A data-model dual-driven distributed resource participation power quality treatment method and device

Abstract

The application discloses a data-model double-driven distributed resource participation power quality treatment method and device, relates to the technical field of intelligent power grid optimal dispatching and power quality control, and comprises the following steps: constructing a multi-time scale optimal dispatching model; training a neural network aging response model based on historical operation data of energy storage; converting the neural network into a mixed integer linear constraint by adopting a large-M method; embedding the linearized aging constraint into the optimal model; and solving the model to generate optimal dispatching instructions. The application realizes accurate modeling of the aging behavior of energy storage, considers power quality and battery life in optimal dispatching, and significantly improves the economic efficiency and reliability of dispatching. The solvability of the model is enhanced through constraint linearization, and the real-time dispatching demand is met. The application supports multi-time scale cooperation and online model updating, and has good adaptability and engineering applicability.

Description

Technical Field

[0001] This invention relates to a data-model dual-driven distributed resource participation power quality governance method and device, belonging to the field of smart grid optimization scheduling and power quality control technology. Background Technology

[0002] With the increasing penetration of new energy sources in power distribution networks, a large number of renewable distributed power sources, represented by distributed photovoltaic (PV), wind power, and energy storage systems, are being connected to the distribution network, providing important support for achieving the goals of clean and low-carbon energy. However, the output of distributed power sources exhibits significant volatility and uncertainty, and their operating characteristics change dynamically with weather, load, and network conditions, posing new challenges to the voltage stability, power quality, and safe and economical operation of the distribution network. In actual operation, the control and scheduling of reactive power compensation equipment such as on-load tap-changing transformers, capacitor banks, and static var compensators (SVCs) can improve voltage levels to some extent, but there are shortcomings such as limited voltage regulation accuracy or excessively high voltage regulation costs. In addition, the partial conversion of photovoltaic (PV) active power output through inverters also provides considerable reactive power resources for system voltage regulation, but it requires full consideration of user preferences and lacks flexibility. In contrast, utilizing the inherent flexible location, flexible scheduling, and rapid response characteristics of energy storage systems (ESS) allows for optimized voltage distribution and improved system operational stability through reasonable energy storage configuration.

[0003] However, the lifespan of energy storage systems is significantly affected by charging and discharging behavior. Frequent deep charging and discharging, high-rate operation, and temperature changes can all accelerate battery aging. Traditional optimization models often ignore these aging characteristics, causing scheduling strategies to deviate from the optimal balance between long-term economic efficiency and reliability.

[0004] Currently, multi-timescale optimization scheduling has become the mainstream method to deal with the uncertainty of distributed power sources. However, how to accurately characterize the aging behavior of energy storage in the model and achieve the integration of data-driven and mechanism models remains a technical challenge that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a data-model dual-driven distributed resource participation power quality governance method. By integrating a data-driven energy storage aging model with multi-timescale optimization scheduling, it achieves synergistic optimization of power quality governance and energy storage lifespan, thereby improving the economic efficiency and reliability of distribution network operation.

[0006] To achieve the above objectives / to solve the above technical problems, the present invention is implemented using the following technical solution.

[0007] On the one hand, this invention provides a data-model dual-driven method for distributed resource participation in power quality governance, comprising the following steps:

[0008] The operation data generated by the target power grid is input into a pre-constructed multi-timescale optimization scheduling model embedded with a linearized aging constraint model for solution, generating the optimal scheduling instruction;

[0009] The method for constructing the multi-timescale optimization scheduling model embedding the linearized aging constraint model specifically includes:

[0010] Construct a multi-timescale optimization scheduling model for distributed resources to participate in power quality governance;

[0011] Based on historical operating data of energy storage systems under different charge / discharge rates, states of charge, and temperature conditions, a data-driven aging response model based on neural networks was constructed and trained.

[0012] An equivalent transformation strategy based on the Big-M method is adopted to transform the trained data-driven aging response model into a set of mixed integer linear constraints, thereby obtaining an embeddable linearized aging constraint model.

[0013] The linearized aging constraint model is embedded into the multi-timescale optimization scheduling model to obtain the constructed multi-timescale optimization scheduling model with embedded linearized aging constraint model.

[0014] Furthermore, the distributed resources include: renewable energy generation, controllable distributed power generation, and energy storage systems;

[0015] Where: renewable energy power generation, expressed as:

[0016] ;

[0017] in: Let be the photovoltaic active power output at time t of the i-th node. Let be the photovoltaic reactive power output at time t of the i-th node. For photovoltaic power generation devices, Let be the maximum usable photovoltaic power at time t of the i-th node. Let be the photovoltaic power output reduction rate at time t of the i-th node. Let be the maximum apparent power of the i-th node of the photovoltaic inverter;

[0018] Controllable distributed power generation, the expression is:

[0019] ;

[0020] in: Let be the active power output of the i-th node micro gas turbine at time t. Let be the active power output of the i-th node micro gas turbine at time t+1. Let be the reactive power output of the i-th node micro gas turbine at time t. Let i be the installed capacity of the micro gas turbine at the i-th node. The upward ramp rate limit for the i-th node micro gas turbine; The downhill ramp rate limit for the i-th node micro gas turbine;

[0021] Energy storage system, expressed as:

[0022] ;

[0023] in, The energy storage charging power at time t of the i-th node. Let be the energy storage and discharge power at time t of the i-th node. The maximum charging power for energy storage at time t of the i-th node. The maximum discharge power of the stored energy at time t at the i-th node. Let t represent the energy storage charging state of the i-th node at time t. This represents the energy storage and discharge state of the i-th node at time t. Store the energy for the i-th node at time t. For energy storage charging efficiency, For energy storage and discharge efficiency, For time step, Let be the battery degradation rate, q be the battery depth of discharge, b be the discharge rate, Q be the current usable battery capacity, z be an empirical coefficient in the battery model, and A be the battery degradation model parameters. This is a calibration constant for the battery's depth of discharge. The calibration constant for the attenuation relationship, The activation energy constant is... The gas constant is... The current operating temperature. This is the nominal reference temperature. Battery degradation model The calibration parameters, The attenuation coefficient is related to the discharge rate. Treat it as an exponential function; The maximum available capacity for storing energy in node t at time t. Let be the amount of energy storage capacity reduction at time t for the i-th node. Let be the battery degradation rate of the i-th node at time t. For the battery capacity at the next moment, The initial capacity for energy storage in the i-th node.

