present disclosure, the optimization goal of the OPF model is to minimize the value of the objective function. Specifically, it is as follows:

[0063] For the embodiment shown in FIG. The objective function OBJ(x, PF(x), y) is minimized for the same period as the forecasted load and DER power 127 period, where OBJ(x, PF(x), y) = power purchase cost of the main power grid 119 - power sales revenue of power grid customers + incentives for distributed energy devices. The constraints include the following three conditions. (1) Hardware capacity and security operation restrictions (2) Grid code limitations, including line power and current limits at the monitored location (3) Limitation conditions for the controllable range of the DER device 129 Here, the objective function OBJ is a numerical function. The decision variables are the power of the distributed energy (DER) devices with controllability in the power grid 101, including active power and reactive power.

[0064] The above objective function for the embodiment shown in FIG. 2 can be modified to the objective function for the embodiment shown in FIG. 1 in the following manner. For example, if the relationship between hardware aging and device power in 101 is known, the hardware aging can be added to OBJ as part of the cost. Also, for example, if the power emission index of the main power grid and DERs in the power grid 101 is known, the total carbon emission cost of the power grid 101 can be calculated and added to OBJ. Also, the constraints can be increased.

[0065] For the embodiment shown in FIG. The objective function OBJ(x, PF(x), y) is minimized over the same period as the forecasted load and DER power 127 period, where OBJ(x, PF(x), y) = power purchase cost of the main power grid 119 - power sales revenue of power grid customers + incentives for distributed energy devices + total carbon emission cost of the power grid 101 + hardware aging cost of the power grid 101. The constraints include the following four conditions: (1) Hardware capacity and security operation restrictions (2) Grid code limitations, including line power and current limits at the monitored location (3) Limitation conditions for the controllable range of the DER device 129 (4) Limitation of 128 scheduling commands from the main power grid where OBJ is a vector-valued function. The decision variables are the powers of the distributed energy (DER) devices with controllability in the power grid 101, including active and reactive powers.

[0066] The OPF model can be computed using a distributed algorithm, such as the Benders decomposition algorithm, which splits the OPF problem into multiple parts that run on different hardware, or it can be computed using a traditional centralized algorithm.

[0067] The Jacobian matrix in 208 is a matrix of partial derivatives of each output value of the objective function OBJ with respect to the decision variables x, where x denotes the decision variables, including the decision variables for PF modeling (e.g., power of distributed energy (DER) devices with controllability) and may also include the remaining decision variables other than those for PF modeling described above. The remaining decision variables are selected as needed or empirically.

[0068] The Jacobian matrix can be shown as follows:

[0069]

number

[0070] where m is the number of decision variables x in the objective function OBJ, and n is the number of output values ​​of OBJ.

[0071] To calculate the Jacobian matrix, we can employ the usual chain rule for the calculation of partial derivatives of composite functions. i If is one of the output values ​​of OBJ, then:

[0072]

number

[0073] Here, ∂(PF(x)) / ∂x, can be easily calculated by the pre-trained, ML-based PF function analytical equation without the need for iterations.

[0074] 211 and 213 are both decision variables of 210 in the current iterative calculation.

[0075] 212 shows the output result of the PF model.

[0076] 214 indicates the calculated Jacobian matrix values, which are used in the optimization process of 210.

[0077] 215 shows the output of 210, some of which are the values ​​of decision variables that are sent as control commands to the corresponding distributed energy devices, and another part is the output of the ML-based PF model that can be output to a visualization front-end for display.

[0078] 207, 208, 211, 212, 213, and 214 are executed in multiple iterative calculations. See Figure 4 for the flow of one iterative calculation.

[0079] As shown in FIG. 4, the OPF iterative calculation process at 210 includes:

[0080] 301 indicates the data input defined in 209.

[0081] 302 indicates creating an initial solution. If the rated power of the distributed energy device is included in the predicted power of 127, the initial solution for the decision variable is that the decision variable is equal to the rated power of the distributed energy device. If the rated power of the distributed energy device is not included in the predicted power of 127, the decision variable is equal to the power of the distributed energy device in the previous time period, for example, the most recently measured power.

