Park power grid source network load storage collaborative response characteristic optimization evaluation method and system

Abstract

The invention relates to the technical field of power grid system collaborative optimization, in particular to a park power grid source grid load storage collaborative response characteristic optimization evaluation method and system, and the method comprises the steps: obtaining a multi-dimensional standard data set of a to-be-optimized and evaluated park power grid in a certain time window; the multi-dimensional standard data set is input to a cooperative control optimization model, an optimized cooperative control parameter combination is obtained, and the cooperative control optimization model comprises a feature extraction model, a source network load storage cooperative degree output model and a multi-target optimization model. According to the method, the global feature vector of the multi-dimensional standard data set is extracted, the dynamic characteristics of the source-grid-load-storage cooperative operation are comprehensively represented, the load permeability, the load fluctuation stabilizing amount and the consumption capacity of the distributed power supply are maximized as the target, the improved particle swarm optimization algorithm is combined for iterative optimization, and the load-grid-load-storage cooperative operation performance is improved. And a cooperative control parameter combination corresponding to target maximization is obtained, deep cooperation and dynamic optimization control among park power grid multi-element systems is realized, and the stability and operation efficiency of a park energy system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid system collaborative optimization, and is a park power grid source-grid-load-storage collaborative response characteristic optimization evaluation method and system. BACKGROUND

[0002] The "source-grid-load-storage" collaborative management of the park power grid is to coordinate the operating states of the power source (distributed energy, etc.), the power grid (distribution network), the load (various electrical equipment), and the energy storage (energy storage system), so as to achieve the goals of energy efficient utilization, reliability improvement, and low carbonization. Therefore, with the development of smart grid and distributed energy technology, the collaborative operation of the distributed power source, the power grid, the load, and the energy storage device in the park power grid has become the key to improving energy utilization efficiency and power grid stability.

[0003] However, the existing park power grid source-grid-load-storage collaborative management technology faces the following challenges:

[0004] Insufficient multi-source data fusion: The distributed power output, load demand, and energy storage state data are collected in a scattered manner, and lack unified spatio-temporal alignment and standardized processing, resulting in low data utilization.

[0005] One-sided feature analysis: Traditional methods cannot effectively fuse time-domain transient characteristics and frequency-domain periodic characteristics, and cannot comprehensively represent the dynamic characteristics of the source-grid-load-storage collaborative operation.

[0006] Incomplete evaluation system: Lack of multi-dimensional quantitative indicators for comprehensive evaluation of power source and load matching degree, energy storage regulation capacity, and power grid accommodation capacity, making it difficult to accurately locate the operation bottleneck.

[0007] Optimization strategy blindness: The generation of optimization control parameter combinations lacks data-driven intelligent algorithm support, and cannot adapt to the dynamic changes of the park power grid operating state.

[0008] Lack of closed-loop feedback: The effect after the strategy execution lacks real-time feedback and iterative optimization mechanism, making it difficult to continuously improve the collaborative control effect.

[0009] Therefore, the existing technology cannot effectively solve the technical bottlenecks of spatio-temporal fusion of multi-source data, cross-domain feature extraction, multi-dimensional evaluation, and self-adaptive optimization, resulting in insufficient evaluation accuracy of the park power grid source-grid-load-storage collaborative response characteristics and limited effectiveness of the optimization strategy. SUMMARY

[0010] The present application provides a park power grid source-grid-load-storage collaborative response characteristic optimization evaluation method and system, which overcomes the deficiencies of the existing technology and effectively solves the problems of insufficient evaluation accuracy and limited effectiveness of the optimization strategy in the existing park power grid source-grid-load-storage collaborative response characteristic evaluation method.

[0011] One of the technical solutions of the present application is realized by the following measures: a park power grid source network load storage collaborative response characteristic optimization evaluation method, comprising:

[0012] Obtain a multi-dimensional standard data set of the park power grid to be optimized and evaluated in a certain time window, wherein the multi-dimensional standard data set includes multi-dimensional data of a plurality of monitoring points at a plurality of sampling times, and the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge;

[0013] Input the multi-dimensional standard data set into a collaborative control optimization model to obtain an optimized collaborative control parameter combination, wherein the collaborative control optimization model is obtained by training a plurality of samples, each sample including multi-dimensional standard data set and corresponding label data of collaborative control parameter combination, and the collaborative control optimization model includes a feature extraction model, a source network load storage collaboration degree output model, and a multi-objective optimization model;

[0014] The feature extraction model extracts a global feature vector in the multi-dimensional standard data set;

[0015] The response evaluation model predicts the load penetration rate, load fluctuation suppression amount, and distributed power consumption capacity under the current operating state of the park power grid to be optimized and evaluated based on the global feature vector in the multi-dimensional standard data set;

[0016] The multi-objective optimization model maximizes the load penetration rate, load fluctuation suppression amount, and distributed power consumption capacity, and iteratively optimizes the improved particle swarm algorithm to obtain the corresponding collaborative control parameter combination when the target is maximized, wherein the collaborative control parameter combination includes active power, reactive power, voltage, current, and energy storage state of charge.