[0024] Furthermore, the objective function of the multi-timescale optimization scheduling model includes three stages: day-ahead scheduling stage, intraday rolling scheduling stage, and real-time feedback correction stage;

[0025] In the day-ahead scheduling phase, the expression is:

[0026] ;

[0027] in, The total operating cost for the entire day and night period is as follows. For the day-ahead scheduling time set, Let N be the power exchange cost between the regional distribution network and the main grid at time t during the current period, and N be the total number of nodes in the distribution network. The voltage deviation penalty cost for the i-th node in the current phase is... Let $t$ be the distributed generation cost of the $i$-th node at time $t$ in the current phase. Let $t$ be the operating cost of the distributed power storage system at the $i$-th node at time $t$ during the current phase. For OLTC costs, To account for the total regulation cost of the parallel capacitors, OLTC is an on-load tap changer. For the unit control cost of OLTC, The unit control cost of parallel capacitor banks, This is the voltage deviation penalty coefficient. Let be the voltage deviation at the i-th node. For the number of OLTC actions, This refers to the number of operations of the parallel capacitor bank;

[0028] During the intraday rolling scheduling phase, the expression is:

[0029] ;

[0030] in, The objective function for the total operating cost during the intraday rolling scheduling phase is... For the optimized time window of intraday rolling scheduling, For the current moment, For the predicted step size of rolling scheduling, The length of the time interval. Intraday rolling phase The cost of power exchange between the distribution network and the main grid at any given time. Let $\frac{i}{i}$ be the voltage deviation cost of the $i$ node during the intraday rolling phase. For the i-th node of the intraday rolling phase The cost of distributed power generation at any given time. For the i-th node of the intraday rolling phase The operating cost of energy storage at any time This is the time index for the intraday rolling scheduling phase.

[0031] In the real-time feedback correction phase, the expression is:

[0032] ;

[0033] in, Let C be the objective function for the real-time feedback correction phase, and let C be the controllable variable for adjusting resources during the feedback correction phase. The relevant values ​​for the adjustable resources at the current moment. This is the actual output feedback value of the adjustable resources at the current moment. To process the adjustment amount at the current moment, This is the reference output value obtained during the intraday rolling optimization phase at the current moment. These are the actual output feedback values ​​of adjustable resources at the current moment. Active power switched by the main network The active power output of the distributed power source at the i-th node. The reactive power output of the distributed power source at the i-th node. The charging power of the energy storage system at the i-th node. Let be the discharge power of the energy storage system at the i-th node. The reactive power support provided for the energy storage system at the i-th node. Let be the reactive power adjustment of the static var compensator at the i-th node.

[0034] This invention achieves hierarchical optimization of long-term economic efficiency, short-term adjustment, and real-time correction by setting objective functions for three stages: day-ahead, intraday, and real-time. This effectively coordinates the accuracy and real-time performance of the scheduling plan and improves the overall response capability of the system to fluctuations in renewable energy and load changes.

[0035] Furthermore, the constraints of the multi-time-scale optimization scheduling model include: power flow constraints and distribution network operation safety constraints;

[0036] Where: power flow constraint, expressed as:

[0037] ;

[0038] in: These are the distribution network node numbers, Let t be the active power flow of the branch from node i to node j at time t. Let t be the reactive power flow of the branch from node i to node j at time t. To sum all branches that start from the j-th node and point to the k-th downstream node, Let be the resistances of the branches from the i-th node to the j-th node. Let be the reactances of the branches from node i to node j. Let be the current magnitude of the branch from node i to node j. Let be the total net active power injection of the j-th node at time t. Let be the total net reactive power injection at the j-th node at time t. Let $t$ represent the active power demand of the $j$-th node at time $t$, the charging power of the energy storage system, the discharging power of the energy storage system, the active power output of the distributed generation, and the controllable load reduction, respectively. These represent the reactive load demand of the j-th node at time t, the reactive power provided by the energy storage system, the reactive output of the static var compensator, the reactive output of the controllable distributed power source, and the reactive compensation amount of the parallel capacitor, respectively.

[0039] The safety constraints for power distribution network operation are expressed as follows:

[0040] ;

[0041] in: Let be the voltage amplitude of the j-th node at time t. Let be the voltage amplitude of the i-th node at time t. Let be the lower and upper voltage limits of the i-th node, respectively. Let be the upper limit of the current in the branch from node i to node j, respectively. These are the minimum and maximum limits for the main network switching power, respectively. It switches active power to the main network.

[0042] This invention incorporates strict power flow equations and safety constraints such as node voltage, branch current, and main grid switching power into the optimization model, ensuring the physical feasibility of dispatching commands and the safety of system operation, and avoiding the problem of optimization results deviating from actual power grid operating conditions.

[0043] Furthermore, the construction and training of a data-driven aging response model based on neural networks, using historical operating data of the energy storage system under different charge / discharge rates, states of charge, and temperatures, specifically includes:

[0044] Constructing a data-driven aging response model based on neural networks, specifically:

[0045] Using historical operating data of the energy storage system under different charge / discharge rates, states of charge, and temperatures as the input layer, the predicted value of battery degradation rate as the output layer, and ReLU-based neurons as the hidden layer to construct a feedforward neural network;

[0046] Training a data-driven aging response model based on a neural network, specifically:

[0047] The constructed feedforward neural network uses mean squared error as the loss function, and the stochastic gradient descent algorithm is used to optimize the neural network parameters to minimize the loss function until a well-trained data-driven aging response model is obtained.

