Distributed resource flexibility assessment method based on double-layer assessment model

Abstract

The invention provides a distributed resource flexibility assessment method based on a double-layer assessment model, and the method comprises the steps: dividing feature data obtained from a distributed energy system into dynamic operation data, equipment health state data and static inherent feature data, carrying out the feature extraction of each type of data, forming a feature matrix, and carrying out the fusion, carrying out Transform extraction on the fused feature matrix to obtain a global feature vector; constructing a node resource flexibility assessment framework, taking the global feature vector as framework input, designing assessment indexes covering five key dimensions of equipment performance, health state, operation efficiency, absorption capability and elastic capacity, and generating a node flexibility score by calculating each index; a double-layer evaluation model composed of a connection layer and a margin score mapping layer is constructed, connectivity and parameters between nodes in the distributed energy system are defined through the connection layer, and the margin score mapping layer quantifies influences between the nodes by means of a power flow propagation matrix through a global feature vector and a node resource flexibility evaluation framework.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed resource flexibility assessment in power systems, and particularly to a distributed resource flexibility assessment method based on a two-layer assessment model. Background Art

[0002] In the context of the accelerating global energy transition, traditional centralized resources are gradually transforming into green and low-carbon distributed resources. Distributed resources themselves have volatility and uncertainty, and scientific assessment of their flexibility is of great significance for ensuring the sustainability and security of energy supply. The use of distributed energy flexibility assessment methods can not only deeply reveal the dynamic characteristics of the system, but also provide theoretical support for the coordinated scheduling of various distributed resources, promote the complementarity and resource sharing between different energy forms, thereby enhancing the reliability and operation efficiency of the overall energy system.

[0003] Currently, flexibility assessment methods mainly analyze from three dimensions: supply-demand balance within a region, power flow distribution within a region, and inter-regional transmission capacity. In terms of supply-demand balance within a region, the assessment method mainly focuses on the matching of power production and consumption in each region, and identifies potential risks of supply-demand imbalance by analyzing the differences between power generation capacity and load demand. However, with the continuous expansion of the power system scale and the gradual increase in distributed energy, the deficiency of simple supply-demand balance analysis in coping with complex and changing system dynamics. In the assessment of power flow distribution within a region, it focuses on analyzing the power flow direction and transmission path of each node to evaluate the flexibility of the power grid in coping with local load fluctuations and power flow scheduling. However, with the increase in the complexity of the power system, traditional power flow distribution analysis methods often fail to reflect the changes in the power grid in real time, resulting in insufficient accuracy of the assessment results. And the assessment of inter-regional transmission capacity mainly examines the power flow capacity between different regions, aiming to enhance the flexibility of the system by improving the interconnection capacity between regions. Although the current flexibility assessment methods can analyze the operation of the power grid from different dimensions, with the increasing complexity and dynamic changes of the system, a single assessment dimension has become difficult to comprehensively and accurately reflect the flexibility of the power grid. Future flexibility assessment methods need to more comprehensively and dynamically consider multi-dimensional factors to cope with the challenges of increasing distributed resources and the growing complexity of the power system.

[0004] With the continuous progress of data-driven technologies, the demand for power system flexibility assessment has been continuously increasing, and existing methods have deficiencies in addressing system complexity. This is mainly manifested as follows: the complex non-linear relationships and dynamic characteristics within the system cannot be comprehensively captured, and the response ability and adaptability are poor when facing complex network structures and changing operating conditions; the effects of actual factors such as equipment aging, environmental impacts, and line scheduling on transmission capacity are not considered, resulting in the assessment results being difficult to meet the requirements of modern power grids in terms of accuracy and reliability; the assessment means are limited to a single dimension and the index system is not comprehensive enough, making it difficult to quantify defects such as direct connections and indirect impacts between nodes, and thus there are deficiencies in the analysis of the overall performance and stability of the system. Summary of the Invention

[0005] To solve the above technical problems, the object of the present invention is to provide a distributed resource flexibility assessment method based on a two-layer assessment model, which improves the management efficiency and operation reliability of distributed energy systems by dynamically integrating system characteristics, equipment health, and time series data.