[0017] The following is a further optimization or / and improvement of the above-mentioned technical solutions of the present application:

[0018] The above-mentioned feature extraction model extracts a global feature vector in the multi-dimensional standard data set, comprising:

[0019] The multi-dimensional standard data set is represented as an original tensor, and the original tensor is input into a hierarchical residual dense network to extract transient features, thereby obtaining a time domain tensor;

[0020] Discrete Fourier transform is performed on the time domain tensor to extract frequency domain features of load periodic fluctuations, and the frequency domain features are input into a hierarchical residual dense network to obtain a frequency domain tensor;

[0021] The global average pooling of the time domain tensor and the frequency domain tensor is performed by using a cross-domain attention fusion mechanism to obtain a global feature vector;

[0022]

[0023] Among them, is the time-domain tensor of the i-th monitoring point at the k-th sampling time, is the frequency-domain tensor of the i-th monitoring point at the k-th sampling time, N t and N f are respectively the number of sampling points of the time-domain tensor sequence and the frequency-domain tensor sequence, ω t (i) and ω f (i) are respectively attention weighting coefficients, monitoring point i∈[1,M],M is the total number of monitoring points.

[0024] The above response evaluation model predicts three indicators, i.e., load penetration rate, load fluctuation suppression amount and distributed power consumption capacity, in the current operating state of the park power grid based on the global feature vector in the multi-dimensional standard data set, including:

[0025] The global feature vector is taken as the digital representation of the current operating state;

[0026] The deep learning network model is used to learn the correlation between the time-frequency domain features in the global feature vector and the three indicators, and the three indicators, i.e., load penetration rate, load fluctuation suppression amount and distributed power consumption capacity, in the current operating state of the park power grid are predicted;

[0027] The load penetration rate, load fluctuation suppression amount and distributed power consumption capacity are taken as the collaborative feature data set x, and the corresponding source-grid-load-storage collaboration degree H(x) is obtained;

[0028] The collaborative degree expectation y of the current operating state of the park power grid is determined, and the loss function is determined as shown below, and the source-grid-load-storage collaboration degree H(x) when the loss function is minimized is output;

[0029]

[0030] The above multi-objective optimization model maximizes the load penetration rate, load fluctuation suppression amount and distributed power consumption capacity, and iteratively optimizes the improved particle swarm algorithm to obtain the corresponding collaborative control parameter combination when the target is maximized, including:

[0031] The load penetration rate, load fluctuation suppression amount and distributed power consumption capacity output by the response evaluation model are input into the optimization degree function to obtain the corresponding optimization degree, wherein the optimization degree formula is as shown below:

[0032]

[0033] wherein, v x is the optimization coefficient of the multi-dimensional evaluation index system, which corresponds to H(x) one by one;

[0034] The active power, the reactive power, the voltage, the current, and the energy storage state of charge are respectively iteratively optimized based on the improved particle swarm algorithm, and the active power, the reactive power, the voltage, the current, and the energy storage state of charge corresponding to the maximum optimization degree are output, thereby forming a coordinated control parameter combination.

[0035] The above-mentioned obtaining the multi-dimensional standard data set of the park power grid to be optimized and evaluated in a certain time window includes:

[0036] In a certain time window, the multi-dimensional data of each monitoring point in the park power grid to be optimized and evaluated at multiple sampling times is obtained, forming a multi-dimensional original data set, and the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge.

[0037] The multi-dimensional original data set is time-domain aligned, and data with a deviation value exceeding three times the standard deviation is filtered by using the 3σ principle, the filtered multi-dimensional original data set is mapped to a numerical interval of [0, 1] through linear transformation, and a multi-dimensional standard data set is formed.

[0038] The above-mentioned also includes issuing and executing the coordinated control parameter combination obtained by the optimization and evaluation as an optimization strategy, and collecting the park power grid operation data after the optimization strategy is executed in real time, and performing optimization and evaluation feedback adjustment.

[0039] The second technical solution of the application is realized by the following measures: a park power grid source-network-load-storage collaborative response characteristic optimization and evaluation system, comprising:

[0040] A data acquisition unit acquires a multi-dimensional standard data set of the park power grid to be optimized and evaluated in a certain time window, wherein the multi-dimensional standard data set includes multi-dimensional data of a plurality of monitoring points at a plurality of sampling times, and the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge.

[0041] An optimization and evaluation unit inputs the multi-dimensional standard data set into a collaborative control optimization model to obtain an optimized coordinated control parameter combination, wherein the collaborative control optimization model is obtained by training a plurality of samples, each sample includes multi-dimensional standard data set and corresponding label data of the coordinated control parameter combination, and the collaborative control optimization model includes a feature extraction model, a source-network-load-storage collaboration degree output model, and a multi-objective optimization model.

[0042] The feature extraction model extracts a global feature vector in the multi-dimensional standard data set.

[0043] A response evaluation model predicts the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity under the current operating state of the park power grid to be optimized and evaluated based on the global feature vector in the multi-dimensional standard data set.

[0044] A multi-objective optimization model is used to maximize the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, and an improved particle swarm algorithm is used for iterative optimization to obtain a corresponding collaborative control parameter combination when the target is maximized, wherein the collaborative control parameter combination includes active power, reactive power, voltage, current and energy storage state of charge.

[0045] The following is a further optimization or / and improvement of the above technical solutions:

[0046] The above also includes a closed-loop execution feedback unit, which issues and executes the collaborative control parameter combination obtained by optimization evaluation as an optimization strategy, and real-time collects park power grid operation data after the execution of the optimization strategy for optimization evaluation feedback adjustment.