[0048] Furthermore, the expression for the feedforward neural network is:

[0049] ;

[0050] in: For the first The output value of the m-th neuron in the layer, where ReLU(·) is the linear rectified activation function.

[0051] For the first The output of the h-th neuron in layer h, where L is the number of layers in the neural network. The current layer number. No. The collection of neurons in a layer This is the output of the neurons in the layer preceding the output layer. For the input feature vector, The first The weights of neurons in layer mh, the first layer The bias term of the m-th neuron in the layer, the weight term of the corresponding output layer, and the bias term of the output layer. For the predicted battery degradation rate, The output of the output layer neurons;

[0052] The loss function expression is as follows:

[0053] ;

[0054] in: The deviation between the neural network's predicted value and the actual value. denoted as the total number of training samples, and 'a' as the actual decay rate.

[0055] This invention employs a data-driven feedforward neural network to model battery aging, avoiding the difficulty of parameter calibration in complex mechanism models. It automatically learns aging patterns from historical data, improving the model's generalization ability and prediction accuracy under varying operating conditions. Furthermore, by clearly defining the neural network hierarchical structure and adopting the mean squared error loss function, the training stability and convergence efficiency of the aging model are ensured, providing a model foundation for subsequent linearization transformation.

[0056] Furthermore, each ReLU activation function in the trained data-driven battery aging response model is equivalently transformed using the pre-defined Big-M method, mapping the entire neural network to a set of mixed integer linear constraints. This constructs a battery aging response constraint model for the energy storage system, expressed as:

[0057] ;

[0058] in, Let be the input value of the m-th neuron in the input layer. It is the bias term of the m-th neuron in the input layer. For the input features of the neural network, For the first The neuron weights connecting layer mh, Let be the linear input value of the m-th neuron in the l-th layer. For the first The bias term of the m-th neuron in the layer. No. The set of neurons in a layer, where L is the number of layers in the neural network. For the first The output of the h-th neuron in layer 1 For the first The output value of the m-th neuron in the layer. and Let be the lower and upper bounds of the linear input of the l-th layer, respectively. The l-th level binary variable represents whether the ReLU function is activated. Let m be the predicted output value of the m-th neuron in the output layer. This is the bias term for the m-th neuron in the output layer. The weights of the neurons connecting mh to the output layer. This is the ReLU output value of the h-th neuron connected to m in layer L-1.

[0059] The default Big-M method has the following expression:

[0060] ;

[0061] Among them, M L For the minimum value, M U d is the maximum value, d is the auxiliary 0-1 variable, x is the linear input of the neuron, i.e. the input value of the ReLU function, and v is the output value of the ReLU function.

[0062] Equivalence analysis: When the input x < 0, d is 0 and the output v is 0; when the input x > 0, d is 1 and the output v is x, verifying the equivalence of the constraints;

[0063] Furthermore, the embedding of the linearized aging constraint model into the multi-time-scale optimization scheduling model specifically involves:

[0064] The linearized aging constraint model replaces the original nonlinear battery degradation rate expression and is embedded in the multi-time-scale optimization scheduling model.

[0065] This invention directly replaces the original nonlinear expression with linearized aging constraints and embeds them into the optimization model, achieving a seamless integration of the data-driven model and the physical optimization model. This improves the model accuracy while maintaining the solvable framework of the optimization problem.

[0066] Furthermore, the step of inputting the operational data generated by the target power grid operation into a pre-constructed multi-timescale optimization scheduling model embedded with a linearized aging constraint model for solution, and generating the optimal scheduling instruction, specifically includes:

[0067] New operational data generated by the target power grid is used to continuously update and train a more accurate data-driven aging response model. The updated data-driven aging response model is then transformed into a new mixed-integer linear constraint through the equivalent transformation strategy of the Big-M method, and fed back to the multi-time-scale optimization scheduling model to generate a better scheduling strategy.

[0068] This invention introduces a model continuous update mechanism based on new operating data, which enables long-term adaptive optimization capabilities. It can track battery performance degradation and changes in the operating environment, dynamically adjust scheduling strategies, and extend the effective service life of the method.

[0069] Secondly, the present invention provides a data-model dual-driven distributed resource-based power quality management device, comprising:

[0070] The governance module is used to input the operation data generated by the target power grid into a pre-built multi-timescale optimization scheduling model embedded with a linearized aging constraint model for solution, and generate the optimal scheduling instruction.

[0071] The method for constructing the multi-timescale optimization scheduling model embedding the linearized aging constraint model specifically includes:

[0072] Construct a multi-timescale optimization scheduling model for distributed resources to participate in power quality governance;

[0073] Based on historical operating data of energy storage systems under different charge / discharge rates, states of charge, and temperature conditions, a data-driven aging response model based on neural networks was constructed and trained.

[0074] An equivalent transformation strategy based on the Big-M method is adopted to transform the trained data-driven aging response model into a set of mixed integer linear constraints, thereby obtaining an embeddable linearized aging constraint model.

[0075] The linearized aging constraint model is embedded into the multi-timescale optimization scheduling model to obtain the constructed multi-timescale optimization scheduling model with embedded linearized aging constraint model.