[0006] The present invention provides a distributed resource flexibility assessment method based on a two-layer assessment model, including:

[0007] Step 1: Divide the feature data obtained from the distributed energy system into dynamic operation data, equipment health status data, and static inherent feature data, extract features for each type of data to form a feature matrix and fuse them, and obtain a global feature vector by extracting the fused feature matrix through a Transformer.

[0008] Step 2: Construct a node resource flexibility assessment framework, use the global feature vector as the framework input, design assessment indicators covering five key dimensions of equipment performance, health status, operation efficiency, consumption capacity, and elastic capacity, and generate a node flexibility score by calculating each indicator.

[0009] Step 3: Construct a two-layer assessment model composed of a connection layer and a margin score mapping layer. Define the connectivity and parameters between nodes in the distributed energy system through the connection layer, and the margin score mapping layer uses the global feature vector and the node resource flexibility assessment framework to quantify the influence between nodes with the help of the power flow propagation matrix.

[0010] The distributed resource flexibility assessment method based on a two-layer assessment model of the present invention has the following

[0011] Advantages:

[0012] (1) Through the hierarchical processing and matrix storage of dynamic operation data, equipment health status data, and static inherent characteristic data in the distributed energy system, and by combining local convolutional neural network and global Transformer technology, the extraction and fusion of multi-level features are realized, retaining the detailed information of the original operation data of the system, providing a data basis for subsequent evaluation.

[0013] (2) By designing evaluation indicators covering multiple dimensions such as equipment performance, health status, operation efficiency, consumption capacity, and flexible capacity, and combining dynamic operation data, equipment health status data, and static inherent characteristic data, the present invention meets the need for comprehensive evaluation of the complex characteristics of distributed resources.

[0014] (3) By constructing a two-layer evaluation model composed of a connection layer and a margin score mapping layer, and considering the influence of power flow propagation at the same time, the direct and indirect influences of each node resource in the system are quantified, providing a quantitative analysis basis for the operation efficiency and stability of the system, supporting the comparison and selection of system optimization schemes, and improving the decision-making efficiency under multi-scenario working conditions.

[0015] (4) The method of the present invention combines the fusion of multi-source data, deep learning technology, and topological model analysis, giving full play to the advantages of multi-level feature extraction and multi-dimensional evaluation, ensuring that the flexibility of resources can be evaluated under complex operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of a method for evaluating the flexibility of distributed resources based on a two-layer evaluation model of the present invention;

[0017] Figure 2 is a schematic diagram of the two-layer evaluation model;

[0018] Figure 3 is a projection of the node resource polyhedron and a distribution diagram of node resource flexibility in a five-dimensional coordinate system;

[0019] Figure 4 is a distribution diagram of node resource flexibility considering direct and indirect influences in the system. DETAILED DESCRIPTION OF THE INVENTION

[0020] As Figure 1 shown, a method for evaluating the flexibility of distributed resources based on a two-layer evaluation model of the present invention includes:

[0021] Step 1: Divide the feature data obtained from the distributed energy system into dynamic operation data, equipment health status data, and static inherent characteristic data, extract features for each type of data to form a feature matrix and perform fusion, and extract the global feature vector from the fused feature matrix through Transformer. Specifically:

[0022] Step 1.1: Matrixize the obtained dynamic operation data (such as output power, response time) inside the distributed energy system to form a time series data matrix, and perform data cleaning and preprocessing on the time series data matrix. Process the time series data matrix through a local convolutional neural network to obtain a dynamic data feature matrix. Specifically:

[0023] Step 1.1.1: Fill in the missing values of the obtained dynamic operation data inside the distributed energy system by linear interpolation, and then matrixize it to form an original time series data matrix X of p*n dynamic :

[0024]

[0025] where d p,n represents the eigenvalue at each moment, where p is the time step and n is the variable dimension.

[0026] Step 1.1.2: Remove outliers from the original time series data matrix X dynamic That is, for each feature column, calculate the mean μ j and the standard deviation σ j as the basis for outlier judgment. Its calculation method:

[0027]

[0028] where d i,j is the jth eigenvalue of the ith row, μ j is the mean of this feature, σ j is the standard deviation of this feature. For each data point d i,j , if it satisfies the following conditions, it is considered an outlier:

[0029] |d i,j -μ j |>3σ j

[0030] For outliers, use the mean to replace the outliers for substitution processing. The calculation method of the replaced data is:

[0031]

[0032] Step 1.1.3: Normalize each feature data:

[0033]

[0034] where min(d' *j ) is the minimum value in the jth column feature, max(d' *j) is the maximum value in the j-th column feature.

