An optimal configuration system for energy storage devices based on ac-dc power grid

Abstract

The application relates to the technical field of power system energy storage planning, and specifically discloses a system for optimizing configuration of energy storage equipment based on an AC-DC power grid, which acquires a basic data set; adopts a density peak clustering algorithm to reduce scenes of wind and light output and load time series data, and generates a multi-scene time series data set; obtains electrical quantities of each node through power flow calculation, and calculates modal sensitivity indexes of each node on network loss and voltage stability based on a Jacobian matrix modal decoupling analysis, so as to screen a set of energy storage configuration nodes to be selected; takes the set of energy storage configuration nodes to be selected as the basis, takes minimization of the whole life cycle cost of the system as the target, and solves to obtain preliminary configuration power and capacity; substitutes the preliminary configuration scheme into electromagnetic transient simulation for dynamic response verification, and if the safety and stability requirements are met, the final scheme is output, otherwise, the constraint conditions are corrected according to response deviation and re-iterative optimization is carried out; the application realizes collaborative optimization between economy and dynamic safety and stability of the energy storage configuration scheme.

Description

Technical Field

[0001] This invention relates to the field of power system energy storage planning technology, and specifically to an optimized configuration system for energy storage devices based on AC / DC power grids. Background Technology

[0002] With the continuous expansion of new energy power generation, AC / DC hybrid power grids have become an important form of the current power system. Energy storage devices, as a key means to improve grid flexibility and the ability to absorb new energy, have their location and capacity directly affecting the system's operational economy and safety stability. Currently, energy storage optimization configuration methods for AC / DC power grids are generally divided into two categories: one is a planning method based on steady-state economic indicators, aiming to minimize system investment and operating costs, and determining the access location and capacity of energy storage through power flow calculations; the other is a screening method based on sensitivity analysis, which selects key nodes for configuration by calculating the sensitivity indicators of nodes to network losses or voltage.

[0003] Existing AC / DC grid energy storage optimization configuration methods only focus on steady-state economic indicators based on multi-scenario time-series data. After obtaining a preliminary configuration scheme by minimizing the total life cycle cost, it cannot ensure that the grid voltage dynamic trajectory and converter station power response can meet the preset safety and stability threshold requirements when encountering extreme and severe faults such as AC three-phase short circuits or DC unipolar blocking. This leads to the potential risk of voltage instability or equipment overload under actual large disturbance impacts, resulting in a serious disconnect between the economically optimal scheme and dynamic safety requirements. Summary of the Invention

[0004] The purpose of this invention is to provide an optimized configuration system for energy storage devices based on AC / DC power grids to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] An optimized configuration system for energy storage devices based on AC / DC power grids includes a data acquisition module that acquires topology data, historical operation data, and technical and economic parameters of energy storage devices from AC / DC power grids to form a basic dataset.

[0007] The scene generation module, based on historical operating data in the basic dataset, uses the density peak clustering algorithm to reduce the time series data of wind and solar power output and load to generate multiple typical operating scenarios and their corresponding probability weights, and outputs a multi-scenario time series dataset.

[0008] The sensitivity analysis module inputs multi-scenario time-series datasets into the AC / DC power flow calculation process, obtains the voltage amplitude, phase angle, and active and reactive power of each node under various typical scenarios, and calculates the modal sensitivity index of each node to the network loss and voltage stability of the whole network. Based on the sensitivity index, the module sorts and selects the set of nodes to be selected for energy storage configuration.

[0009] The capacity optimization module is based on the set of candidate energy storage configuration nodes, with the goal of minimizing the system's total life cycle cost. It also considers constraints on energy storage charging and discharging power, state of charge, and safe operation of AC / DC power grids. It constructs an optimization configuration objective function and uses a multi-objective particle swarm evolution strategy to solve the objective function, thereby obtaining the preliminary configuration power and capacity of the energy storage device.

[0010] The dynamic verification module substitutes the initial power and capacity configuration into the electromagnetic transient simulation process of the AC / DC power grid, simulates the dynamic response characteristics under the preset fault set, and outputs the final optimized configuration scheme if the transient response meets the safety and stability requirements; otherwise, it corrects the constraints based on the transient response deviation and re-triggers the capacity optimization module.

[0011] As a further aspect of the present invention: the output multi-scene time-series dataset specifically includes:

[0012] The wind and solar power output time series data and load time series data in the historical operation data are normalized and abnormal fluctuation points are removed to form a standardized time series sample set.

[0013] Based on a standardized time series sample set, the similarity between any two time series samples is calculated using dynamic time curvature distance, and a distance matrix is ​​constructed.

[0014] Based on the distance matrix, determine the local density of each time series sample and the relative distance between the time series sample and the higher density sample, and draw a decision map based on the product of local density and relative distance. Select the time series sample whose product value is located in the upper right region of the decision map as the cluster center.

[0015] The remaining time series samples are assigned to the nearest cluster centers to form multiple typical operating scenarios. The probability weights corresponding to each typical operating scenario are calculated based on the proportion of the original time series samples covered by each cluster center to the total number of samples, and a multi-scenario time series dataset is output.

[0016] As a further aspect of the present invention: the construction of the distance matrix specifically includes:

[0017] The wind and solar power output time series data and load time series data in the standardized time series sample set are aligned by timestamp, and then a multi-dimensional time series vector group is formed after alignment.

[0018] For any two multidimensional time series vector groups, a dynamic time warping algorithm is used to calculate an optimal warping path that minimizes the cumulative distance between the two vector groups, while satisfying the boundary conditions and continuity constraints. The corresponding cumulative distance is then used as the dynamic time warping distance between the two vector groups.