[0076] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention constructs a data-driven aging response model based on actual operating data, realistically depicting the battery degradation characteristics under different charge / discharge rates, states of charge, and temperatures. By embedding the data-driven aging model into multi-timescale optimization scheduling, it achieves synergistic optimization of power quality management and energy storage lifespan loss. This invention's method can comprehensively consider operating costs and battery health status at the day-ahead, intra-day, and real-time stages, avoiding energy storage performance degradation caused by overuse or improper scheduling, thereby extending equipment lifespan, reducing long-term operation and maintenance costs, and improving the economy and reliability of system operation. Simultaneously, the dual-drive mode provides a new paradigm for intelligent dispatching of distribution networks, helping to promote the transformation of dispatching decisions from relying on experience-based rules to data- and model-based scientific decision-making.

[0077] This invention uses the Big-M method to convert the neural network model into a mixed-integer linear constraint, thus transforming the original nonlinear nonconvex optimization problem into a standard mixed-integer linear programming problem. It can be solved quickly with the help of a mature and efficient solver, meeting the real-time requirements of multi-timescale scheduling.

[0078] This invention not only provides a static optimization model but also designs a model update mechanism based on new operational data. By continuously collecting actual operational data and retraining the aging model, the scheduling strategy can be dynamically adjusted and optimized, enabling the system to adapt to long-term evolution processes such as equipment aging and changes in the operating environment, exhibiting strong adaptability and robustness. Attached Figure Description

[0079] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0080] Figure 2 This is a schematic diagram of a neural network structure. Detailed Implementation

[0081] It should be noted that:

[0082] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0083] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0084] Example 1

[0085] like Figure 1 The embodiment shown provides a data-model dual-driven distributed resource participation power quality governance method, including the following steps:

[0086] Step 1: Construct a multi-time-scale optimization scheduling model for distributed resources to participate in power quality governance, specifically including:

[0087] Step 1.1: Based on the latest power forecast information, a closed-loop optimization control is constructed from three stages: day-ahead scheduling, intraday rolling scheduling, and real-time feedback correction, to promptly correct the deviation of the scheduling results.

[0088] In the day-ahead scheduling phase, the expression is:

[0089] ;

[0090] in, The total operating cost for the entire day and night period is as follows. For the day-ahead scheduling time set, Let N be the power exchange cost between the regional distribution network and the main grid at time t during the current period, and N be the total number of nodes in the distribution network. The voltage deviation penalty cost for the i-th node in the current phase is... Let $t$ be the distributed generation cost of the $i$-th node at time $t$ in the current phase. Let $t$ be the operating cost of the distributed power storage system at the $i$-th node at time $t$ during the current phase. For OLTC costs, To account for the total regulation cost of the parallel capacitors, OLTC is an on-load tap changer. For the unit control cost of OLTC, The unit control cost of parallel capacitor banks, This is the voltage deviation penalty coefficient. Let be the voltage deviation at the i-th node. For the number of OLTC actions, This represents the number of times the parallel capacitor bank operates.

[0091] The day-ahead optimization phase aims to minimize the total operating cost of the distribution network. During the day-ahead dispatch phase, active and reactive power coordination and optimization dispatch of the distribution network are carried out based on day-ahead load and renewable energy forecast information. The unit time interval of the day-ahead optimization dispatch phase is 1 hour. The decision variables include the active and reactive power output of distributed generation, flexible load adjustment, active and reactive power charging and discharging power of energy storage, switching position of compensation capacitors, and output of static var compensator (SVC).

[0092] During the intraday rolling scheduling phase, the expression is:

[0093] ;

[0094] in, The objective function for the total operating cost during the intraday rolling scheduling phase is... For the optimized time window of intraday rolling scheduling, For the current moment, For the predicted step size of rolling scheduling, The length of the time interval. Intraday rolling phase The cost of power exchange between the distribution network and the main grid at any given time. Let $\frac{i}{i}$ be the voltage deviation cost of the $i$ node during the intraday rolling phase. For the i-th node of the intraday rolling phase The cost of distributed power generation at any given time. For the i-th node of the intraday rolling phase The operating cost of energy storage at any time This is the time index for the intraday rolling scheduling phase.

[0095] During the intraday rolling scheduling phase, the optimization objective remains minimizing the distribution network operating cost, and the solution time is shortened from the previous 24 hours to... During the time period, the control variables are the active and reactive power output of the distributed power source, the active and reactive power charging and discharging power of the energy storage, and the reactive power output of the SVC.

[0096] In the real-time feedback correction phase, the expression is:

[0097] ;

[0098] in, Let C be the objective function for the real-time feedback correction phase, and let C be the controllable variable for adjusting resources during the feedback correction phase. The relevant values ​​for the adjustable resources at the current moment. This is the actual output feedback value of the adjustable resources at the current moment. To process the adjustment amount at the current moment, This is the reference output value obtained during the intraday rolling optimization phase at the current moment. These are the actual output feedback values ​​of adjustable resources at the current moment. Active power switched by the main network The active power output of the distributed power source at the i-th node. The reactive power output of the distributed power source at the i-th node. The charging power of the energy storage system at the i-th node. Let be the discharge power of the energy storage system at the i-th node. The reactive power support provided for the energy storage system at the i-th node. Let be the reactive power adjustment of the static var compensator at the i-th node.

[0099] During the real-time feedback correction phase, the output of each adjustable resource in the rolling phase is adjusted. The optimization goal is to minimize the adjustment of the adjustable resource output at the current moment, so as to ensure the stability of the system in response to renewable energy fluctuations.

[0100] Step 1.2: Construct the constraints of the multi-time-scale optimization scheduling model, including: power flow constraints and distribution network operation security constraints;

[0101] Where: power flow constraint, expressed as:

[0102] ;

[0103] in: These are the distribution network node numbers, Let t be the active power flow of the branch from node i to node j at time t. Let t be the reactive power flow of the branch from node i to node j at time t. To sum all branches that start from the j-th node and point to the k-th downstream node, Let be the resistances of the branches from the i-th node to the j-th node. Let be the reactances of the branches from node i to node j. Let be the current magnitude of the branch from node i to node j. Let be the total net active power injection of the j-th node at time t. Let be the total net reactive power injection at the j-th node at time t. Let $t$ represent the active power demand of the $j$-th node at time $t$, the charging power of the energy storage system, the discharging power of the energy storage system, the active power output of the distributed generation, and the controllable load reduction, respectively. These represent the reactive load demand of the j-th node at time t, the reactive power provided by the energy storage system, the reactive output of the static var compensator, the reactive output of the controllable distributed power source, and the reactive compensation amount of the parallel capacitor, respectively.