[0035] Step 1.1.4: Process the dynamic operation data through a local convolutional neural network to extract the short-term dynamic features of the nodes, and obtain the dynamic data feature matrix X1:

[0036]

[0037] where f p,m represents the short-term dynamic features extracted through the convolution operation.

[0038] Step 1.2: For the equipment health status data of the distributed energy system, construct the original health status data matrix, and obtain the health status data feature matrix by linearly repeating and aligning it to the row dimension of the dynamic data feature matrix. Specifically:

[0039] Step 1.2.1: Construct the original health status data matrix X state :

[0040]

[0041] where A a represents the number of the a-th device, B a represents the usage rate encoding of the a-th device, C a represents the environmental condition encoding of the a-th device, and D a represents the health status encoding of the a-th device.

[0042] Usage rate encoding of the device: idle 0, low load 1, medium load 2, high load 3, overload 4 for the device operation status.

[0043] Environmental condition encoding of the device: normal temperature 0, too high temperature 1, normal humidity 2, too high humidity 3, abnormal vibration 4.

[0044] Health status encoding of the device: normal operation 0, to be repaired 1, fault shutdown 2, temporary shutdown 3, overload 4.

[0045] Step 1.2.1: Linearly interpolate and repeat the original health status data matrix X state to align it to the row dimension of the dynamic data feature matrix X1, and obtain the health status data feature matrix X2 with dimensions p*y.

[0046] The calculation formula for the supplementary data a i,j in the matrix is:

[0047]

[0048] where Δvi is the linear increment of the initial vector v0 in each transformation, and p - x is the number of repetitions of the transformation.

[0049] Step 1.3: For the static inherent characteristic data related to the distributed energy system, covering invariant state parameters such as energy types (wind energy, photovoltaic, etc.), equipment types (battery energy storage, generator, etc.), and geographical distribution and relative positions, construct an initial static data matrix, and obtain a static data feature matrix by linearly repeating and aligning the matrix to the row dimension of the dynamic data feature matrix. Specifically:

[0050] Step 1.3.1: For the static inherent characteristic data related to the distributed energy system, construct an initial static data matrix X based on energy type coding, type coding, and geographical location coding static :

[0051]

[0052] where E a represents the energy type coding of the a-th device, F a represents the type coding of the a-th device, and G a represents the geographical location coding of the a-th device.

[0053] Energy type coding: wind energy 0, photovoltaic 1, electricity 2, gas 3, cold 4, heat 5, two - energy coupling 6, three - energy coupling 7.

[0054] Device type coding: battery energy storage 0, generator 1, CCHP 2, electric boiler 3.

[0055] Geographical location coding: commercial (0), residential (1), industrial (2), educational (3).

[0056] Step 1.3.2: Linearly interpolate and repeat the initial static data matrix X static to align it to the row dimension of the dynamic data feature matrix X1, obtaining a static data feature matrix X3 with dimensions p * z.

[0057] Step 1.4: Extract the dynamic data feature matrix, health status data feature matrix, and static data feature matrix through a global Transformer, model the long - term dependencies and global associations in the time - series data, and form a global feature vector for efficient input through the unified fusion of multi - level features. Specifically:

[0058] Step 1.4.1: In the feature matrices with dimensions p * m, p * y, and p * z respectively, find the maximum number of columns q:

[0059] q = max(m, y, z)

[0060] Step 1.4.2: Dynamically weight and fuse these three feature matrices to obtain a feature matrix X of p*q trans :

[0061] X trans = α1·X1 + α2·X2 + α3·X3

[0062]

[0063] where α i is the dynamic weighting coefficient, and ||X i || F is the Frobenius norm of matrix X1.

[0064] Step 1.4.3: Extract long-term dependence features from X trans through the global Transformer model to obtain a global feature vector X transformer = [x 1,1 … x 1,q of dimension 1*q.