[0019] Based on the dynamic time bending distance of all pairwise combinations, a real symmetric distance matrix is ​​constructed. Each element in the distance matrix is ​​used to characterize the morphological similarity of the corresponding two time series samples under nonlinear time axis scaling.

[0020] As a further aspect of the present invention: the calculation of the modal sensitivity index of each node to the overall network loss and voltage stability specifically includes:

[0021] Based on the power flow calculation results under various typical scenarios, the Jacobian matrix of the AC power grid is extracted, and modal decoupling analysis is performed on the Jacobian matrix to obtain several dominant eigenvalues ​​that are strongly correlated with network loss and voltage stability.

[0022] The participation factor of the injected power of each node on the dominant eigenvalue is calculated, and the participation factor is normalized to obtain the first sensitivity value and the second sensitivity value of each node on the impact of network loss and voltage stability.

[0023] The first sensitivity value and the second sensitivity value are weighted and fused according to a preset weight ratio to form a comprehensive modal sensitivity index for each node. The nodes are sorted from high to low according to the corresponding comprehensive modal sensitivity index, and the nodes with the highest ranking are selected as the set of candidate energy storage configuration nodes.

[0024] As a further aspect of the present invention: the normalization of the participation factors to obtain a first sensitivity value characterizing the impact of each node on network loss and a second sensitivity value characterizing the impact on voltage stability specifically includes:

[0025] For each dominant eigenvalue, a small perturbation is applied to the injected power of each node obtained from the power flow calculation, and the ratio between the change in injected power of each node and the change in dominant eigenvalue is solved by the perturbation method.

[0026] The absolute value of the ratio is taken as the original participation factor of the corresponding node with respect to the current dominant eigenvalue, and the original participation factors of the same node under all dominant eigenvalues ​​are summed.

[0027] The summation results of the dominant feature root group of network loss association are normalized by the maximum value to obtain the first sensitivity value of each node to the impact of network loss.

[0028] The summation results of the dominant eigenvalues ​​related to voltage stability are normalized to obtain the second sensitivity value of each node to the influence of voltage stability.

[0029] As a further aspect of the present invention: obtaining the preliminary configuration power and capacity of the energy storage device specifically includes:

[0030] Each node in the set of candidate energy storage configuration nodes is taken as the location to be connected to the energy storage device. The rated power and rated capacity of the energy storage device are taken as the variables to be optimized. A population containing multiple particles is initialized, and each particle represents an energy storage configuration scheme.

[0031] Each energy storage configuration scheme is substituted into the AC / DC power grid time-series operation simulation under various typical operating scenarios. The sum of investment cost and operating cost over the entire life cycle of the system is calculated, and the reciprocal of the corresponding cost value is used as the fitness value of the particle.

[0032] Based on the fitness value of each particle, update the individual historical best position of the particle and the global best position of the population, and iteratively update the position and velocity of the particles according to the inertia weight and learning factor in the particle swarm evolution strategy.

[0033] When the number of iterations reaches the preset upper limit or the global optimal fitness value remains unchanged for multiple consecutive times, the iteration is terminated, and the rated power and rated capacity corresponding to the current global optimal position are used as the initial configuration power and capacity output of the energy storage device.

[0034] As a further aspect of the present invention: the iterative update of the particle's position and velocity specifically includes:

[0035] Obtain the current iteration number, and dynamically calculate the value of the inertia weight based on the ratio of the current iteration number to the total iteration number, so that the inertia weight decreases non-linearly from the initial value to the final value as the iteration progresses;

[0036] Compare the current fitness value of each particle with the fitness value of its historical best position. If the current fitness value is better, update the current position to the historical best position of the corresponding particle. At the same time, compare the fitness values ​​of the historical best positions of all particles and update the position with the highest fitness value to the global best position of the population.

[0037] Based on the current inertia weight and two learning factors, and combining the difference between each particle's current position and its individual historical best position and the population's global best position, the movement speed of the corresponding particle in the solution space is calculated, and the particle's position is updated according to the movement speed to generate a new generation of energy storage configuration scheme.

[0038] As a further aspect of the present invention: the electromagnetic transient simulation process of substituting the initial power and capacity configuration into the AC / DC power grid to simulate the dynamic response characteristics under a preset fault set specifically includes:

[0039] The initial power and capacity configurations are used as parameters for the energy storage device. They are then substituted into the electromagnetic transient simulation program of the AC / DC power grid. Various AC short-circuit faults and DC blocking faults are applied one by one in the preset fault set to simulate the dynamic response process of the power grid and extract the voltage trajectory of key nodes and the power trajectory of the converter station under each fault.

[0040] The extracted voltage and power trajectories are compared with preset safety and stability thresholds. If the dynamic response under all faults meets the threshold requirements, the preliminary power and capacity configuration is output as the final optimized configuration scheme.

[0041] If the dynamic response under any fault does not meet the threshold requirement, the deviation between the dynamic response under the corresponding fault and the safety and stability threshold is calculated. Based on the magnitude of the deviation, the voltage stability constraint boundary or network loss constraint boundary in the objective function is adjusted in reverse. The adjusted constraint conditions are then re-input into the capacity optimization module for iterative solution.

[0042] The beneficial effects of this invention are:

[0043] (1) In this invention, the preliminary configuration scheme obtained from capacity optimization is substituted into electromagnetic transient simulation for dynamic response verification, and the constraint boundary in the objective function of optimization configuration is corrected in reverse according to the deviation between the transient response and the safety and stability threshold. This closed-loop feedback mechanism enables the final output energy storage configuration scheme to not only meet the requirement of minimum cost over the entire life cycle in terms of economy, but also to ensure that the voltage and power dynamic response of the power grid meets the safety and stability requirements under preset faults such as AC short circuit or DC blockage, which significantly improves the reliability and adaptability of the configuration scheme in actual operation.