[0047] The optimization evaluation unit includes:

[0048] A feature extraction module extracts a global feature vector in a multi-dimensional standard data set using a feature extraction model, including:

[0049] The multi-dimensional standard data set is represented as an original tensor, and the original tensor is input into a hierarchical residual dense network to extract transient features to obtain a time domain tensor;

[0050] The time domain tensor is subjected to a discrete Fourier transform to extract frequency domain features of load periodic fluctuations, and the frequency domain features are input into a hierarchical residual dense network to obtain a frequency domain tensor;

[0051] The time domain tensor and the frequency domain tensor are subjected to global average pooling using a cross-domain attention fusion mechanism to obtain a global feature vector;

[0052]

[0053] wherein, is a time domain tensor of the i-th monitoring point at the k-th sampling time, is a frequency domain tensor of the i-th monitoring point at the k-th sampling time, N t and N f are the number of sampling points of the time domain tensor sequence and the frequency domain tensor sequence, respectively, ω t (i) and ω f (i) are attention weighting coefficients, respectively, monitoring point i∈[1,M], M is the total number of monitoring points;

[0054] A response evaluation module uses a response evaluation model to predict three indicators, i.e., load penetration rate, load fluctuation suppression amount and distributed power consumption capacity, based on the global feature vector in the multi-dimensional standard data set under the current operating state of the park power grid, including:

[0055] The global feature vector is used as a digital representation of the current operating state;

[0056] The deep learning network model is used to learn the correlation between the time-frequency domain features in the global feature vector and the three indexes, and the three indexes, i.e., the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, in the current operation state of the park power grid are predicted;

[0057] The load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity are taken as the collaborative feature data set x, and the corresponding source-grid-load-storage collaboration degree H(x) is obtained;

[0058] The collaborative degree expectation y of the current operation state of the park power grid is determined, and the loss function is determined as shown below, and the source-grid-load-storage collaboration degree H(x) when the loss function is minimum is output;

[0059]

[0060] The multi-objective optimization module uses a multi-objective optimization model to maximize the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, and combines the improved particle swarm algorithm to iteratively optimize to obtain the corresponding collaborative control parameter combination when the target is maximized, including:

[0061] The load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity output by the input response evaluation model are input to the optimization degree function to obtain the corresponding optimization degree, and the optimization degree formula is as shown below:

[0062]

[0063] Wherein, v x is the optimization coefficient of the multi-dimensional evaluation index system, which corresponds to H(x) one by one;

[0064] The improved particle swarm algorithm is used to iteratively optimize the active power, the reactive power, the voltage, the current and the energy storage state of charge, and the active power, the reactive power, the voltage, the current and the energy storage state of charge corresponding to the maximum optimization degree are output to form the collaborative control parameter combination.

[0065] The above data acquisition unit includes a distributed monitoring terminal and a data preprocessing module;

[0066] The distributed monitoring terminal is deployed at the park substation, the distributed power grid connection point, the important load node and the energy storage converter, and in a certain time window, the multi-dimensional data of each monitoring point in the park power grid to be optimized and evaluated at multiple sampling times are acquired to form a multi-dimensional original data set, and the multi-dimensional data includes active power, reactive power, voltage, current and energy storage state of charge;

[0067] A data preprocessing unit performs time domain alignment on a multi-dimensional original data set, filters data with a deviation value exceeding three times a standard deviation using a 3 sigma principle, maps the filtered multi-dimensional original data set to a numerical interval of [0, 1] through linear transformation, and forms a multi-dimensional standard data set.

[0068] The multi-dimensional standard data set includes active power, reactive power, voltage, current, and energy storage state of charge.

[0069] The multi-dimensional standard data set includes active power, reactive power, voltage, current, and energy storage state of charge. Figure 1 A flowchart of a park power grid source network load storage collaborative response characteristic optimization evaluation method provided by the present application is shown.

[0070] A flowchart of a multi-dimensional standard data acquisition method provided by the present application is shown. Figure 2 A flowchart of a multi-dimensional standard data acquisition method provided by the present application is shown.

[0071] A flowchart of another park power grid source network load storage collaborative response characteristic optimization evaluation method provided by the present application is shown. Figure 3 A flowchart of another park power grid source network load storage collaborative response characteristic optimization evaluation method provided by the present application is shown.

[0072] Figure 4 A structure diagram of a park power grid source network load storage collaborative response characteristic optimization evaluation system provided by the present application is shown.

[0073] A structure diagram of a park power grid source network load storage collaborative response characteristic optimization evaluation system provided by the present application is shown. Figure 5 ​​A structure diagram of an optimization evaluation unit provided by the application is shown.

[0074] The application further provides a garden power grid source network load storage collaborative response characteristic optimization evaluation system structure diagram. Figure 6 The application further provides a garden power grid source network load storage collaborative response characteristic optimization evaluation system structure diagram. DETAILED DESCRIPTION

[0075] The application is not limited by the following examples, and the specific implementation can be determined according to the technical scheme of the application and the actual situation.

[0076] Those skilled in the art can understand that, unless specifically stated, the "module" or "unit" in the embodiments of the application refers to a computer program or part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined target, and can be implemented entirely or partially by using software, hardware (such as a processing circuit or a memory) or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the functions of the module or unit.

[0077] In addition, "multiple" in the embodiments of the application refers to two or more, and "first" and "second" are used for differentiation and description, and cannot be understood as implying relative importance.

[0078] Based on this, the technical scheme of the application will be introduced and described below in combination with several examples.