[0065] Step 2: Construct a node resource flexibility evaluation framework, with the global feature vector as the framework input, design evaluation indicators covering 5 key dimensions of device performance, health status, operation efficiency, accommodation capacity, and elastic capacity, and generate a node flexibility score by calculating each indicator, specifically:

[0066] Step 2.1: Use the global feature vector extracted in Step 1 as the input, construct device performance indicators from two aspects of device utilization rate and power output stability, and calculate the device performance indicator score.

[0067] Step 2.1.1: Calculate the device utilization rate U b , which is the ratio of the actual running time of the device to its total available time within a certain time period, while considering non-linear factors (such as load changes, device response delays, etc.):

[0068]

[0069] After calculating U b , use the min-max normalization method to convert the result into a score:

[0070]

[0071] where Δt is the time increment of device operation, and T available is the total available time of the device; the increase of Δt directly affects the system regulation ability. If the device does not run during a certain period, the R bt value for this period is 0; if the device runs normally, the R bt value is 1; U min is U for this periodi The minimum value, U max For this period U b The maximum value; is the equipment utilization score. The higher the equipment utilization, the higher the operating efficiency and the stronger the system flexibility. Calculate the utilization rate of the equipment during the specified time period. When dealing with data from multiple time periods, merge the results through weighted average or summation: if the weights of each time period are the same, calculate the average value of the operating ratios of each time period; if the weights are different, use weighted average.

[0072] Step 2.1.2: Calculate the power output stability Q i , by calculating the standard deviation of the equipment power output, collect the power data of the equipment at multiple time points, and use the standard deviation formula to measure the stability:

[0073]

[0074] where, P bt is the power output of equipment b at time t, is the average power output during this period, α is the load factor, usually set as a constant value (0.5 - 0.8) in combination with the actual situation, L t is the load fluctuation amount, and n is the number of time series; is the power output stability score; Q min For this period Q b The minimum value, Q max For this period Q b The maximum value;

[0075] Step 2.1.3: Calculate the equipment performance index, through and Weighted to get the score of this index:

[0076]

[0077] where, w U and w S are the weights of the two set indexes respectively, and the sum of the two weights is 1.

[0078] Step 2.2: Use the global feature vector extracted in Step 1 as the input, construct the health status index from two aspects of the failure rate and the remaining life, and calculate the score of the constructed health status index.

[0079] Step 2.2.1: Calculate the failure rate, which is composed of the number of failures of different equipment in multiple time periods, and obtain the final failure rate F through weighted average b :

[0080]

[0081] Among them, E f is the severity index of the f-th failure mode, generally set as a constant (0 - 1), T f is the operating time of the device under the f-th failure mode, T total is the total operating time of the device; is the failure incidence rate score; F min is the minimum value of F during this period, F b is the maximum value of F during this period, F max is the maximum value of F during this period, F b is the maximum value of F during this period.

[0082] Step 2.2.2: Calculate the remaining life. For multiple health data (such as health score, number of repairs, service life, etc.), calculate the remaining life through weighted average, and judge the life expectancy L of the device according to the current health status and operating data of the device residual,b :

[0083]

[0084] Among them, R bd is the health status score of device b under the d-th condition, μ b is the original design life of the device; γ and δ are adjustment parameters, adjusted according to engineering needs, γ is generally set as a constant (0.5 - 2), and δ is generally set as a constant (0.01 - 0.05); is the remaining life score; L residual,min is the minimum value of L during this period, L residual,b is the minimum value of L during this period, L residual,max is the maximum value of L during this period, L residual,b is the maximum value of L during this period. exp() is used to adjust the attenuation or gain degree of the remaining life according to the overall situation of the deviation of the health status from the threshold.

[0085] Step 2.2.3: Calculate the health status index, through and weighted to obtain the score of this index:

[0086]

[0087] Among them, w F and w L are respectively set as the weights of the two indexes.

[0088] Step 2.3: Take the global feature vector extracted in Step 1 as the input, construct the operating efficiency index from two aspects of power transmission efficiency and unit energy consumption ratio, and calculate the operating efficiency index score.

[0089] Step 2.3.1: Calculate the power transmission efficiency E p, considering the power loss and non-linear transmission loss in the system, the calculation method is defined as:

[0090]

[0091] Among them, P load is the power at the load end, P generation is the power at the power generation end, β loss is the transmission loss coefficient, generally set as a constant (0.95 - 0.98), L loss is the power loss during transmission, L load is the power demand at the load end; is the power transmission efficiency score; E pmin is the minimum value of E p during this period, E pmax is the maximum value of E p during this period.