[0044] (2) This invention performs modal decoupling analysis on the Jacobian matrix of power flow calculation, calculates the modal sensitivity index of each node to the network loss and voltage stability of the entire network, and sorts and selects the set of nodes to be selected for energy storage configuration based on the index. This technical means can accurately identify the key nodes in the power grid that have the greatest impact on the economic efficiency and stability of the system operation, effectively reduce the search space for subsequent capacity optimization, avoid blindly configuring all nodes in the entire network, thereby improving the optimization solution efficiency and reducing the computational complexity. Attached Figure Description

[0045] The invention will now be further described with reference to the accompanying drawings.

[0046] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1 As shown, the present invention is an optimized configuration system for energy storage devices based on AC / DC power grids, comprising:

[0049] The data acquisition module acquires topology data of AC / DC power grids, historical operation data, and technical and economic parameters of energy storage devices to form a basic dataset;

[0050] The scene generation module, based on historical operating data in the basic dataset, uses the density peak clustering algorithm to reduce the time series data of wind and solar power output and load to generate multiple typical operating scenarios and their corresponding probability weights, and outputs a multi-scenario time series dataset.

[0051] The sensitivity analysis module inputs multi-scenario time-series datasets into the AC / DC power flow calculation process, obtains the voltage amplitude, phase angle, and active and reactive power of each node under various typical scenarios, and calculates the modal sensitivity index of each node to the network loss and voltage stability of the whole network. Based on the sensitivity index, the module sorts and selects the set of nodes to be selected for energy storage configuration.

[0052] The capacity optimization module is based on the set of candidate energy storage configuration nodes, with the goal of minimizing the system's total life cycle cost. It also considers constraints on energy storage charging and discharging power, state of charge, and safe operation of AC / DC power grids. It constructs an optimization configuration objective function and uses a multi-objective particle swarm evolution strategy to solve the objective function, thereby obtaining the preliminary configuration power and capacity of the energy storage device.

[0053] The dynamic verification module substitutes the initial power and capacity configuration into the electromagnetic transient simulation process of the AC / DC power grid, simulates the dynamic response characteristics under the preset fault set, and outputs the final optimized configuration scheme if the transient response meets the safety and stability requirements; otherwise, it corrects the constraints based on the transient response deviation and re-triggers the capacity optimization module.

[0054] The data acquisition module obtains topology data of AC / DC power grids, historical operating data, and technical and economic parameters of energy storage devices to form a basic dataset, which specifically includes:

[0055] The construction process of the basic dataset is as follows: First, the topology data of the AC and DC power grids is exported from the energy management system database of the power grid dispatch center. This topology data specifically includes the impedance parameters, ground susceptance, and rated voltage level of AC lines; the resistance value and rated current of DC lines; the converter transformer capacity and bridge arm reactance value of converter stations; and the numbering and connection relationship of each bus node.

[0056] Secondly, historical operating data of the area to be optimized over the past three consecutive full calendar years are retrieved from the historical server of the Supervisory Control and Data Acquisition (SCADA) system. This historical operating data includes: active power output sequences of photovoltaic power plants at 15-minute intervals uploaded by photovoltaic inverter monitoring units; active power output sequences of wind farms at 15-minute intervals uploaded by wind turbine monitoring systems; and active and reactive load data of each node of the power grid at 15-minute intervals collected by the monitoring and control devices of each substation.

[0057] Finally, the technical and economic parameters of the energy storage equipment are obtained from the technical specifications provided by the equipment supplier and historical price information from the electricity market trading platform. These technical and economic parameters include: rated power cost, rated capacity cost, charge and discharge efficiency, cycle life, operation and maintenance cost, and the maximum charge and discharge rate of the power conversion system. The collected topology data, historical operating data, and technical and economic parameters are associated and formatted according to a unified grid node numbering rule to form a basic dataset for subsequent optimization configuration calculations.

[0058] In the scene generation module, based on historical operational data in the base dataset, the density peak clustering algorithm is used to reduce the time-series data of wind and solar power output and load to generate multiple typical operational scenes and their corresponding probability weights, outputting a multi-scene time-series dataset, specifically including:

[0059] First, the historical operational data in the basic dataset is preprocessed. Active power output sequences for photovoltaic power plants, wind farms, and load active and reactive power sequences for each node are extracted from the basic dataset every 15 minutes. All time-series data are aligned according to a unified timestamp to form an original time-series sample set containing multiple complete daily cycles. For each sample in the original time-series sample set, deviation standardization is used to linearly transform the data to the zero-to-one range, eliminating the influence of dimensions. Simultaneously, abnormal fluctuations caused by communication interruptions or measurement errors, such as power abruptly dropping to zero or power values ​​exceeding twice the rated capacity, are identified and removed. The data processed in the above way forms a standardized time-series sample set.

[0060] Secondly, the morphological similarity between any two samples in the standardized time-series sample set is calculated. Each sample in the standardized time-series sample set is considered as a multi-dimensional time-series vector, which simultaneously contains the wind power output, photovoltaic power output, and load values ​​at that moment. For any two multi-dimensional time-series vectors, the dynamic time bending algorithm is used to calculate their similarity. The specific implementation of this algorithm is as follows: a two-dimensional matrix is ​​constructed, where the number of rows and columns of the matrix correspond to the time series lengths of the two vectors, respectively; starting from the starting point of the matrix, an optimal bending path is found that satisfies the boundary conditions (from the starting point to the ending point), continuity (each step can only move to the right, down, or down-right), and monotonicity (the path monotonically increases with time), such that the sum of the absolute differences between the two vector elements corresponding to all points traversed by the path is minimized; this minimum sum of absolute differences is defined as the dynamic time bending distance between the two multi-dimensional time-series vectors. By iterating through all pairwise combinations of samples in the standardized time series sample set, the calculated dynamic time curvature distance is filled into the corresponding position of the matrix to construct a real symmetric distance matrix. Each element in this distance matrix is ​​used to characterize the degree of morphological similarity between two corresponding time series samples under the condition that the time axis allows nonlinear scaling. The smaller the distance, the closer the morphology.