[0079] Embodiment 1: As shown in the accompanying drawings, the embodiment of the application discloses a garden power grid source network load storage collaborative response characteristic optimization evaluation method, which comprises the following steps: Figure 1

[0080] Step S110, a multi-dimensional standard data set of a garden power grid to be optimized and evaluated in a certain time window is acquired, wherein the multi-dimensional standard data set comprises multi-dimensional data of a plurality of monitoring points at a plurality of sampling time points, and the multi-dimensional data comprises active power, reactive power, voltage, current and energy storage state of charge;

[0081] Step S120, the multi-dimensional standard data set is input into a collaborative control optimization model to obtain an optimized collaborative control parameter combination, wherein the collaborative control optimization model is obtained by training a plurality of samples, each sample comprises multi-dimensional standard data set and corresponding label data of the collaborative control parameter combination, and the collaborative control optimization model comprises a feature extraction model, a source network load storage collaboration degree output model and a multi-objective optimization model;

[0082] The feature extraction model extracts a global feature vector in the multi-dimensional standard data set.

[0083] ​In response to the evaluation model, based on the global feature vector in the multi-dimensional standard data set, the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity under the current running state of the to-be-optimized evaluation park power grid are predicted;

[0084] The multi-objective optimization model maximizes the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, and iteratively optimizes the improved particle swarm algorithm to obtain the corresponding collaborative control parameter combination when the target is maximized, wherein the collaborative control parameter combination includes active power, reactive power, voltage, current, and energy storage state of charge.

[0085] The application discloses a park power grid source-network-load-storage collaborative response characteristic optimization evaluation method, establishes a park power grid multi-dimensional standard data set acquisition, global feature vector extraction, multi-objective optimization optimization evaluation process, extracts the global feature vector in the multi-dimensional standard data set, comprehensively represents the dynamic characteristics of the collaborative operation of the source-network-load-storage, provides stable data support for subsequent optimization evaluation, maximizes the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, and iteratively optimizes the improved particle swarm algorithm to obtain the corresponding collaborative control parameter combination when the target is maximized, realizes deep collaboration and dynamic optimization control among the multi-element systems of the park power grid, and significantly improves the stability, flexibility and operation efficiency of the park energy system.

[0086] Embodiment 2: as shown in the accompanying Figure 2 Embodiment 2: as shown in the accompanying

[0087] Step S210, in a certain time window, the multi-dimensional data of each monitoring point in the to-be-optimized evaluation park power grid at a plurality of sampling time points is acquired to form a multi-dimensional original data set, and the multi-dimensional data includes active power, reactive power, voltage, current and energy storage state of charge.

[0088] The multi-dimensional original data set is represented as

[0089] Wherein, D i is the data set of the i th monitoring point i [1, M] M is the total number of monitoring points, t k is the k th sampling time, P i，k is the active power, Q i，k is the reactive power, U i，k is the voltage, I i，k is the current, SOC i，k is the energy storage state of charge, and N is the number of sampling points.

[0090] Step S220, time domain alignment is performed on the multi-dimensional original data set, and data with a deviation value exceeding three times of the standard deviation is filtered by using the 3σ principle. The filtered multi-dimensional original data set is mapped to a numerical interval of [0, 1] through linear transformation to form a multi-dimensional standard data set.

[0091] For example, 2 monitoring points are arranged in the evaluation park power grid to be optimized. In a certain time window, the following multi-dimensional standard data set is obtained, as shown in Table 1:

[0092] Table 1: Multi-dimensional standard data set

[0093] i <![CDATA[t k ]]> P i，k ]]> Q i，k ]] U i，k ]]> I i，k ]] SOC i，k ]]> 1 ​ 0.500 0.286 0.714 0.286 0.667 1 [t2] 0.750 0.571 0.857 0.571 0.800 1 [t3] 1.000 1.000 1.000 1.000 1.000 2 ​ 0.000 0.000 0.000 0.000 0.000 2 [t2] 0.100 0.143 0.143 0.143 0.133 2 [t3] 0.250 0.286 0.286 0.286 0.333

[0094] Embodiment 3: The embodiment of the application is a further optimization of the above-mentioned embodiment, wherein the specific execution process of the optimized collaborative control parameter combination obtained by the collaborative control optimization model according to the input multi-dimensional standard data set includes:

[0095] (I) The feature extraction model extracts the global feature vector in the multi-dimensional standard data set, including:

[0096] (1) The multi-dimensional standard data set is represented as an original tensor, and the original tensor is input into a hierarchical residual dense network to extract transient features to obtain a time domain tensor, i.e., a time domain feature map of the lth layer;

[0097]

[0098] wherein, is the time domain tensor, and the time domain feature map of the lth layer; is the time domain feature map of the (l-1)th layer; is the nonlinear transformation of the lth layer, and [·] represents a channel concatenation operation;

[0099] The above-mentioned hierarchical residual dense network can include 3 residual blocks;

[0100] (2) The time domain tensor is subjected to discrete Fourier transform to extract frequency domain features of load periodic fluctuations, and the frequency domain features are input into a hierarchical residual dense network to obtain a frequency domain tensor, i.e., a frequency domain tensor of the lth layer

[0101] (3) The time domain tensor and the frequency domain tensor are subjected to global average pooling by using a cross-domain attention fusion mechanism to obtain a global feature vector;

[0102]

[0103] wherein, is the time domain tensor of the i th monitoring point at the k th sampling time, is the frequency domain tensor of the i th monitoring point at the k th sampling time, and Nt and N f N and N are the number of sampling points of the time domain tensor sequence and the frequency domain tensor sequence respectively, ω t (i) and ω f (i) are attention weighting coefficients respectively, the monitoring point i ∈ [1, M], M is the total number of monitoring points.