[0092] Step 2.3.2: Calculate the unit energy consumption ratio C u , considering the influence of factors such as equipment energy efficiency, load fluctuation, and climate change, the calculation method is:

[0093]

[0094] Among them, E total,t is the total energy consumption at time t, α t is the energy efficiency adjustment coefficient at time t (considering climate change, load fluctuation, etc.), generally set as a constant (0.9 - 0.95), P total is the total power output of the system, and n is the total number of time periods; is the unit energy consumption ratio score; C umin is the minimum value of C u during this period, C umax is the maximum value of C u during this period.

[0095] Step 2.3.3: Calculate the operation efficiency index, and obtain the score of this index through and weighting:

[0096]

[0097] Among them, and are respectively set as the weights of the two indexes.

[0098] Step 2.4: Use the feature vector extracted in Step 1 as the input, and calculate the consumption capacity index according to the renewable energy consumption ratio. The calculation method is as follows:

[0099]

[0100] Among them, P renewable,t is the renewable energy output at time t, and β t is the weather or seasonal fluctuation adjustment coefficient, usually set to a constant value (0.8 - 1.2), and γ t is the grid load demand adjustment coefficient, usually set to a constant value (0.95 - 1.05), and P total,t is the total output power; S renewable is the absorption capacity index score, and R cmin is the minimum value of R during this period c and R cmax is the maximum value of R during this period c ;

[0101] Step 2.5: Use the global feature vector extracted in Step 1 as the input, and construct an elastic capacity index from two aspects: load regulation ability and dynamic reserve capacity allocation ability, and calculate the elastic capacity index score.

[0102] Step 2.5.1: Calculate the load regulation ability, that is, calculate the response time of the system to load changes considering factors such as equipment regulation ability and load fluctuation degree:

[0103]

[0104] Among them, ΔP load is the change in load, Δt is the response time, σ load is the standard deviation of load fluctuation, and α l is the load fluctuation correction coefficient, generally set to a constant (0.9 - 1); is the load regulation ability score; L rmin is the minimum value of L during this period r and L rmax is the maximum value of L during this period r ;

[0105] Step 2.5.2: Calculate the dynamic reserve capacity allocation ability, and measure the dynamic reserve capacity allocation ability considering factors such as equipment failures and reserve capacity allocation:

[0106]

[0107] Among them, P backup is the reserve capacity, λ is the scheduling efficiency coefficient of the reserve capacity, generally set to a constant (0.9 - 1), and Δt is the time required for allocation; is the dynamic reserve capacity allocation ability score; C backupmin is the minimum value of C during this period backup and C backupmax is the maximum value of C during this period backupThe maximum value.

[0108] Step 2.5.3: Calculate the elastic capacity index, through and weighted to obtain the score of this index:

[0109]

[0110] Among them, and are respectively set as the weights of the two indicators.

[0111] Step 3: As Figure 2 shown, construct a two-layer evaluation model composed of a connection layer and a margin scoring mapping layer. Define the connectivity and parameters between nodes in the distributed energy system through the connection layer. The margin scoring mapping layer uses the global feature vector and the node resource flexibility evaluation framework, quantifies the influence between nodes with the help of the power flow propagation matrix, and finally shows the distribution of the target node resource flexibility in the system. Specifically:

[0112] Step 3.1: Construct the connection layer according to the physical structure and line topology of the distributed energy system. Each building is regarded as a node in the connection layer. The power generation equipment, energy storage equipment, and loads in the system are associated according to the building location and the corresponding equipment information. The connections in the connection layer represent the actual line connections between building nodes, intuitively showing the direct connection situation of each node in the system.

[0113] Step 3.1.1: Define the connection layer G=(N, A, L, C), use N={n1, n2,..., n n} to represent the set of nodes in the system. For each node n i assign relevant physical attributes to it.

[0114] Step 3.1.2: Construct the attribute set of nodes A={A1, A2,..., An}. The attribute set Ai of each node n i has physical parameters describing performance and response ability according to different types of equipment.