[0082] 303 indicates a solution for updating the OPF. If the Jacobian matrix has already been calculated, the current solution is updated according to the descending direction of the output value of the objective function based on the Jacobian matrix. If the Jacobian matrix is ​​not available, the current solution is not updated. The current solution after processing by 303 is input to 304.

[0083] 304 indicates the current solution input based on 303, and executes the PF model in 207 to obtain the output result of the PF model.

[0084] 305 shows the output result based on the PF model, and performs the calculation of the objective function OBJ in 210 to obtain the latest solution.

[0085] 306 indicates the calculation of the Jacobian matrix as defined in 208.

[0086] 307 indicates determining whether the current solution satisfies the constraints and convergence criteria of the OPF defined in 210. If the constraints and convergence criteria of the OPF defined in 210 are satisfied, the decision variables of the current solution are output as control commands 308 to the distributed energy devices of the power grid 101, and the iterative calculation process is terminated. If the constraints or convergence criteria of the OPF defined in 210 are not satisfied, the process returns to 303 to start the next iterative calculation.

[0087] Based on the above embodiments, the optimization method for a power management system according to an embodiment of the present disclosure, as shown in FIG. 5, includes the following steps.

[0088] In step 501, training data collected from a power grid is input into a machine learning model, a feedforward neural network, to identify a power flow model for the power grid.

[0089] Wherein, the input data in the training data includes state data collected from a plurality of pre-defined locations in the power grid, the state data includes one or more of current, voltage, and power, and the output data in the training data includes at least current or voltage of pre-defined vulnerable locations in the power grid.

[0090] Furthermore, the input data in the training data may further include at least one of environmental information, scheduling commands from a main power grid, and power grid switching commands, where the environmental information includes one or more of air temperature, humidity, precipitation, wind speed, and solar radiation information.

[0091] In step 502, an initial solution for the decision variables in the optimal power flow model is obtained, the initial solution is taken as the current solution for the optimal power flow model, and an iterative calculation process is started.

[0092] Wherein, the inputs of the optimal power flow model include price information, hardware capacity, grid code and controllable bounds. Optionally, the inputs of the optimal power flow model further include at least one of forecasted load, forecasted power of distributed energy devices, scheduling commands from the main power grid and power grid switching commands.

[0093] Corresponding to the embodiment shown in Fig. 2, the objective function of the optimal power flow model is configured to subtract the power sales revenue of the power grid customers from the power purchase cost of the main power grid and add the incentives of the distributed energy devices. The constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits of the monitored locations, and controllable range constraints of the distributed energy devices. The decision variables of the optimal power flow model include the power of the distributed energy devices with controllability in the power grid.

[0094] Corresponding to the embodiment shown in Fig. 1, the objective function of the optimal power flow model is configured to subtract the power sales revenue of the power grid customers from the power purchase cost of the main power grid, and add the incentives, carbon emission cost and hardware aging cost of the distributed energy devices. The constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits of the monitored locations, controllable range constraints of the distributed energy devices, and scheduling command constraints of the main power grid. The decision variables of the optimal power flow model include the power of the distributed energy devices with controllability in the power grid.

[0095] Here, the iterative calculation process includes the following steps, as shown in FIG.

[0096] In step a, the current solution is updated using gradient information for the decision variables of the output result of the objective function of the optimal power flow model obtained in the previous iterative calculation, and an updated solution is obtained.

[0097] Here, when performing the first iterative calculation, the output result of the objective function obtained in the previous iterative calculation does not exist. In this case, the measurement result of each decision variable obtained in the latest measurement is set as the updated solution, or the rated power of the distributed energy device corresponding to each decision variable is set as the updated solution, and then proceed to step b.

[0098] The gradient information is the Jacobian matrix of each output value of the objective function of the optimal power flow model for each decision variable. In step a, the Jacobian matrix is ​​calculated, and the Jacobian matrix is ​​used to update the current solution according to the descending direction of the output value of the objective function to obtain an updated solution.