[0061] Next, cluster centers are determined based on the distance matrix. For each time series sample in the standardized time series sample set, its local density is first calculated. The local density is calculated by counting the number of other samples whose dynamic time curvature distance from the sample is less than a preset cutoff distance, and this number is taken as the local density of that sample. Then, for the sample with the highest local density, its relative distance is defined as the maximum dynamic time curvature distance between that sample and all other samples; for the remaining samples, their relative distance is defined as the minimum dynamic time curvature distance between that sample and all samples with higher local densities. For each time series sample, the product of its local density and relative distance is calculated. All samples are sorted according to the magnitude of this product value, and a decision graph is plotted with local density as the x-axis and relative distance as the y-axis. In the decision graph, several points located in the upper right region with a product value significantly greater than other samples are selected as cluster centers.

[0062] Finally, typical operating scenarios are formed and probability weights are assigned. For each remaining time-series sample other than the cluster centers, the dynamic time curvature distance between it and each cluster center is calculated, and it is assigned to the cluster represented by the cluster center with the smallest distance. All the original time-series samples in each cluster constitute a typical operating scenario. The number of original time-series samples in each cluster is counted, and the proportion of this number to the total number of samples in the standardized time-series sample set is calculated. This proportion is used as the probability weight of the typical operating scenario corresponding to that cluster. All typical operating scenarios and their corresponding probability weights are combined to form the final multi-scenario time-series dataset for subsequent power flow calculation and analysis.

[0063] First, a multi-dimensional time-series vector set is constructed. Data from the standardized time-series sample set are aligned according to timestamps. For any given sampling time, the wind power output, photovoltaic power output, and active power load values ​​at that time are combined to form a three-dimensional time-series vector. All three-dimensional vectors within the entire sampling period are arranged in chronological order to constitute the multi-dimensional time-series vector set for that sample.

[0064] Next, the optimal bending path and cumulative distance are calculated. For any two different multidimensional time-series vector groups, denoted as the first vector group and the second vector group, a two-dimensional cumulative distance matrix is ​​constructed. The number of rows in this matrix is ​​equal to the length of the first vector group, and the number of columns is equal to the length of the second vector group. Starting from the first element in the upper left corner of the matrix, the cumulative distance at each position is calculated row by row and column by column. For the position with coordinates i in the i-th row and j in the j-th column of the matrix, the cumulative distance is calculated as follows: calculate the Euclidean distance between the i-th vector in the first vector group and the j-th vector in the second vector group, and add this Euclidean distance to the minimum of the cumulative distances of the three adjacent positions (left position, top position, and upper left corner position) to obtain the cumulative distance at that position. After traversing the entire matrix according to this recursive relationship, the value of the last element in the lower right corner of the matrix is ​​the minimum cumulative distance between the two multidimensional time-series vector groups. This minimum cumulative distance is defined as the dynamic time bending distance between the two.

[0065] Finally, the distance matrix is ​​assembled. After calculating the dynamic time-warping distances between all pairs of samples in the standardized time-series sample set, the distance values ​​are filled into a matrix with the same number of rows and columns as the total number of samples, according to the original sample numbers. The elements on the diagonal of the matrix are set to zero, and the element in the m-th row and n-th column represents the dynamic time-warping distance between the m-th sample and the n-th sample. Since the dynamic time-warping distance is symmetric, this matrix is ​​a real symmetric matrix, and the distance matrix is ​​now constructed.

[0066] In the sensitivity analysis module, multi-scenario time-series datasets are input into the AC / DC power flow calculation process to obtain the voltage amplitude, phase angle, and active and reactive power of each node under various typical scenarios. The module also calculates the modal sensitivity index of each node to network losses and voltage stability. Based on the sensitivity index, a set of candidate energy storage configuration nodes is selected, specifically including:

[0067] First, multi-scenario power flow calculations are performed. The generated multi-scenario time-series datasets are sequentially substituted into the AC / DC power flow calculation program. Under each typical operating scenario, the power flow equations of the AC / DC system are solved using an alternating iterative method. After the power flow calculation converges, the voltage magnitude, voltage phase angle, and active and reactive power injected into each node in the AC network under that scenario are recorded. After traversing all typical operating scenarios, a time-series dataset containing all node electrical quantities is obtained.

[0068] Secondly, the Jacobian matrix is ​​extracted and analyzed. For each typical operating scenario, the polar coordinate form of the Jacobian matrix of the AC power grid is extracted from the convergence point of the power flow calculation. Eigenvalue decomposition is performed on this Jacobian matrix to obtain a series of eigenvalues ​​and their corresponding left and right eigenvectors. Among all the obtained eigenvalues, several eigenvalues ​​with the smallest magnitudes are selected as dominant eigenvalues. The modes corresponding to these dominant eigenvalues ​​are strongly correlated with the changes in network losses and voltage stability of the system. Among them, eigenvalues ​​with smaller absolute values ​​of real parts are mainly associated with voltage stability, while eigenvalues ​​whose real and imaginary parts combine to affect active power loss are mainly associated with network losses.