[0104] (II) The response evaluation model predicts three indexes, i.e., the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, in the current operating state of the park power grid based on the global feature vector in the multi-dimensional standard data set, including:

[0105] (1) The global feature vector is taken as the digital representation of the current operating state;

[0106] (2) The deep learning network model is used to learn the correlation between the time-frequency domain features in the global feature vector and the three indexes, and the three indexes, i.e., the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, in the current operating state of the park power grid are predicted;

[0107] The deep learning network model is not limited in the embodiment and can be set as needed.

[0108] (3) The load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity are taken as the collaborative feature data set x, and the corresponding source-grid-load-storage collaboration degree H(x) is obtained, i.e., H(1), H(2), and H(3);

[0109] (4) The collaboration degree expectation y of the current operating state of the park power grid is determined, the loss function is determined as shown below, and the source-grid-load-storage collaboration degree H(x) when the loss function is minimized is output, i.e., H(1), H(2), and H(3);

[0110]

[0111] The goal of the above loss function is to make H(x) as close to y as possible. When the loss function does not reach the minimum value, it means that there is a large deviation between the source-grid-load-storage collaboration degree output by the model and the collaboration degree expectation, at this time, the mapping weight of the correlation between the time-frequency domain features in the global feature vector and the three indexes is adjusted, and the three indexes, i.e., the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, in the current operating state of the park power grid are predicted again until the loss function takes the minimum value.

[0112] (III) The multi-objective optimization model takes the maximization of the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity as the goal, and iteratively optimizes the improved particle swarm algorithm to obtain the corresponding collaborative control parameter combination when the goal is maximized, including:

[0113] (1) Input the load penetration rate, load fluctuation mitigation amount, and distributed power absorption capacity output by the input response evaluation model to the optimization degree function to obtain the corresponding optimization degree, where the optimization degree formula is as follows:

[0114]

[0115] Among them, v x The optimization coefficients of the multidimensional evaluation index system correspond one-to-one with H(x);

[0116] (2) Based on the improved particle swarm optimization algorithm, active power, reactive power, voltage, current, and energy storage state of charge are iteratively optimized, and the active power, reactive power, voltage, current, and energy storage state of charge corresponding to the maximum optimization degree are output to form a coordinated control parameter combination, specifically including:

[0117] In the defined multidimensional standard dataset, any variable is E. i (The variable can be any one of active power, reactive power, voltage, current, or energy storage state of charge), determine the variable E. i The value at sampling time t Based on the improved particle swarm algorithm calculation Specifically as follows:

[0118]

[0119] in, Let be the global feature vector at sampling time t, and γ be the convergence factor;

[0120] Will Updated to The corresponding z is recalculated using the feature extraction model and the response evaluation model. fusion , H(x), and obtain the corresponding optimization degree;

[0121] Repeat the above steps until convergence, and output the active power, reactive power, voltage, current and energy storage state of charge corresponding to the maximum optimization degree (the maximum value among all optimization degrees in steps (1) and (2)) to form a combination of coordinated control parameters.

[0122] Example 4: As shown in the appendix Figure 3 As shown in the embodiment of the present invention, this embodiment discloses a method for optimizing and evaluating the coordinated response characteristics of power grid sources, grids, loads, and storage in a power park, including:

[0123] Step S310: Obtain a multi-dimensional standard dataset of the power grid in the park to be optimized and evaluated within a certain time window. The multi-dimensional standard dataset includes multi-dimensional data from multiple monitoring points at multiple sampling times. The multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge.

[0124] Step S320, input the multi-dimensional standard data set to the collaborative control optimization model to obtain an optimized collaborative control parameter combination, wherein the collaborative control optimization model is obtained by training a plurality of samples, each sample including a multi-dimensional standard data set and corresponding label data of the collaborative control parameter combination, and the collaborative control optimization model includes a feature extraction model, a source-grid-load-storage collaboration degree output model, and a multi-objective optimization model.

[0125] The feature extraction model extracts a global feature vector in the multi-dimensional standard data set.

[0126] The response evaluation model predicts the load penetration rate, load fluctuation suppression amount, and distributed power consumption capacity under the current operating state of the evaluation park power grid based on the global feature vector in the multi-dimensional standard data set.

[0127] The multi-objective optimization model maximizes the load penetration rate, load fluctuation suppression amount, and distributed power consumption capacity, and iteratively optimizes the combination of collaborative control parameters by combining an improved particle swarm algorithm to obtain the corresponding combination of collaborative control parameters when the target is maximized, wherein the combination of collaborative control parameters includes active power, reactive power, voltage, current, and energy storage state of charge.

[0128] Step S330, the collaborative control parameter combination obtained by optimization and evaluation is used as an optimization strategy for distribution and execution, and real-time collection of park power grid operation data after execution of the optimization strategy is performed for optimization and evaluation feedback adjustment.

[0129] The optimization and evaluation feedback adjustment can be used to determine whether the park power grid operation data after execution of the optimization strategy meets the requirements, if it meets the requirements, no operation is performed, and if it does not meet the requirements, the collaborative control optimization model is retrained.

[0130] Embodiment 5: as shown in the accompanying drawings Figure 4 The embodiment of the present application discloses a park power grid source-grid-load-storage collaborative response characteristic optimization and evaluation system, which comprises:

[0131] The data acquisition unit acquires a multi-dimensional standard data set of the park power grid to be optimized and evaluated in a certain time window, wherein the multi-dimensional standard data set includes multi-dimensional data of a plurality of monitoring points at a plurality of sampling times, and the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge.