[0099] When calculating the Jacobian matrix, the chain rule of partial derivatives of a composite function is used to convert the Jacobian matrix into a calculation equation including the load flow model, and the Jacobian matrix can be calculated using the calculation equation. The embodiment of the present disclosure can obtain the Jacobian matrix without requiring multiple iterative calculations, and can easily calculate gradient information for OPF calculation.

[0100] In step b, the updated solution is input into the power flow model and an output result of the power flow model is obtained.

[0101] In step c, the output result of the power flow model is used as an input, the objective function of the optimal power flow model is calculated, the latest solution of the optimal power flow model is obtained, and it is determined whether it satisfies the constraints and convergence criteria of the optimal power flow model.

[0102] In step d, if the constraints or convergence criteria of the optimal power flow model are not satisfied, the latest solution is taken as the current solution, and the process returns to step a for the next iterative calculation.

[0103] In step e, if the constraints and convergence criteria of the optimal power flow model are satisfied, the iterative process is terminated and the decision variables in the current solution are output as control commands to the power grid.

[0104] As can be seen from the above, the embodiment of the present disclosure can calculate gradient information (Jacobian matrix) for OPF calculation without requiring multiple iterations, and can improve the performance and speed of OPF calculation by using the gradient information, thereby reducing the amount of calculation for OPF calculation and improving the calculation efficiency. In addition, the embodiment of the present disclosure can model the PF model using a machine learning model, and more types of input data can be adopted as training data for machine learning, thereby improving the robustness and accuracy of the PF model.

[0105] Referring to Fig. 7, an embodiment of the present disclosure also provides an optimization device for a power management system, which corresponds to 120 in Fig. 1 and Fig. 2. As shown in Fig. 7, the device includes the following modules: The power flow model identification module 701 inputs training data collected from a power grid into a machine learning model, which is a feedforward neural network, to identify the power flow model of the power grid. The iterative processing module 702 obtains an initial solution for the decision variables in the optimal power flow model, sets the initial solution as the current solution of the optimal power flow model, and proceeds with an iterative calculation process, which includes the following steps: Step a: Update the current solution using gradient information for the decision variables of the output result of the objective function of the optimal power flow model obtained in the previous iterative calculation, and obtain an updated solution. Step b: Input the updated solution into the load flow model and obtain the output result of the load flow model. Step c: Using the output result of the power flow model as input, calculate the objective function of the optimal power flow model, obtain the latest solution of the optimal power flow model, and determine whether it satisfies the constraints and convergence criteria of the optimal power flow model. Step d: If the constraints or convergence criteria of the optimal power flow model are not satisfied, the latest solution is set as the current solution, and the process returns to step a for the next iterative calculation. Step e: If the constraints and convergence criteria of the optimal power flow model are satisfied, the iterative calculation process is terminated and the decision variables in the current solution are output as control commands to the power grid.

[0106] Optionally, in step a, if the current iteration is the first iteration in the iteration process, the rated power of the distributed energy devices in the power grid is directly taken as the updated solution, and then proceed to step b.

[0107] Optionally, the gradient information is a Jacobian matrix of each output value of the objective function of the optimal power flow model for each decision variable. In step a, the Jacobian matrix is ​​calculated, and the Jacobian matrix is ​​used to update the current solution according to a descending direction of the output value of the objective function to obtain an updated solution.

[0108] Optionally, the Jacobian matrix is ​​converted into computational equations including the load flow model using the chain rule for partial derivatives of a composite function, and the Jacobian matrix is ​​calculated using the computational equations.

[0109] Optionally, the input data in the training data includes state data collected from a plurality of pre-defined locations in the power grid, the state data including one or more of current, voltage, and power, and the output data in the training data includes at least current or voltage of pre-defined vulnerable locations in the power grid.

[0110] Optionally, the input data in the training data further comprises at least one of environmental information, scheduling commands and grid switching commands from a main power grid, where the environmental information comprises one or more of air temperature, humidity, precipitation, wind speed and solar radiation information.