[0069] Next, the degree of participation of node injection power in each dominant eigenvalue is calculated. For each selected dominant eigenvalue, the participation factor of each node in the system for that eigenvalue is calculated. The specific calculation method for the participation factor is as follows: take the left and right eigenvectors corresponding to the dominant eigenvalue, multiply the element in the left eigenvector corresponding to the node by the element in the right eigenvector corresponding to the node, and the absolute value of the product is the participation factor of that node for that dominant eigenvalue. This participation factor quantifies the degree of excitation of the eigenvalue mode by the change in node injection power.

[0070] Then, the participating factors are normalized to form sensitivity values. For a set of dominant eigenvalues ​​related to network loss, the participating factors of each node for all eigenvalues ​​in the set are summed. The maximum value is found among the sums of all nodes, and the sum of each node is divided by this maximum value to obtain a value between 0 and 1. This value is the first sensitivity value characterizing the node's impact on the overall network loss. Similarly, for a set of dominant eigenvalues ​​related to voltage stability, the same summation and maximum value normalization operations are used to obtain the second sensitivity value characterizing the node's impact on voltage stability.

[0071] Finally, sensitivity indices are integrated and candidate nodes are selected. A first weighting coefficient and a second weighting coefficient are assigned to each node. The first sensitivity value is multiplied by the first weighting coefficient, and then added to the second sensitivity value multiplied by the second weighting coefficient to obtain the node's comprehensive modal sensitivity index. The first and second weighting coefficients are set according to the actual operation requirements of the power grid, with values ​​ranging from 0 to 1, and their sum is 1. All nodes in the power grid are sorted from highest to lowest according to the comprehensive modal sensitivity index values. The top 20% or a predetermined number of nodes are selected to form a candidate energy storage configuration node set for subsequent capacity optimization modules.

[0072] The specific implementation method for normalizing the participating factors to obtain the first sensitivity value and the second sensitivity value is as follows:

[0073] First, the perturbation method is used to solve the ratio between the injected power at each node and the change in the eigenvalue. For each dominant eigenvalue, based on the convergence point of the power flow calculation, a small positive perturbation, such as an increase of 0.01 MW, is applied to the injected active power of a single node. Keeping the injected power of other nodes constant, the power flow calculation is repeated to solve for the new Jacobian matrix and its eigenvalues, and the change in the dominant eigenvalue is recorded. Dividing the change in the dominant eigenvalue by the perturbation in the injected power yields the first-order sensitivity coefficient of the node's injected power to the dominant eigenvalue.

[0074] Secondly, the original participation factors are constructed. The absolute value of the first-order sensitivity coefficient obtained above is taken as the original participation factor of the node for that dominant eigenvalue. Following this method, perturbations are applied to each node in turn to obtain the original participation factor of each node under the current dominant eigenvalue. After the calculation of a dominant eigenvalue is completed, the above perturbation process is repeated for the next dominant eigenvalue until all dominant eigenvalues ​​have been processed.

[0075] Next, perform intra-group summation. Based on the physical properties of the dominant eigenvalues, all eigenvalues ​​are divided into a network loss correlation group and a voltage stability correlation group. For the network loss correlation group, the original participation factors of each node under all intra-group eigenvalues ​​are summed to obtain a sum value. For the voltage stability correlation group, the summation is also performed for each node to obtain another sum value.

[0076] Finally, maximum value normalization is performed. In the network loss correlation group, the maximum value among the sums of all nodes is found. The sum of each node is divided by this maximum value, and the result is defined as the first sensitivity value of that node. In the voltage stability correlation group, the maximum value among the sums of all nodes is also found, and the sum of each node is divided by this maximum value, and the result is defined as the second sensitivity value of that node. After the above normalization process, the first and second sensitivity values ​​of all nodes are mapped to the interval between 0 and 1, which facilitates subsequent weighted fusion and ranking comparison.

[0077] In the capacity optimization module, based on the set of candidate energy storage configuration nodes, and with the goal of minimizing the system's total lifecycle cost, while considering constraints on energy storage charging and discharging power, state of charge, and AC / DC grid safety operation, an optimization objective function is constructed. A multi-objective particle swarm optimization strategy is then used to solve the objective function, yielding the preliminary configuration power and capacity of the energy storage devices. Specifically, this includes:

[0078] First, the particle swarm is initialized. Each node in the set of candidate energy storage configuration nodes is taken as the potential access location for energy storage devices. The rated power and rated capacity of the energy storage devices that can be installed at each candidate node are set as variables to be optimized. For a system with M candidate nodes, the position vector of each particle consists of 2M dimensions, where the first M dimensions correspond to the rated power of each candidate node, and the last M dimensions correspond to the rated capacity of each candidate node. The swarm size is set to N, i.e., containing N particles. Within the allowable range of rated power and rated capacity, initial values ​​are assigned to each dimension of each particle using random number generation, forming the initial position matrix of the swarm. At the same time, within the allowable range of velocity, the velocity vector of each particle is randomly initialized, and the dimension of the velocity vector is the same as that of the position vector. In addition, the current position of each particle is initialized to the individual historical best position of that particle, and after calculating the fitness of all particles, the position of the particle with the highest fitness is initialized as the global best position of the swarm.