[0104] The safety constraints for power distribution network operation are expressed as follows:

[0105] ;

[0106] in: Let be the voltage amplitude of the j-th node at time t. Let be the voltage amplitude of the i-th node at time t. Let be the lower and upper voltage limits of the i-th node, respectively. Let be the upper limit of the current in the branch from node i to node j, respectively. These are the minimum and maximum limits for the main network switching power, respectively. It switches active power to the main network.

[0107] The aforementioned distributed resources include: renewable energy generation, controllable distributed power generation, and energy storage systems;

[0108] Where: renewable energy power generation, expressed as:

[0109] ;

[0110] in: Let be the photovoltaic active power output at time t of the i-th node. Let be the photovoltaic reactive power output at time t of the i-th node. For photovoltaic power generation devices, Let be the maximum usable photovoltaic power at time t of the i-th node. Let be the photovoltaic power output reduction rate at time t of the i-th node. Let be the maximum apparent power of the i-th node of the photovoltaic inverter;

[0111] Controllable distributed power generation, the expression is:

[0112] ;

[0113] in: Let be the active power output of the i-th node micro gas turbine at time t. Let be the active power output of the i-th node micro gas turbine at time t+1. Let be the reactive power output of the i-th node micro gas turbine at time t. Let i be the installed capacity of the micro gas turbine at the i-th node. The upward ramp rate limit for the i-th node micro gas turbine; The downhill ramp rate limit for the i-th node micro gas turbine;

[0114] Energy storage system, expressed as:

[0115] ;

[0116] in, The energy storage charging power at time t of the i-th node. Let be the energy storage and discharge power at time t of the i-th node. The maximum charging power for energy storage at time t of the i-th node. The maximum discharge power of the stored energy at time t at the i-th node. Let t represent the energy storage charging state of the i-th node at time t. This represents the energy storage and discharge state of the i-th node at time t. Store the energy for the i-th node at time t. For energy storage charging efficiency, For energy storage and discharge efficiency, For time step, Let be the battery degradation rate, q be the battery depth of discharge, b be the discharge rate, Q be the current usable battery capacity, z be an empirical coefficient in the battery model, and A be the battery degradation model parameters. These are the calibration constants for the relationship between battery discharge depth and degradation. The activation energy constant is... The gas constant is... The current operating temperature. This is the nominal reference temperature. Battery degradation model The calibration parameters, The attenuation coefficient is related to the discharge rate. Treat it as an exponential function; The maximum available capacity for storing energy in node t at time t. Let be the amount of energy storage capacity reduction at time t for the i-th node. Let be the battery degradation rate of the i-th node at time t. For the battery capacity at the next moment, The initial capacity for energy storage in the i-th node.

[0117] Step 2: Based on historical operating data of the energy storage system under different charge / discharge rates, states of charge, and temperature conditions, construct and train a data-driven aging response model based on a neural network, specifically including:

[0118] Step 2.1: As Figure 2 As shown, a data-driven aging response model based on neural networks is constructed as follows:

[0119] Using historical operating data of the energy storage system under different charge / discharge rates, states of charge, and temperatures as the input layer, the predicted value of battery degradation rate as the output layer, and ReLU-based neurons as the hidden layer to construct a feedforward neural network;

[0120] The expression for the feedforward neural network is:

[0121] ;

[0122] in: For the first The output value of the m-th neuron in the layer, where ReLU(·) is the linear rectified activation function.

[0123] For the first The output of the h-th neuron in layer h, where L is the number of layers in the neural network. The current layer number. No. The collection of neurons in a layer This is the output of the neurons in the layer preceding the output layer. For the input feature vector, The first The weights of neurons in layer mh, the first layer The bias term of the m-th neuron in the layer, the weight term of the corresponding output layer, and the bias term of the output layer. For the predicted battery degradation rate, This refers to the output of neurons in the output layer.

[0124] Step 2.2: Train a data-driven aging response model based on a neural network, specifically as follows:

[0125] The constructed feedforward neural network uses mean squared error as the loss function, and the stochastic gradient descent algorithm is used to optimize the neural network parameters to minimize the loss function until a well-trained data-driven aging response model is obtained.

[0126] The loss function expression is as follows:

[0127] ;

[0128] in: The deviation between the neural network's predicted value and the actual value. denoted as the total number of training samples, and 'a' as the actual decay rate.

[0129] Through model training, a data-driven aging response model based on neural networks is constructed to fit and infer the aging characteristics of energy storage systems under different operating conditions. The trained neural network model only relies on historical operating data and experimental samples and does not directly expose the specific operating parameters of the energy storage system, thereby achieving privacy protection for equipment parameters and operating data.