[0111] Optionally, inputs for the optimal power flow model include pricing information, hardware capacity, grid code and control horizon.

[0112] Optionally, the inputs of the optimal power flow model further include at least one of the forecasted load, the forecasted power of distributed energy devices, scheduling commands from the main power grid and power grid switching commands.

[0113] Optionally, the objective function of the optimal power flow model includes the power purchase cost of the main grid plus incentives for distributed energy devices and minus the power sales revenue of grid customers. The optimal power flow model constraints include hardware capacity and security operation constraints, grid code constraints including line power and current limits of monitored locations, and distributed energy device controllability constraints. The optimal power flow model decision variables include the power of distributed energy devices with controllability in the power grid.

[0114] Optionally, the objective function of the optimal power flow model includes the power purchase cost of the main grid plus incentives, carbon emission costs, and hardware aging costs of distributed energy devices, and subtracts the power sales revenue of grid customers. The constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits of monitored locations, controllable range constraints of distributed energy devices, and scheduling command constraints of the main power grid. Decision variables of the optimal power flow model include the power of distributed energy devices with controllability in the power grid.

[0115] 8, an embodiment of the present disclosure provides yet another block diagram of an optimization device 800 for a power management system, including a processor 801, a transceiver 802, a memory 803, and a bus interface. Here, in the embodiment of the present disclosure, the optimization device 800 for a power management system further includes a program stored in the memory 803 and executable by the processor 801. When the program is executed by the processor 801, it realizes the following steps: Training data collected from a power grid is input into a machine learning model, a feedforward neural network, to identify a power flow model for the power grid. Obtain an initial solution for the decision variables in the optimal power flow model, take the initial solution as the current solution of the optimal power flow model, and proceed to an iterative calculation process, which includes the following steps: Step a: Update the current solution using gradient information for the decision variables of the output result of the objective function of the optimal power flow model obtained in the previous iterative calculation, and obtain an updated solution. Step b: Input the updated solution into the load flow model and obtain the output result of the load flow model. Step c: Using the output result of the power flow model as input, calculate the objective function of the optimal power flow model, obtain the latest solution of the optimal power flow model, and determine whether it satisfies the constraints and convergence criteria of the optimal power flow model. Step d: If the constraints or convergence criteria of the optimal power flow model are not satisfied, the latest solution is set as the current solution, and the process returns to step a for the next iterative calculation. Step e: If the constraints and convergence criteria of the optimal power flow model are satisfied, the iterative calculation process is terminated and the decision variables in the current solution are output as control commands to the power grid.

[0116] It can be understood that in the embodiments of the present disclosure, when the computer program is executed by the processor 801, the processes of each embodiment of the optimization method of the power management system described above can be realized and the same technical effect can be achieved, and will not be described further here to avoid duplication.

[0117] In Fig. 8, the bus architecture includes any number of interconnecting buses and bridges, specifically connecting various types of circuits, such as one or more processors, such as processor 801, and memory, such as memory 803. The bus architecture may also connect various types of other circuits, such as peripheral equipment, regulators, power management circuits, etc., all of which are known in the art and will not be described further herein. The interface is provided by a bus interface. The transceiver 802 may be multiple components, i.e., includes a transmitter and a receiver, and is provided as a unit that communicates with various types of other devices over a transmission medium.

[0118] The processor 801 manages the bus architecture and normal processing. The memory 803 can store data used during the operations of the processor 801.

[0119] It should be noted that the terminal in this embodiment is a device corresponding to the method shown in FIG. 3, and the implementation forms in each of the above embodiments can be applied to the embodiment of the terminal and achieve the same technical effects. In this device, the transceiver 802 and the memory 803, and the transceiver 802 and the processor 801 can be communicatively connected through a bus interface, and the function of the processor 801 can also be realized by the transceiver 802, and the function of the transceiver 802 can also be realized by the processor 801. It is necessary to explain here that the above-mentioned device provided by the embodiment of the present disclosure can realize all the method steps realized by the above-mentioned method embodiment and can achieve the same technical effects. Here, the same parts and beneficial effects of the embodiment of the method in this embodiment will not be specifically described.