[0115] Use the set L={l ij , l ij' ,...} to represent the energy transmission paths between system nodes in the connection layer and the physical connection relationships of interactions between different nodes. l ij or l ij' represents the physical connection between node n i and node n j or the physical connection between node n i and node n j' ;

[0116] Step 3.1.4: For each physical connection lij ∈(n i , n j ) endows it with actual physical properties and defines its physical constraint set C ij ={P max , L loss}, where P max represents the maximum power transfer capacity of this connection, and L loss represents the loss during the power transfer process.

[0117] Step 3.2: Extract the feature data of each node in the connection layer. According to the method in Step 1, divide the feature data of each node n i into dynamic operation data, device health status data, and static inherent feature data, perform feature extraction on each type of data to form a feature matrix and fuse them, and extract the global feature vector through Transformer for the fused feature matrix.

[0118] In specific implementation, matrixize the obtained dynamic operation data inside the distributed energy system to form a time series data matrix. The original 8760*30 time series data matrix X dynamic obtained after processing is processed through Step 1.1 to obtain an 8760*20 dynamic data feature matrix X1. The original 8700*4 health status data matrix X state is processed through Step 1.2 to obtain an 8760*4 health status data feature matrix X2. The initial 8690*6 static data matrix X static is processed through Step 1.3 to obtain an 8760*6 static data feature matrix X3. Fuse the feature matrices X1, X2, and X3 and then extract through the Transformer model to obtain the global feature vector.

[0119] Step 3.3: Use the node resource flexibility evaluation framework in Step 2. Input the global feature vector of each node into the evaluation framework, calculate the node flexibility score of the node, and use it as the score S = [S p S ef S h S el S r of the node in the margin score mapping layer.

[0120] After selecting the target node, calculate the five scoring indicators respectively according to Step 2, and the obtained results are S p = 63, S h = 80, S ef = 95, S r = 81, S el= 82, map the five scoring metrics to the axes of a five-dimensional spatial coordinate system, form a polyhedron with the scores of each metric, and generate a flexibility distribution map through the two-dimensional projection of the polyhedron, as shown in Figure 3 as shown.

[0121] Step 3.4: For nodes that have mutual influence but no direct physical connection in the connection layer, reflect the indirect influence propagation between nodes in the margin scoring mapping layer through the five propagation matrices corresponding to the five scoring metrics, and quantify how the change in the state of the indirect node propagates to the corresponding target node through the intermediate nodes in the system and causes an impact:

[0122]

[0123] where, is the propagation matrix, k = 1, 2, 3, 4, 5; S k” is an element in S, R is the influence matrix, the five metrics respectively correspond to five S k' , and at the same time S k' ·R represents the interactive influence between the margin score and the influence matrix, d ij represents the shortest path between the target node n i and the indirect node n j ; Δθ represents the voltage phase angle difference between the two nodes, ΔV represents the voltage amplitude difference between the two nodes, ΔP represents the active power difference between the two nodes, and ΔQ represents the reactive power difference between the two nodes.

[0124] Step 3.5: When the system is updated, solve the updated propagation matrix P through the iterative propagation algorithm, calculate the active power flow of each node through the topological data in the connection layer to form a power flow matrix, and consider the multi-hop propagation effect of the system. The specific iterative propagation process is defined as:

[0125]

[0126] where, P (c) represents the propagation matrix of the c-th iteration, F ij is the active power flow from node i to node j, τ is the power flow influence coefficient, P (c-1) is the transpose of the propagation matrix, representing the reverse propagation of the influence.

[0127] Step 3.6: Calculate the five correlation coefficients and five coupling coefficients between the target node and the indirect node:

[0128]

[0129] where, is the correlation coefficient, is the coupling coefficient, represents node n i and node nj Propagation matrix of the interaction between and Covariance of and are respectively the standard deviations of the interaction matrices of the target node n i and the indirect node n j ; is the score of the target node n i , is the score of the indirect node n j .