[0130] Step 3: Using an equivalent transformation strategy based on the Big-M method, the trained data-driven aging response model is transformed into a set of mixed integer linear constraints to obtain a linearized aging constraint model that can be embedded in the multi-timescale optimization scheduling model. Specifically, this includes:

[0131] The trained data drives each ReLU activation function in the battery aging response model. An equivalent transformation is performed using the pre-defined Big-M method, mapping the entire neural network to a set of mixed-integer linear constraints. This constructs a battery aging response constraint model for the energy storage system, expressed as:

[0132] ;

[0133] in, Let be the input value of the m-th neuron in the input layer. It is the bias term of the m-th neuron in the input layer. For the input features of the neural network, For the first The neuron weights connecting layer mh, Let be the linear input value of the m-th neuron in the l-th layer. For the first The bias term of the m-th neuron in the layer. No. The set of neurons in a layer, where L is the number of layers in the neural network. For the first The output of the h-th neuron in layer 1 For the first The output value of the m-th neuron in the layer. and Let be the lower and upper bounds of the linear input of the l-th layer, respectively. The l-th level binary variable represents whether the ReLU function is activated. Let m be the predicted output value of the m-th neuron in the output layer. This is the bias term for the m-th neuron in the output layer. The weights of the neurons connecting mh to the output layer. This is the ReLU output value of the h-th neuron connected to m in layer L-1.

[0134] The default Big-M method has the following expression:

[0135] ;

[0136] Among them, M L For the minimum value, M U d is the maximum value, d is the auxiliary 0-1 variable, x is the linear input of the neuron, i.e. the input value of the ReLU function, and v is the output value of the ReLU function.

[0137] Equivalence analysis: When the input x < 0, d is 0 and the output v is 0; when the input x > 0, d is 1 and the output v is x, verifying the equivalence of the constraints.

[0138] Due to the strong nonlinear and nonconvex characteristics of neural networks, the resulting aging response models are difficult to directly embed into multi-timescale optimization scheduling frameworks. To address this issue, this invention analyzes the internal structure and hidden layer connection characteristics of neural networks and, combined with the output characteristics of the battery aging rate model, proposes a method for constructing an aging response constraint model based on neural networks. This constraint model transforms the neural network structure into a Mixed Integer Linear Constraint (MILP) model through an equivalent linearization transformation, enabling direct use in scheduling optimization involving distributed power sources. This achieves the solvability of the aging mechanism and the engineering integrability of the model.

[0139] Step 4: Embed the linearized aging constraint model into the multi-time-scale optimization scheduling model. Specifically, replace the original nonlinear battery degradation rate expression with the linearized aging constraint model and embed it into the multi-time-scale optimization scheduling model.

[0140] Step 5: Input the operational data generated by the target power grid into a pre-constructed multi-timescale optimization scheduling model embedded with a linearized aging constraint model for solution, generating the optimal scheduling command, specifically including:

[0141] New operational data generated by power grid operation is used to continuously update and train a more accurate data-driven aging response model. The updated data-driven aging response model is then transformed into a new mixed-integer linear constraint through the equivalent transformation strategy of the Big-M method, and fed back to the multi-time-scale optimization scheduling model to generate a better scheduling strategy.

[0142] Based on the constraint learning algorithm, a battery aging response constraint model for energy storage systems is constructed. A data-model dual-driven distributed power source participation power quality governance framework under multiple time scales is proposed to achieve coordinated control of photovoltaic-storage system management and power quality optimization. The framework embeds a data-driven aging response model linearized by the constraint learning algorithm into the day-ahead and intraday optimization models to achieve dynamic balance between energy dispatch and aging constraints in the energy storage system.

[0143] The proposed data-model dual-drive mechanism achieves deep coupling between energy storage system management and distribution network multi-timescale scheduling model: the neural network model is trained based on the operation data generated by the optimization model, and then transformed into an embeddable linear constraint model through the constraint learning process, realizing the closed-loop integration of data-driven and optimization solution, thereby improving the economy, reliability and sustainability of distribution network operation.

[0144] Example 2

[0145] This embodiment provides a data-model dual-driven distributed resource-based power quality management device, including:

[0146] The governance module is used to input new operational data generated by the target power grid into a pre-embedded multi-timescale optimization scheduling model for solving and generating optimal scheduling instructions.

[0147] Wherein: the embedding method of the pre-embedded multi-timescale optimization scheduling model specifically includes:

[0148] Construct a multi-timescale optimization scheduling model for distributed resources to participate in power quality governance;

[0149] Based on historical operating data of energy storage systems under different charge / discharge rates, states of charge, and temperature conditions, a data-driven aging response model based on neural networks was constructed and trained.

[0150] An equivalent transformation strategy based on the Big-M method is adopted to transform the trained data-driven aging response model into a set of mixed integer linear constraints, thereby obtaining an embeddable linearized aging constraint model.

[0151] The linearized aging constraint model is embedded into the multi-time-scale optimization scheduling model to obtain the embedded multi-time-scale optimization scheduling model.

[0152] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0153] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0154] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0155] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0156] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A data-model dual-driven distributed resource participation power quality governance method, characterized in that, Includes the following steps: The operation data generated by the target power grid is input into a pre-constructed multi-timescale optimization scheduling model embedded with a linearized aging constraint model for solution, generating the optimal scheduling instruction; The method for constructing the multi-timescale optimization scheduling model embedding the linearized aging constraint model specifically includes: Construct a multi-timescale optimization scheduling model for distributed resources to participate in power quality governance; Based on historical operating data of energy storage systems under different charge / discharge rates, states of charge, and temperature conditions, a data-driven aging response model based on neural networks was constructed and trained. An equivalent transformation strategy based on the Big-M method is adopted to transform the trained data-driven aging response model into a set of mixed integer linear constraints, thereby obtaining an embeddable linearized aging constraint model, specifically including: The trained data drives each ReLU activation function in the battery aging response model. An equivalent transformation is performed using the pre-defined Big-M method, mapping the entire neural network to a set of mixed-integer linear constraints. This constructs a battery aging response constraint model for the energy storage system, expressed as: ； in, Let be the input value of the m-th neuron in the input layer. It is the bias term of the m-th neuron in the input layer. For the input features of the neural network, For the first The neuron weights connecting layer mh, Let be the linear input value of the m-th neuron in the l-th layer. For the first The bias term of the m-th neuron in the layer. No. The set of neurons in a layer, where L is the number of layers in the neural network. For the first The output of the h-th neuron in layer 1 For the first The output value of the m-th neuron in the layer. and Let be the lower and upper bounds of the linear input of the l-th layer, respectively. The l-th level binary variable represents whether the ReLU function is activated. Let m be the predicted output value of the m-th neuron in the output layer. This is the bias term for the m-th neuron in the output layer. The weights of the neurons connecting mh to the output layer. The ReLU output value of the h-th neuron connected to m in layer L-1 is given. The linearized aging constraint model is embedded into the multi-timescale optimization scheduling model to obtain the constructed multi-timescale optimization scheduling model with embedded linearized aging constraint model.