[0120] In some embodiments of the present disclosure, a computer-readable storage medium is also provided having stored thereon a program that, when executed by a processor, performs the following steps: Training data collected from a power grid is input into a machine learning model, a feedforward neural network, to identify a power flow model for the power grid. Obtain an initial solution for the decision variables in the optimal power flow model, take the initial solution as the current solution of the optimal power flow model, and proceed to an iterative calculation process, which includes the following steps: Step a: Update the current solution using gradient information for the decision variables of the output result of the objective function of the optimal power flow model obtained in the previous iterative calculation, and obtain an updated solution. Step b: Input the updated solution into the load flow model and obtain the output result of the load flow model. Step c: Using the output result of the power flow model as input, calculate the objective function of the optimal power flow model, obtain the latest solution of the optimal power flow model, and determine whether it satisfies the constraints and convergence criteria of the optimal power flow model. Step d: If the constraints or convergence criteria of the optimal power flow model are not satisfied, the latest solution is set as the current solution, and the process returns to step a for the next iterative calculation. Step e: If the constraints and convergence criteria of the optimal power flow model are satisfied, the iterative calculation process is terminated and the decision variables in the current solution are output as control commands to the power grid.

[0121] When this program is executed by a processor, it can realize all the implementation methods of the optimization method of the power management system described above, and can achieve the same technical effects, which will not be repeated here in order to avoid duplication.

[0122] Those skilled in the art can understand that each example unit and algorithm step described in the embodiments disclosed herein can be realized by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software is determined by the specific application and design constraints of the technical means. Those skilled in the art can realize the described functions by different methods for each specific application, but these realizations should not be considered as going beyond the scope of the present disclosure.

[0123] For convenience and brevity, the specific operation processes of the above-described systems, devices and units are not described here repeatedly, and those are obvious to those skilled in the art, and should be referred to the corresponding processes in the method embodiments.

[0124] In some embodiments provided in the present application, it is to be understood that the disclosed apparatus and method may be implemented in other ways. The above-described apparatus embodiments are merely exemplary. For example, the division of the described units is merely a division of logical functions, and there are other division ways in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not implemented. In addition, the couplings, direct couplings, and communication connections between each component shown or discussed may be indirect couplings or communication connections via interfaces, devices, or units, and may be electrical, mechanical, or other types.

[0125] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units, i.e., they may be located in one place or distributed among multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the embodiments of the present disclosure.

[0126] Furthermore, each functional unit in each embodiment of the present disclosure may be integrated into a single processing unit, may be provided physically separately, or two or more may be integrated together.

[0127] The functions may be stored in a computer-readable storage medium when they are realized in the form of a software functional unit and sold or used as an independent product. Based on this understanding, the substantial or prior art contribution of the technical means of the present disclosure, or a part of the technical means, appears in the form of a software product. The computer software product is stored in a storage medium and includes some instructions that cause a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to each embodiment of the present disclosure. The storage medium includes various media that can store program code, such as U disk, mobile hard disk, read-only memory ROM (Read-Only Memory), random access memory RAM (Random Access Memory), magnetic disk, or optical disk.

[0128] The above description is a specific embodiment of the present disclosure, but the scope of protection of the present disclosure is not limited thereto. Any modifications or replacements that a person skilled in the art can easily think of within the technical scope disclosed by the present disclosure are included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims.

Claims

1. 1. A method for optimizing a power management system, comprising: inputting training data collected from a power grid into a machine learning model, the machine learning model being a feedforward neural network, to identify a power flow model of the power grid; Obtaining an initial solution for decision variables in the optimal power flow model, and setting the initial solution as a current solution for the optimal power flow model; a step a of updating a current solution using gradient information for a decision variable of an output result of an objective function of the optimal power flow model obtained in a previous iterative calculation, and obtaining an updated solution; b) inputting the updated solution into the load flow model and obtaining an output of the load flow model; a step c) of calculating an objective function of the optimal power flow model using the output result of the power flow model as an input, obtaining an updated solution of the optimal power flow model, and determining whether the updated solution satisfies the constraints and convergence criteria of the optimal power flow model; a step d of determining that the latest solution is a current solution if the constraints or convergence criteria of the optimal power flow model are not satisfied, and returning to step a to perform a next iterative calculation; and e) terminating the iterative process if the constraints and convergence criteria of the optimal power flow model are satisfied, and outputting decision variables in the current solution as control commands to the power grid.