[0130] Step 3.7: Determine the connection lines of the influence relationship of the node margin score on each node in the margin mapping layer through the correlation coefficient and the coupling coefficient, and use the connection weight W to represent this influence:

[0131]

[0132] Among them, is the weight of the margin influence between the target node n i and the indirect node n j , ε is the constraint influence coefficient, is the constraint influence weight between the target node n i and the indirect node n j ; Considering the margin and constraint influences corresponding to the five indicators comprehensively, the margin connection layer obtains the connection weight W between the target node n i and the node n j , reflecting the strength of the node connection in the margin layer; where γ i and the node n j is a normalization coefficient, usually set to a constant value (such as 0.5 or 1.0). g

[0133] Step 3.8: Calculate the power flow propagation influence coefficient according to the connection weight W and the shortest path d ij in the margin mapping layer in Step 3.2. For any index, calculate the flexibility S I of other nodes on the target node after the influence of the power flow propagation:

[0134]

[0135] By calculating the influence of the direct and indirect nodes on the flexibility of the target node, the five flexibility indicators S I of the node are finally obtained, so as to intuitively display the flexibility distribution of the distributed resources of the node in the whole system.

[0136] In this embodiment, through the propagation matrix P ij and the correlation coefficient ρ ij ​and coupling coefficient β ij Calculate the connection weights between each node and the target node, quantify the direct and indirect power flow propagation impacts of each node on the target node. Finally, five scoring metrics of the target node in the system considering both direct and indirect impacts in the system are obtained through calculation, and the metric scoring results are S p = 68, S h = 79, S ef = 65, S r = 80, S el = 65. Generate a flexibility distribution map as shown in Figure 4 shown.

[0137] The present invention proposes a distributed resource flexibility evaluation method based on a two-layer evaluation model, aiming to construct a comprehensive flexibility comprehensive scoring method by combining dynamic operation data, static inherent characteristics, and device health status. This method improves the management efficiency and operation reliability of distributed energy systems by dynamically integrating system characteristics, device health, and time series data, solves the problems of insufficient dynamic adaptability and lack of comprehensive evaluation in existing methods, and provides an efficient and accurate flexibility evaluation for distributed resources.

[0138] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A distributed resource flexibility evaluation method based on a double-layer evaluation model, characterized in that Including: Step 1: Divide the feature data obtained from the distributed energy system into dynamic operation data, equipment health status data, and static inherent feature data. Extract features from each type of data to form a feature matrix and fuse them. Then, obtain the global feature vector by extracting the fused feature matrix through a Transformer. Step 2: Construct a node resource flexibility evaluation framework with the global feature vector as the input of the framework. Design evaluation indicators covering five key dimensions: equipment performance, health status, operation efficiency, accommodation capacity, and elastic capacity. Generate the node flexibility score by calculating each indicator. Step 3: Construct a two-layer evaluation model consisting of a connection layer and a margin score mapping layer. Define the connectivity and parameters between nodes in the distributed energy system through the connection layer. The margin score mapping layer uses the global feature vector and the node resource flexibility evaluation framework to quantify the influence between nodes with the help of the power flow propagation matrix.

2. The distributed resource flexibility evaluation method based on a two-layer evaluation model according to claim 1, characterized in that The specific content of Step 1 is as follows: Step 1.1: Matrixize the obtained dynamic operation data inside the distributed energy system to form a time series data matrix, and perform data cleaning and preprocessing on the time series data matrix. Process the time series data matrix through a local convolutional neural network to obtain a dynamic data feature matrix. Step 1.2: For the equipment health status data of the distributed energy system, construct an original health status data matrix based on the equipment number, equipment usage rate coding, equipment environmental condition coding, and equipment health status coding. Align the matrix linearly repeatedly to the row dimension of the dynamic data feature matrix to obtain a health status data feature matrix. Step 1.3: For the static inherent feature data related to the distributed energy system, construct an initial static data matrix based on the energy type coding, type coding, and geographical location coding. Align the matrix linearly repeatedly to the row dimension of the dynamic data feature matrix to obtain a static data feature matrix. Step 1.4: Extract the dynamic data feature matrix, health status data feature matrix, and static data feature matrix through a global Transformer, model the long-term dependencies and global correlations in the time series data, and form a global feature vector through the unified fusion of multi-level features.