2. The data-model dual-driven distributed resource participation power quality governance method according to claim 1, characterized in that, The distributed resources include: renewable energy generation, controllable distributed power generation, and energy storage systems; Where: renewable energy power generation, expressed as: ； in: Let be the photovoltaic active power output at time t of the i-th node. Let be the photovoltaic reactive power output at time t of the i-th node. For photovoltaic power generation devices, Let be the maximum usable photovoltaic power at time t of the i-th node. Let be the photovoltaic power output reduction rate at time t of the i-th node. Let be the maximum apparent power of the i-th node of the photovoltaic inverter; Controllable distributed power generation, the expression is: ； in: Let be the active power output of the i-th node micro gas turbine at time t. Let be the active power output of the i-th node micro gas turbine at time t+1. Let be the reactive power output of the i-th node micro gas turbine at time t. Let i be the installed capacity of the micro gas turbine at the i-th node. The upward ramp rate limit for the i-th node micro gas turbine; The downhill ramp rate limit for the i-th node micro gas turbine; Energy storage system, expressed as: ； in, The energy storage charging power at time t of the i-th node. Let be the energy storage and discharge power at time t of the i-th node. The maximum charging power for energy storage at time t of the i-th node. The maximum discharge power of the stored energy at time t at the i-th node. Let t represent the energy storage charging state of the i-th node at time t. This represents the energy storage and discharge state of the i-th node at time t. Store the energy for the i-th node at time t. For energy storage charging efficiency, For energy storage and discharge efficiency, For time step, Let be the battery degradation rate, q be the battery depth of discharge, b be the discharge rate, Q be the current usable battery capacity, z be an empirical coefficient in the battery model, and A be the battery degradation model parameters. This is a calibration constant for the battery's depth of discharge. The calibration constant for the attenuation relationship, The activation energy constant is... The gas constant is... The current operating temperature. This is the nominal reference temperature. Battery degradation model The calibration parameters, The attenuation coefficient is related to the discharge rate. Treat it as an exponential function; Let the maximum available energy storage capacity of the i-th node at time t be determined. Let be the amount of energy storage capacity reduction at time t for the i-th node. Let be the battery degradation rate of the i-th node at time t. For the battery capacity at the next moment, The initial capacity for energy storage in the i-th node.

3. The data-model dual-driven distributed resource participation power quality governance method according to claim 1, characterized in that, The objective function of the multi-timescale optimization scheduling model includes three stages: day-ahead scheduling stage, intraday rolling scheduling stage, and real-time feedback correction stage. In the day-ahead scheduling phase, the expression is: ； in, The total operating cost for the entire day and night period is as follows. For the day-ahead scheduling time set, Let N be the power exchange cost between the regional distribution network and the main grid at time t during the current period, and N be the total number of nodes in the distribution network. The voltage deviation penalty cost for the i-th node in the current phase is... Let $t$ be the distributed generation cost of the $i$-th node at time $t$ in the current phase. Let $t$ be the operating cost of the distributed power storage system at the $i$-th node at time $t$ during the current phase. For OLTC costs, To account for the total regulation cost of the parallel capacitors, OLTC is an on-load tap changer. For the unit control cost of OLTC, The unit control cost of parallel capacitor banks, This is the voltage deviation penalty coefficient. Let be the voltage deviation at the i-th node. For the number of OLTC actions, This refers to the number of operations of the parallel capacitor bank; During the intraday rolling scheduling phase, the expression is: ； in, The objective function for the total operating cost during the intraday rolling scheduling phase is... For the optimized time window of intraday rolling scheduling, For the current moment, For the predicted step size of rolling scheduling, The length of the time interval. Intraday rolling phase The cost of power exchange between the distribution network and the main grid at any given time. Let $\frac{i}{i}$ be the voltage deviation cost of the $i$ node during the intraday rolling phase. For the i-th node of the intraday rolling phase The cost of distributed power generation at any given time. For the i-th node of the intraday rolling phase The operating cost of energy storage at any time This serves as the time index for the intraday rolling scheduling phase. In the real-time feedback correction phase, the expression is: ； in, Let C be the objective function for the real-time feedback correction phase, and let C be the controllable variable for adjusting resources during the feedback correction phase. The relevant values ​​for the adjustable resources at the current moment. This is the actual output feedback value of the adjustable resources at the current moment. To process the adjustment amount at the current moment, This is the reference output value obtained during the intraday rolling optimization phase at the current moment. These are the actual output feedback values ​​of adjustable resources at the current moment. Active power switched by the main network The active power output of the distributed power source at the i-th node. The reactive power output of the distributed power source at the i-th node. The charging power of the energy storage system at the i-th node. Let be the discharge power of the energy storage system at the i-th node. The reactive power support provided for the energy storage system at the i-th node. Let be the reactive power adjustment of the static var compensator at the i-th node.