2. The method of claim 1 , wherein in step a, if the current iteration is the first iteration in the iterative calculation process, the rated power of the distributed energy devices in the power grid is directly taken as the updated solution, and the method proceeds to step b.

3. The gradient information is a Jacobian matrix of each output value of an objective function of the optimal power flow model for each decision variable, 2. The method according to claim 1, wherein in the step a, the Jacobian matrix is ​​calculated, and the Jacobian matrix is ​​used to update the current solution according to a descending direction of the output value of the objective function to obtain an updated solution.

4. The method of claim 3 , further comprising converting the Jacobian matrix into computational equations including the load flow model using a chain rule for partial derivatives of a composite function, and using the computational equations to calculate the Jacobian matrix.

5. the input data in the training data set includes state data collected from a plurality of predefined location points in the power grid; the status data includes one or more of a current, a voltage, and a power; The method of claim 1 , wherein the output data in the training data includes at least a current or a voltage at a pre-defined vulnerable location in the power grid.

6. The input data in the training data further includes at least one of environmental information, scheduling commands from a main power grid, and power grid switching commands; The method of claim 5 , wherein the environmental information includes one or more of air temperature, humidity, precipitation, wind speed and solar radiation information.

7. The method of claim 1 , wherein inputs for the optimal power flow model include pricing information, hardware capacity, grid code, and control horizon.

8. 8. The method of claim 7, wherein inputs of the optimal power flow model further include at least one of a forecasted load, a forecasted power of distributed energy devices, scheduling commands from the main power grid, and power grid switching commands.

9. an objective function of the optimal power flow model is configured to add an incentive for distributed energy devices to the power purchase cost of the main power grid and to subtract the power sales revenue of the power grid customers; the constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits at the monitored locations, and controllable range constraints of distributed energy devices; The method of claim 1 , wherein decision variables of the optimal power flow model include power of distributed energy devices with controllability in a power grid.

10. The objective function of the optimal power flow model is configured to add incentives, carbon emission costs and hardware aging costs of distributed energy devices to the power purchase cost of the main power grid, and to subtract the power sales revenue of power grid customers; the constraints of the optimal power flow model include hardware capacity and security operation constraints, grid code constraints including line power and current limits of the monitored locations, controllable range constraints of distributed energy devices, and scheduling command constraints of the main power grid; The method of claim 1 , wherein decision variables of the optimal power flow model include power of distributed energy devices with controllability in a power grid.

11. An optimization device for a power management system, comprising: a power flow model identification module for inputting training data collected from a power grid into a machine learning model, which is a feedforward neural network, and identifying a power flow model of the power grid; Obtaining an initial solution for decision variables in the optimal power flow model, and setting the initial solution as a current solution for the optimal power flow model; a step a of updating a current solution using gradient information for a decision variable of an output result of an objective function of the optimal power flow model obtained in a previous iterative calculation, and obtaining an updated solution; b) inputting the updated solution into the load flow model and obtaining an output of the load flow model; a step c) of calculating an objective function of the optimal power flow model using the output result of the power flow model as an input, obtaining an updated solution of the optimal power flow model, and determining whether the updated solution satisfies the constraints and convergence criteria of the optimal power flow model; a step d of determining that the latest solution is a current solution if the constraints or convergence criteria of the optimal power flow model are not satisfied, and returning to step a to perform a next iterative calculation; and e) terminating the iterative calculation process and outputting decision variables in the latest solution as control commands to the power grid if the constraints and convergence criteria of the optimal power flow model are satisfied.

Images