[0079] Secondly, the fitness value of each particle is calculated. For each particle in the population, the energy storage configuration scheme it represents (i.e., the rated power and rated capacity of each candidate node) is substituted into the multi-scenario time-series dataset. For each typical operating scenario, a full-time time-series operation simulation is conducted, considering energy storage charging and discharging power constraints, state of charge constraints, and AC / DC grid safety operation constraints. The time scale of the time-series operation simulation is one year, with a step size of fifteen minutes. During the simulation, the energy storage charging and discharging strategy is optimized based on the renewable energy output and load levels at each moment to minimize the system's operating cost. The operating cost includes: the cost of purchasing electricity from the upstream grid, grid loss costs, and wind and solar curtailment penalty costs. After completing the simulation of all typical scenarios, the expected annual operating cost over the entire life cycle of the system is obtained by weighting and summing the values ​​according to the probability weights of each scenario. The life-cycle cost consists of two parts: investment cost and operating cost. The investment cost includes the power cost (rated power multiplied by unit power cost) and capacity cost (rated capacity multiplied by unit capacity cost) of the energy storage equipment, and the one-time investment is discounted to an equivalent annual value using a discount rate. The equivalent annual investment cost is added to the annual operating cost to obtain the total equivalent annual cost over the system's entire lifecycle. The reciprocal of the total cost is defined as the fitness value of the particle; that is, the lower the total cost, the higher the fitness.

[0080] Next, update the individual historical best position and the population global best position. For each particle, compare its current fitness value with the fitness value corresponding to its individual historical best position. If the current fitness value is greater, update the particle's position to its new individual historical best position. After updating the individual historical best positions of all particles, iterate through the fitness values ​​of all particles' individual historical best positions, find the maximum value, and if this maximum value is greater than the fitness value corresponding to the current population global best position, update the individual historical best position corresponding to this maximum value to the new population global best position.

[0081] Then, the particle's velocity and position are iteratively updated. In each iteration, the inertia weight is first calculated based on the current iteration number. Let the total number of iterations be... The current iteration number is The initial value of the inertia weight is The termination value is Then the inertial weight w decreases nonlinearly according to the following formula:

[0082] ;

[0083] In the formula, The value is 0.9. The value is set to 0.4, and the exponent 2 is used to achieve non-linear changes, resulting in a larger inertia weight in the early stages of iteration, which is beneficial for global search, and a smaller inertia weight in the later stages of iteration, which is beneficial for fine-grained local search. Secondly, two learning factors are determined, denoted as the first learning factor. Second learning factor Both are set to 2.0, which are used to adjust the step size of the particle's flight towards the individual's historical best position and the global best position.

[0084] Next, based on the current inertia weight, learning factor, and the relationship between the particle's current position and optimal position, the velocity update of each particle in each dimension is calculated. For the i-th particle... Dimension, its speed The update method is as follows: multiply the inertia weight by the current velocity, add the first learning factor multiplied by a random number between 0 and 1, then multiply by the difference between the individual's historical best position and the current position, add the second learning factor multiplied by another random number between 0 and 1, then multiply by the difference between the global best position and the current position. After the velocity update, if it exceeds the preset velocity limit, it is restricted to the boundary value. The position update method is as follows: add the updated velocity to the particle's current position to obtain the new position. Similarly, if the new position exceeds the allowable range of rated power or rated capacity, it is restricted to the boundary value. After the velocity and position of all particles have been updated, one iteration is completed, generating a new generation of particle population.

[0085] Finally, the iteration termination condition is determined. After each iteration, it is checked whether the preset maximum number of iterations T has been reached, or whether the fitness value of the global optimum has remained unchanged for multiple consecutive iterations (e.g., 50 consecutive iterations). If either condition is met, the iteration process is terminated, and the values ​​of each dimension corresponding to the current global optimum position of the population, i.e., the rated power and rated capacity of each candidate node, are used as the initial configuration power and capacity output of the energy storage device. If the termination condition is not met, the process returns to the fitness calculation step and continues to the next iteration.

[0086] In the dynamic verification module, the initial power and capacity configurations are substituted into the electromagnetic transient simulation process of the AC / DC power grid to simulate the dynamic response characteristics under a preset fault set. If the transient response meets the safety and stability requirements, the final optimized configuration scheme is output; otherwise, the constraints are corrected based on the transient response deviation, and the capacity optimization module is retried. Specifically, this includes:

[0087] First, an electromagnetic transient simulation model is constructed and preset faults are applied. The initial power and capacity configuration of the energy storage device are used as parameters of the energy storage system and substituted into the electromagnetic transient simulation program of the AC / DC power grid. In this simulation program, a preset fault set including AC short-circuit faults and DC blocking faults is established in advance. The AC short-circuit faults include a three-phase metallic short-circuit fault set at a critical AC bus, which is cleared after 0.1 seconds. The DC blocking faults include a unipolar blocking fault simulating the converter station due to the loss of trigger pulses. After the simulation starts, various faults are applied one by one in the order of the fault set to simulate the electromagnetic transient response process of the power grid after being subjected to a large disturbance. The simulation step size is set to 50 microseconds, and the total simulation time is set to 5 seconds to ensure that the entire dynamic trajectory of the fault occurrence, duration, and clearing is completely recorded.

[0088] Secondly, the dynamic response trajectories of key nodes are extracted and compared with safety and stability thresholds. After each fault simulation, the voltage amplitude of key nodes and the active power transmitted by key converter stations over time are extracted from the simulation results. The extracted voltage trajectories are compared with preset voltage safety and stability thresholds, which include the minimum allowable voltage drop (e.g., 0.75 times the rated voltage) and the maximum allowable voltage recovery time (e.g., recovery to above 0.95 times the rated voltage within 2 seconds after fault clearance). Simultaneously, the extracted power trajectories are compared with preset power safety and stability thresholds, which include the maximum allowable converter station overload multiple (e.g., 1.2 times the rated power) and the maximum allowable power oscillation amplitude. The dynamic responses under all faults are checked one by one to ensure that all the above threshold requirements are met simultaneously.