4. The data-model dual-driven distributed resource participation power quality governance method according to claim 1, characterized in that, The constraints of the multi-timescale optimization scheduling model include: power flow constraints and distribution network operation security constraints; Where: power flow constraint, expressed as: ； in: These are the distribution network node numbers, Let t be the active power flow of the branch from node i to node j at time t. Let t be the reactive power flow of the branch from node i to node j at time t. To sum all branches that start from the j-th node and point to the k-th downstream node, Let be the resistances of the branches from the i-th node to the j-th node. Let be the reactances of the branches from node i to node j. Let be the current magnitude of the branch from node i to node j. Let be the total net active power injection of the j-th node at time t. Let be the total net reactive power injection at the j-th node at time t. Let $t$ represent the active power demand of the $j$-th node at time $t$, the charging power of the energy storage system, the discharging power of the energy storage system, the active power output of the distributed generation, and the controllable load reduction, respectively. These represent the reactive load demand of the j-th node at time t, the reactive power provided by the energy storage system, the reactive output of the static var compensator, the reactive output of the controllable distributed power source, and the reactive compensation amount of the parallel capacitor, respectively. The safety constraints for power distribution network operation are expressed as follows: ； in: Let be the voltage amplitude of the j-th node at time t. Let be the voltage amplitude of the i-th node at time t. Let be the lower and upper voltage limits of the i-th node, respectively. Let be the upper limit of the current in the branch from node i to node j, respectively. These are the minimum and maximum limits for the main network switching power, respectively. It switches active power to the main network.

5. The data-model dual-driven distributed resource participation power quality governance method according to claim 1, characterized in that, The aforementioned method involves constructing and training a data-driven aging response model based on neural networks, using historical operating data of the energy storage system under different charge / discharge rates, states of charge, and temperatures. Specifically, this includes: Constructing a data-driven aging response model based on neural networks, specifically: Using historical operating data of the energy storage system under different charge / discharge rates, states of charge, and temperatures as the input layer, the predicted value of battery degradation rate as the output layer, and ReLU-based neurons as the hidden layer to construct a feedforward neural network; Training a data-driven aging response model based on a neural network, specifically: The constructed feedforward neural network uses mean squared error as the loss function, and the stochastic gradient descent algorithm is used to optimize the neural network parameters to minimize the loss function until a well-trained data-driven aging response model is obtained.

6. The data-model dual-driven distributed resource participation power quality governance method according to claim 5, characterized in that, The expression for the feedforward neural network is: ； in: For the first The output value of the m-th neuron in the layer, where ReLU(·) is the linear rectified activation function. For the first The output of the h-th neuron in layer h, where L is the number of layers in the neural network. The current layer number. No. The collection of neurons in a layer This is the output of the neurons in the layer preceding the output layer. For the input feature vector, The first The weights of neurons in layer mh, the first layer The bias term of the m-th neuron in the layer, the weight term of the corresponding output layer, and the bias term of the output layer. For the predicted battery degradation rate, The output of the output layer neurons; The loss function expression is as follows: ； in: The deviation between the neural network's predicted value and the actual value. denoted as the total number of training samples, and 'a' as the actual decay rate.

7. The data-model dual-driven distributed resource participation power quality governance method according to claim 1, characterized in that, The specific steps of embedding the linearized aging constraint model into the multi-timescale optimization scheduling model are as follows: The linearized aging constraint model replaces the original nonlinear battery degradation rate expression and is embedded in the multi-time-scale optimization scheduling model.

8. The data-model dual-driven distributed resource participation power quality governance method according to claim 1, characterized in that, The step of inputting the operational data generated by the target power grid operation into a pre-constructed multi-timescale optimization scheduling model embedded with a linear aging constraint model for solution, and generating the optimal scheduling instruction, specifically includes: New operational data generated by the target power grid is used to continuously update and train a more accurate data-driven aging response model. The updated data-driven aging response model is then transformed into a new mixed-integer linear constraint through the equivalent transformation strategy of the Big-M method, and fed back to the multi-time-scale optimization scheduling model to generate a better scheduling strategy.

9. A data-model dual-driven distributed resource-based power quality management device, characterized in that, include: The governance module is used to input the operation data generated by the target power grid into a pre-built multi-timescale optimization scheduling model embedded with a linearized aging constraint model for solution, and generate the optimal scheduling instruction. The method for constructing the multi-timescale optimization scheduling model embedding the linearized aging constraint model specifically includes: Construct a multi-timescale optimization scheduling model for distributed resources to participate in power quality governance; Based on historical operating data of energy storage systems under different charge / discharge rates, states of charge, and temperature conditions, a data-driven aging response model based on neural networks was constructed and trained. An equivalent transformation strategy based on the Big-M method is adopted to transform the trained data-driven aging response model into a set of mixed integer linear constraints, thereby obtaining an embeddable linearized aging constraint model, specifically including: The trained data drives each ReLU activation function in the battery aging response model. An equivalent transformation is performed using the pre-defined Big-M method, mapping the entire neural network to a set of mixed-integer linear constraints. This constructs a battery aging response constraint model for the energy storage system, expressed as: ； in, Let be the input value of the m-th neuron in the input layer. It is the bias term of the m-th neuron in the input layer. For the input features of the neural network, For the first The neuron weights connecting layer mh, Let be the linear input value of the m-th neuron in the l-th layer. For the first The bias term of the m-th neuron in the layer. No. The set of neurons in a layer, where L is the number of layers in the neural network. For the first The output of the h-th neuron in layer 1 For the first The output value of the m-th neuron in the layer. and Let be the lower and upper bounds of the linear input of the l-th layer, respectively. The l-th level binary variable represents whether the ReLU function is activated. Let m be the predicted output value of the m-th neuron in the output layer. This is the bias term for the m-th neuron in the output layer. The weights of the neurons connecting mh to the output layer. The ReLU output value of the h-th neuron connected to m in layer L-1 is given. The linearized aging constraint model is embedded into the multi-timescale optimization scheduling model to obtain the constructed multi-timescale optimization scheduling model with embedded linearized aging constraint model.

Images