[0132] The optimization and evaluation unit inputs the multi-dimensional standard data set to the collaborative control optimization model to obtain an optimized collaborative control parameter combination, wherein the collaborative control optimization model is obtained by training a plurality of samples, each sample including a multi-dimensional standard data set and corresponding label data of the collaborative control parameter combination, and the collaborative control optimization model includes a feature extraction model, a source-grid-load-storage collaboration degree output model, and a multi-objective optimization model.

[0133] a feature extraction model, which extracts a global feature vector in a multi-dimensional standard data set;

[0134] a response evaluation model, which predicts a load penetration rate, a load fluctuation suppression amount, and a distributed power consumption capacity under a current operating state of the optimized evaluation park power grid based on the global feature vector in the multi-dimensional standard data set;

[0135] a multi-objective optimization model, which maximizes the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, and iteratively optimizes the combination of the coordinated control parameters, including active power, reactive power, voltage, current, and energy storage state of charge, by using an improved particle swarm algorithm.

[0136] The data acquisition unit includes a distributed monitoring terminal and a data preprocessing module.

[0137] The distributed monitoring terminal is deployed at the park substation, the distributed power grid connection point, the important load node, and the energy storage converter. In a certain time window, it acquires multi-dimensional data of each monitoring point in the optimized evaluation park power grid at multiple sampling times to form a multi-dimensional original data set. The multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge.

[0138] The data preprocessing unit aligns the multi-dimensional original data set in the time domain, filters data with a deviation value exceeding three times the standard deviation using the 3σ principle, maps the filtered multi-dimensional original data set to a numerical interval of [0, 1] through linear transformation, and forms a multi-dimensional standard data set.

[0139] As shown in the accompanying Figure 5 The optimization evaluation unit includes:

[0140] The feature extraction module extracts a global feature vector in the multi-dimensional standard data set using the feature extraction model, including:

[0141] The multi-dimensional standard data set is represented as an original tensor, and the original tensor is input into the hierarchical residual dense network to extract transient features and obtain a time-domain tensor.

[0142] The time-domain tensor is subjected to discrete Fourier transform to extract frequency domain features of load periodic fluctuations, and the frequency domain features are input into the hierarchical residual dense network to obtain a frequency domain tensor.

[0143] The time-domain tensor and the frequency domain tensor are subjected to global average pooling using a cross-domain attention fusion mechanism to obtain a global feature vector.

[0144]

[0145] In the above, is the time-domain tensor of the i-th monitoring point at the k-th sampling time, is the frequency-domain tensor of the i-th monitoring point at the k-th sampling time, N t and N f are the sampling point numbers of the time-domain tensor sequence and the frequency-domain tensor sequence respectively, ω t (i) and ω f (i) are attention weighting coefficients respectively, monitoring point i∈[1,M], M is the total number of monitoring points;

[0146] a response evaluation module, which utilizes a response evaluation model to predict three indexes, i.e., a load penetration rate, a load fluctuation suppression amount and a distributed power consumption capacity, in a current operating state of the park power grid based on a global feature vector in a multi-dimensional standard data set, including:

[0147] taking the global feature vector as a digital representation of the current operating state;

[0148] utilizing a deep learning network model to learn an association relationship between time-frequency domain features in the global feature vector and the three indexes, to predict the three indexes, i.e., the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, in the current operating state of the park power grid;

[0149] taking the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity as a coordination feature data set x, to obtain a corresponding source-grid-load-storage coordination degree H(x);

[0150] determining a coordination degree expectation y of the current operating state of the park power grid, determining a loss function as shown below, and outputting a source-grid-load-storage coordination degree H(x) at which the loss function is minimum;

[0151]

[0152] a multi-objective optimization module, which utilizes a multi-objective optimization model to maximize the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, and combines an improved particle swarm algorithm to iteratively optimize, to obtain a corresponding coordination control parameter combination when the target is maximized, including:

[0153] inputting the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity output by the response evaluation model to an optimization degree function, to obtain a corresponding optimization degree, wherein the optimization degree formula is as shown below:

[0154]

[0155] wherein, v x is an optimization coefficient of a multi-dimensional evaluation index system, corresponding to H(x) one by one;

[0156] The active power, the reactive power, the voltage, the current, and the energy storage state of charge are respectively iteratively optimized based on the improved particle swarm algorithm, and the active power, the reactive power, the voltage, the current, and the energy storage state of charge corresponding to the maximum optimization degree are output, thereby forming the collaborative control parameter combination.

[0157] Embodiment 6: As shown in the accompanying Figure 6 The park power grid source network load storage collaborative response characteristic optimization evaluation system disclosed by the embodiment of the present application comprises:

[0158] A data acquisition unit acquires a multi-dimensional standard data set of the park power grid to be optimized and evaluated in a certain time window, wherein the multi-dimensional standard data set comprises multi-dimensional data of a plurality of monitoring points at a plurality of sampling time points, and the multi-dimensional data comprises active power, reactive power, voltage, current, and energy storage state of charge.

[0159] An optimization evaluation unit inputs the multi-dimensional standard data set into a collaborative control optimization model to obtain an optimized collaborative control parameter combination, wherein the collaborative control optimization model is obtained by training a plurality of samples, each sample comprising multi-dimensional standard data set and label data of corresponding collaborative control parameter combination, and the collaborative control optimization model comprises a feature extraction model, a source network load storage collaboration degree output model, and a multi-objective optimization model.

[0160] The feature extraction model extracts a global feature vector in the multi-dimensional standard data set.

[0161] A response evaluation model predicts the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity under the current operating state of the park power grid to be optimized and evaluated based on the global feature vector in the multi-dimensional standard data set.