[0089] Next, the verification results are evaluated, and an output or correction is determined. If the voltage and power trajectories under all preset faults are within the range of the safety and stability threshold, the preliminary configuration scheme is deemed to meet the dynamic safety and stability requirements, and the preliminary configuration power and capacity are directly output as the final optimized configuration scheme. If the voltage trajectory under any fault is lower than the minimum allowable voltage drop, or the power trajectory exceeds the maximum allowable overload multiple of the converter station, the preliminary configuration scheme is deemed not to meet the dynamic safety and stability requirements.

[0090] Finally, the response deviation is calculated and the constraints are adjusted. For fault scenarios that do not meet the requirements, the deviation between the dynamic response and the safety and stability threshold is calculated. Specifically, for voltage dip exceeding the limit, the duration for which the voltage trajectory is below the minimum allowable voltage dip and the maximum voltage deviation depth are statistically analyzed; for power overload, the duration for which the power trajectory exceeds the maximum allowable overload multiple and the maximum overload multiple are statistically analyzed. Based on the magnitude of these deviations, the constraint boundaries in the optimization objective function are adjusted in reverse. If the voltage dip exceeds the limit, the voltage stability constraint boundaries used in the optimization objective function are tightened proportionally to the deviation, for example, by increasing the lower limit of the allowable node voltage. If the power overload exceeds the limit, the network loss constraint boundaries or converter station capacity constraint boundaries are tightened proportionally to the deviation. The adjusted constraints are used as new inputs to re-trigger the capacity optimization solution process in the capacity optimization module for a new round of iterative calculations until the simulation verification passes or the preset maximum number of iterations is reached.

[0091] The working principle of this invention is as follows: First, it acquires topology data, historical operating data, and technical and economic parameters of energy storage devices from the AC / DC power grid to form a basic dataset. Second, based on the historical operating data in the basic dataset, it uses a density peak clustering algorithm to reduce the time-series data of wind and solar power output and load, generating multiple typical operating scenarios and corresponding probability weights, and outputting a multi-scenario time-series dataset. Third, it inputs the multi-scenario time-series dataset into the AC / DC power flow calculation process to obtain the voltage amplitude, phase angle, and active and reactive power of each node under each typical scenario. Based on the power flow calculation results, it extracts the Jacobian matrix for modal decoupling analysis, calculates the modal sensitivity index of each node to the overall network loss and voltage stability, and selects a set of candidate energy storage configuration nodes based on the sensitivity index. Finally, the capacity optimization module, based on the set of candidate energy storage configuration nodes, optimizes the system over its entire lifespan. The optimization objective is to minimize cycle cost. Considering constraints on energy storage charging and discharging power, state of charge, and AC / DC grid safety operation, an optimization configuration objective function is constructed. A multi-objective particle swarm optimization strategy is used to solve the objective function, yielding the initial configuration power and capacity of the energy storage device. This initial configuration power and capacity are then substituted into the electromagnetic transient simulation process of the AC / DC grid to simulate the dynamic response characteristics under a preset fault set. The voltage trajectories of key nodes and the power trajectories of converter stations are extracted and compared with preset safety and stability thresholds. If the transient response meets the safety and stability requirements, the final optimized configuration scheme is output; otherwise, the constraint boundaries in the optimization configuration objective function are adjusted in reverse according to the transient response deviation. The adjusted constraint conditions are then re-triggered to trigger the capacity optimization module for iterative solution until the final optimized configuration scheme that meets the dynamic safety and stability requirements is obtained.