[0162] The multi-objective optimization model maximizes the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, and iteratively optimizes the collaborative control parameter combination by combining the improved particle swarm algorithm, wherein the collaborative control parameter combination comprises active power, reactive power, voltage, current, and energy storage state of charge.

[0163] A closed-loop execution feedback unit issues and executes the collaborative control parameter combination obtained by optimization evaluation as an optimization strategy, and real-time collects park power grid operation data after execution of the optimization strategy, and performs optimization evaluation feedback adjustment.

[0164] The above is only a specific embodiment of the present application, which has strong adaptability and implementation effect, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application, therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A method for optimizing and evaluating source-grid-load-storage collaborative response characteristics of a park power grid, characterized in that, The method comprises the following steps: acquiring a multi-dimensional standard data set of the to-be-optimized evaluation park power grid in a time window, wherein the multi-dimensional standard data set comprises multi-dimensional data of a plurality of monitoring points at a plurality of sampling time points, and the multi-dimensional data comprises active power, reactive power, voltage, current and energy storage state of charge; inputting the multi-dimensional standard data set into a coordinated control optimization model to obtain an optimized coordinated control parameter combination, wherein the coordinated control optimization model is obtained by training a plurality of samples, each sample comprising multi-dimensional standard data set and corresponding label data of the coordinated control parameter combination, and the coordinated control optimization model comprises a feature extraction model, a source-grid-load-storage coordination degree output model and a multi-objective optimization model; the feature extraction model extracts a global feature vector in the multi-dimensional standard data set; a response evaluation model predicts a load penetration rate, a load fluctuation suppression amount and a distributed power consumption capacity under a current operating state of the to-be-optimized evaluation park power grid based on the global feature vector in the multi-dimensional standard data set; a multi-objective optimization model maximizes the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, and iteratively optimizes the maximized target by using an improved particle swarm algorithm to obtain a corresponding coordinated control parameter combination, wherein the coordinated control parameter combination comprises active power, reactive power, voltage, current and energy storage state of charge.

2. The method of claim 1, wherein the method further comprises: The feature extraction model extracts a global feature vector in the multi-dimensional standard data set, comprising: representing the multi-dimensional standard data set as an original tensor, inputting the original tensor into a hierarchical residual dense network to extract transient features and obtain a time domain tensor; performing discrete Fourier transform on the time domain tensor to extract frequency domain features of load periodic fluctuation, inputting the frequency domain features into a hierarchical residual dense network to obtain a frequency domain tensor; performing global average pooling on the time domain tensor and the frequency domain tensor by using a cross-domain attention fusion mechanism to obtain a global feature vector; wherein, is the time-domain tensor of the i-th monitoring point at the k-th sampling time, is the frequency-domain tensor of the i-th monitoring point at the k-th sampling time, N t and N f is the sampling point number of the time-domain tensor sequence and the frequency-domain tensor sequence, respectively, ω t (i) and ω f (i) are attention weighting coefficients, respectively, monitoring point i ∈ [1, M], and M is the total number of monitoring points.

3. The method of claim 1 or 2, wherein the method further comprises: determining the source-grid-load-storage coordination response characteristic of the park grid based on the power consumption of the park grid and the power generation of the park grid. the response evaluation model predicts three indexes, i.e., a load penetration rate, a load fluctuation suppression amount and a distributed power consumption capacity, under a current operating state of the park power grid based on the global feature vector in the multi-dimensional standard data set, comprising: using the global feature vector as a digital representation of the current operating state; learning a correlation between time-frequency domain features in the global feature vector and the three indexes by using a deep learning network model to predict the three indexes, i.e., the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, under the current operating state of the park power grid; using the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity as a coordinated feature data set x to obtain a corresponding source-grid-load-storage coordination degree H(x); determining a coordination degree expectation y of the current operating state of the park power grid, determining a loss function as shown below, and outputting the source-grid-load-storage coordination degree H(x) when the loss function is minimum; 4. The method of claim 1 to 3, wherein, the multi-objective optimization model maximizes the load penetration rate, the load fluctuation suppression amount and the distributed power consumption capacity, and iteratively optimizes the maximized target by using an improved particle swarm algorithm to obtain a corresponding coordinated control parameter combination, comprising: The penetration rate of the input response evaluation model output load, the load fluctuation suppression amount, the distributed power consumption capacity to the optimization degree function, and the corresponding optimization degree are obtained, wherein the optimization degree formula is as follows: where v x is the optimization coefficient of the multi-dimensional evaluation index system, corresponding to H(x). Based on the improved particle swarm algorithm, the active power, the reactive power, the voltage, the current, and the energy storage state of charge are iteratively optimized, and the active power, the reactive power, the voltage, the current, and the energy storage state of charge corresponding to the maximum optimization degree are output, forming a collaborative control parameter combination.

5. The method of claim 1 to 4, wherein, Obtain a multi-dimensional standard data set of the to-be-optimized evaluation park power grid in a certain time window, including: In a certain time window, obtain the multi-dimensional data of each monitoring point in the to-be-optimized evaluation park power grid at multiple sampling times, form a multi-dimensional original data set, and the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge. The multi-dimensional original data set is time-domain aligned, and data with a deviation value exceeding three times the standard deviation is filtered using the 3σ principle. The filtered multi-dimensional original data set is mapped to the numerical interval [0, 1] through linear transformation, forming a multi-dimensional standard data set.