[0092] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A system for optimizing configuration of energy storage devices based on AC-DC grid, characterized in that, include: The data acquisition module acquires topology data of AC / DC power grids, historical operation data, and technical and economic parameters of energy storage devices to form a basic dataset; The scene generation module, based on historical operating data in the basic dataset, uses the density peak clustering algorithm to reduce the time series data of wind and solar power output and load to generate multiple typical operating scenarios and their corresponding probability weights, and outputs a multi-scenario time series dataset. The sensitivity analysis module inputs multi-scenario time-series datasets into the AC / DC power flow calculation process to obtain the voltage amplitude, phase angle, and active and reactive power of each node under various typical scenarios. It also calculates the modal sensitivity index of each node to the overall network loss and voltage stability. Based on the sensitivity index, it sorts and selects a set of nodes for potential energy storage configuration. Specifically, the calculation of the modal sensitivity index of each node to the overall network loss and voltage stability includes: Based on the power flow calculation results under various typical scenarios, the Jacobian matrix of the AC power grid is extracted, and modal decoupling analysis is performed on the Jacobian matrix to obtain several dominant eigenvalues ​​that are strongly correlated with network loss and voltage stability. The participation factor of the injected power of each node on the dominant eigenvalue is calculated, and the participation factor is normalized to obtain the first sensitivity value and the second sensitivity value of each node on the impact of network loss and voltage stability. The first sensitivity value and the second sensitivity value are weighted and fused according to a preset weight ratio to form a comprehensive modal sensitivity index for each node. The nodes are sorted from high to low according to the corresponding comprehensive modal sensitivity index, and the nodes with the highest ranking are selected as the set of candidate energy storage configuration nodes. The normalization of the participating factors to obtain the first sensitivity value characterizing the impact of each node on network loss and the second sensitivity value characterizing the impact on voltage stability specifically includes: For each dominant eigenvalue, a small perturbation is applied to the injected power of each node obtained from the power flow calculation, and the ratio between the change in injected power of each node and the change in dominant eigenvalue is solved by the perturbation method. The absolute value of the ratio is taken as the original participation factor of the corresponding node with respect to the current dominant eigenvalue, and the original participation factors of the same node under all dominant eigenvalues ​​are summed. The summation results of the dominant feature root group of network loss association are normalized by the maximum value to obtain the first sensitivity value of each node to the impact of network loss. The summation results of the dominant eigenvalues ​​related to voltage stability are normalized to obtain the second sensitivity value of each node to the influence of voltage stability. The capacity optimization module is based on the set of candidate energy storage configuration nodes, with the goal of minimizing the system's total life cycle cost. It also considers constraints on energy storage charging and discharging power, state of charge, and safe operation of AC / DC power grids. It constructs an optimization configuration objective function and uses a multi-objective particle swarm evolution strategy to solve the objective function, thereby obtaining the preliminary configuration power and capacity of the energy storage device. The dynamic verification module substitutes the initial power and capacity configuration into the electromagnetic transient simulation process of the AC / DC power grid, simulates the dynamic response characteristics under the preset fault set, and outputs the final optimized configuration scheme if the transient response meets the safety and stability requirements; otherwise, it corrects the constraints based on the transient response deviation and re-triggers the capacity optimization module. 2.The system of claim 1, wherein, The output multi-scenario time-series dataset specifically includes: The wind and solar power output time series data and load time series data in the historical operation data are normalized and abnormal fluctuation points are removed to form a standardized time series sample set. Based on a standardized time series sample set, the similarity between any two time series samples is calculated using dynamic time curvature distance, and a distance matrix is ​​constructed. Based on the distance matrix, determine the local density of each time series sample and the relative distance between the time series sample and the higher density sample, and draw a decision map based on the product of local density and relative distance. Select the time series sample whose product value is located in the upper right region of the decision map as the cluster center. The remaining time series samples are assigned to the nearest cluster centers to form multiple typical operating scenarios. The probability weights corresponding to each typical operating scenario are calculated based on the proportion of the original time series samples covered by each cluster center to the total number of samples, and a multi-scenario time series dataset is output. 3.The system of claim 2, wherein, The construction of the distance matrix specifically includes: The wind and solar power output time series data and load time series data in the standardized time series sample set are aligned by timestamp, and then a multi-dimensional time series vector group is formed after alignment. For any two multidimensional time series vector groups, a dynamic time warping algorithm is used to calculate an optimal warping path that minimizes the cumulative distance between the two vector groups, while satisfying the boundary conditions and continuity constraints. The corresponding cumulative distance is then used as the dynamic time warping distance between the two vector groups. Based on the dynamic time bending distance of all pairwise combinations, a real symmetric distance matrix is ​​constructed. Each element in the distance matrix is ​​used to characterize the morphological similarity of the corresponding two time series samples under nonlinear time axis scaling. 4.The system of claim 1, wherein, The preliminary configuration power and capacity of the energy storage device are obtained, specifically including: Each node in the set of candidate energy storage configuration nodes is taken as the location to be connected to the energy storage device. The rated power and rated capacity of the energy storage device are taken as the variables to be optimized. A population containing multiple particles is initialized, and each particle represents an energy storage configuration scheme. Each energy storage configuration scheme is substituted into the AC / DC power grid time-series operation simulation under various typical operating scenarios. The sum of investment cost and operating cost over the entire life cycle of the system is calculated, and the reciprocal of the corresponding cost value is used as the fitness value of the particle. Based on the fitness value of each particle, update the individual historical best position of the particle and the global best position of the population, and iteratively update the position and velocity of the particles according to the inertia weight and learning factor in the particle swarm evolution strategy. When the number of iterations reaches the preset upper limit or the global optimal fitness value remains unchanged for multiple consecutive times, the iteration is terminated, and the rated power and rated capacity corresponding to the current global optimal position are used as the initial configuration power and capacity output of the energy storage device.

5. The system for optimal configuration of energy storage devices based on AC / DC grid according to claim 4, characterized in that, The iterative update of the particle's position and velocity specifically includes: Obtain the current iteration number, and dynamically calculate the value of the inertia weight based on the ratio of the current iteration number to the total iteration number, so that the inertia weight decreases non-linearly from the initial value to the final value as the iteration progresses; Compare the current fitness value of each particle with the fitness value of its historical best position. If the current fitness value is better, update the current position to the historical best position of the corresponding particle. At the same time, compare the fitness values ​​of the historical best positions of all particles and update the position with the highest fitness value to the global best position of the population. Based on the current inertia weight and two learning factors, and combining the difference between each particle's current position and its individual historical best position and the population's global best position, the movement speed of the corresponding particle in the solution space is calculated, and the particle's position is updated according to the movement speed to generate a new generation of energy storage configuration scheme. 6.The system of claim 1, wherein, The electromagnetic transient simulation process, which substitutes the initial power and capacity configuration into the AC / DC power grid, simulates the dynamic response characteristics under a preset fault set, specifically includes: The initial power and capacity configurations are used as parameters for the energy storage device. They are then substituted into the electromagnetic transient simulation program of the AC / DC power grid. Various AC short-circuit faults and DC blocking faults are applied one by one in the preset fault set to simulate the dynamic response process of the power grid and extract the voltage trajectory of key nodes and the power trajectory of the converter station under each fault. The extracted voltage and power trajectories are compared with preset safety and stability thresholds. If the dynamic response under all faults meets the threshold requirements, the preliminary power and capacity configuration is output as the final optimized configuration scheme. If the dynamic response under any fault does not meet the threshold requirement, the deviation between the dynamic response under the corresponding fault and the safety and stability threshold is calculated. Based on the magnitude of the deviation, the voltage stability constraint boundary or network loss constraint boundary in the objective function is adjusted in reverse. The adjusted constraint conditions are then re-input into the capacity optimization module for iterative solution.

Images