6. The method of claim 1 to 5, wherein, It also includes issuing and executing the collaborative control parameter combination obtained by optimization evaluation as an optimization strategy, and real-time collection of park power grid operation data after optimization strategy execution for optimization evaluation feedback adjustment.

7. A park grid source-grid-load-storage collaborative response characteristic optimization evaluation system applying the method of any one of claims 1 to 6, characterized in that, It includes: A data acquisition unit acquires a multi-dimensional standard data set of a to-be-optimized evaluation park power grid in a certain time window, wherein the multi-dimensional standard data set includes multi-dimensional data of multiple monitoring points at multiple sampling times, and the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge. An optimization evaluation unit inputs the multi-dimensional standard data set into a collaborative control optimization model to obtain an optimized collaborative control parameter combination, wherein the collaborative control optimization model is trained by a plurality of samples, each sample including multi-dimensional standard data set and corresponding collaborative control parameter combination label data, and the collaborative control optimization model includes a feature extraction model, a source-grid-storage coordination degree output model, and a multi-objective optimization model. The feature extraction model extracts the global feature vector in the multi-dimensional standard data set. The response evaluation model predicts the load penetration rate, load fluctuation suppression amount, and distributed power consumption capacity under the current operating state of the to-be-optimized evaluation park power grid based on the global feature vector in the multi-dimensional standard data set. The multi-objective optimization model maximizes the load penetration rate, load fluctuation suppression amount, and distributed power consumption capacity, and iteratively optimizes the collaborative control parameter combination corresponding to the maximum target using an improved particle swarm algorithm, wherein the collaborative control parameter combination includes active power, reactive power, voltage, current, and energy storage state of charge.

8. The park grid source-network-load-storage collaborative response characteristic optimization evaluation system according to claim 7, characterized in that, It also includes a closed-loop execution feedback unit that issues and executes the collaborative control parameter combination obtained by optimization evaluation as an optimization strategy, and real-time collection of park power grid operation data after optimization strategy execution for optimization evaluation feedback adjustment.

9. The park grid source-grid-load-storage collaborative response characteristic optimization evaluation system according to claim 7 or 8, characterized in that, The optimization evaluation unit includes: The feature extraction module extracts the global feature vector in the multi-dimensional standard data set using the feature extraction model, including: The multi-dimensional standard data set is represented as an original tensor, and the original tensor is input into a hierarchical residual dense network to extract transient features, thereby obtaining a time-domain tensor; The time-domain tensor is subjected to discrete Fourier transform to extract frequency-domain features of load periodic fluctuations, and the frequency-domain features are input into a hierarchical residual dense network to obtain a frequency-domain tensor; The time-domain tensor and the frequency-domain tensor are subjected to global average pooling using a cross-domain attention fusion mechanism to obtain a global feature vector; wherein, is the time-domain tensor of the i-th monitoring point at the k-th sampling time, is the frequency-domain tensor of the i-th monitoring point at the k-th sampling time, N t and N f is the sampling point number of the time-domain tensor sequence and the frequency-domain tensor sequence, respectively, ω t (i) and ω f (i) are attention weighting coefficients, respectively, monitoring point i ∈ [1, M], M is the total number of monitoring points; A response evaluation module uses a response evaluation model to predict three indicators, i.e., a load penetration rate, a load fluctuation suppression amount, and a distributed power consumption capacity, in a current operating state of the park power grid based on the global feature vector in the multi-dimensional standard data set, including: The global feature vector is used as a digital representation of the current operating state; A deep learning network model is used to learn the correlation between the time-frequency domain features in the global feature vector and the three indicators, thereby predicting the three indicators, i.e., the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, in the current operating state of the park power grid; The load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity are used as a collaborative feature data set x to obtain a corresponding source-grid-load-storage collaboration degree H(x); A collaboration degree expectation y of the current operating state of the park power grid is determined, and a loss function is determined as shown below, and a source-grid-load-storage collaboration degree H(x) at a minimum loss of the loss function is output; A multi-objective optimization module uses a multi-objective optimization model to maximize the load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity, and iteratively optimizes the improved particle swarm algorithm to obtain a corresponding collaborative control parameter combination when the target is maximized, including: The load penetration rate, the load fluctuation suppression amount, and the distributed power consumption capacity output by the response evaluation model are input into an optimization degree function to obtain a corresponding optimization degree, where the optimization degree formula is as shown below: where v x is the optimization coefficient of the multi-dimensional evaluation index system, corresponding to H(x). The active power, the reactive power, the voltage, the current, and the energy storage state of charge are iteratively optimized based on the improved particle swarm algorithm, and the active power, the reactive power, the voltage, the current, and the energy storage state of charge corresponding to the maximum optimization degree are output to form a collaborative control parameter combination.

10. The park grid source-grid-load-storage collaborative response characteristic optimization evaluation system according to claim 7 or 8, characterized in that, The data acquisition unit includes a distributed monitoring terminal and a data preprocessing module; The distributed monitoring terminal is deployed at a park substation, a distributed power grid connection point, an important load node, and an energy storage converter, and acquires multi-dimensional data of each monitoring point in the park power grid to be optimized and evaluated at multiple sampling times within a certain time window to form a multi-dimensional original data set, where the multi-dimensional data includes active power, reactive power, voltage, current, and energy storage state of charge; The data preprocessing unit aligns the multi-dimensional original data set in the time domain, filters data with a deviation value exceeding three times the standard deviation using the 3σ principle, maps the filtered multi-dimensional original data set to a numerical interval of [0, 1] through linear transformation, and forms a multi-dimensional standard data set.

Images