3. The distributed resource flexibility evaluation method based on the double-layer evaluation model according to claim 2, characterized in that The specific content of Step 2 is as follows: Step 2.1: Use the global feature vector extracted in Step 1 as the input, construct equipment performance indicators from two aspects: equipment utilization rate and power output stability, and calculate the equipment performance indicator score. Step 2.2: Use the global feature vector extracted in Step 1 as the input, construct health status indicators from two aspects: failure incidence rate and remaining life, and calculate the health status indicator score. Step 2.3: Use the global feature vector extracted in Step 1 as the input, construct operation efficiency indicators from two aspects: power transmission efficiency and unit energy consumption ratio, and calculate the operation efficiency indicator score. Step 2.4: Use the feature vector extracted in Step 1 as the input, calculate the accommodation capacity indicator according to the renewable energy accommodation ratio. The calculation method is as follows: Among them, P renewable,t is the renewable energy output at time t, β t is the weather or seasonal fluctuation adjustment coefficient, γ t is the grid load demand adjustment coefficient, and P total,t is the total output power; S renewable is the score of the accommodation capacity index, R cmin is the minimum value of R during this period, R c is the minimum value, R cmax is the maximum value of R during this period, and R c is the maximum value, and n is the total number of time periods; Step 2.5: Use the global feature vector extracted in Step 1 as the input, construct elastic capacity indicators from two aspects: load regulation capacity and dynamic allocation capacity of reserve capacity, and calculate the elastic capacity indicator score.

4. The distributed resource flexibility evaluation method based on a two-layer evaluation model according to claim 3, characterized in that Step 2.1 is specifically as follows: Step 2.1.1: Calculate the equipment utilization rate U b , which is the ratio of the actual running time of the equipment to its total usage time within a certain period, while considering non-linear factors: Calculate U b After that, use the min-max normalization method to convert the result into a score: where Δt is the time increment of the device operation, and T available is the total available time of the device; if the device does not operate during a certain period, the R bt value for this period is 0; if the device operates normally, the R bt value is 1; U min is the minimum value of U b during this period, and U max is the maximum value of U b during this period; is the device utilization score; Step 2.1.2: Calculate the power output stability Q b , by calculating the standard deviation of the device's power output, collecting the device's power data at multiple time points, and using the standard deviation formula to measure stability: Among them, P bt is the power output of device b at time t, is the average power output during this period, α is the load factor, L t is the load fluctuation amount, and n is the number of time series; is the power output stability score; Q min is the minimum value of Q during this period, Q b is the minimum value of Q during this period, Q max is the maximum value of Q during this period, Q b is the maximum value; Step 2.1.3: Calculate the device performance metrics, through and weight them to obtain the score of this metric: where w U and w S are the weights of two set indicators respectively, and the sum of the two weights is 1.

5. The distributed resource flexibility evaluation method based on a two-layer evaluation model according to claim 3, wherein Step 2.2 is specifically as follows: Step 2.2.1: Calculate the failure rate, which consists of the number of failures of different devices in multiple time periods, and obtain the final failure rate F through weighted average b : Among them, E f is the severity index of the f-th failure mode, T f is the operating time of the device under the f-th failure mode, T total is the total operating time of the device; is the failure incidence rate score; F min is the minimum value of F b during this period, F max is the maximum value of F b during this period; Step 2.2.2: Calculate the remaining life. For multiple health data, calculate the remaining life through weighted average, and judge the life expectancy L of the device based on the current health status and operation data of the device residual,b : Among them, R bd is the health status score of device b under the d-th condition, and μ b is the original design life of the device; γ and δ are adjustment parameters, which are adjusted according to engineering needs; is the remaining life score; L residual,min is the minimum value of this period L residual,b , and L residual,max is the maximum value of this period L residual,b ; Step 2.2.3: Calculate the health status indicator, which is weighted by and to obtain the score of this indicator: Among them, w F and w L are respectively set as the weights of two indicators.

6. The distributed resource flexibility evaluation method based on a two-layer evaluation model according to claim 3, characterized in that Step 2.3 is specifically as follows: Step 2.3.1: Calculate the power transfer efficiency E p , considering the power loss and non-linear transmission loss in the system, the calculation method is defined as: Among them, P load is the load - end power, P generation is the power generation - end power, β loss is the transmission loss coefficient, L loss is the power loss during transmission, L load is the power demand at the load end; is the power transmission efficiency score; E pmin is the minimum value of E p during this period, E pmax is the maximum value of E p during this period; Step 2.3.2: Calculate the unit energy consumption ratio C u , considering the influences of factors such as equipment energy efficiency, load fluctuation, and climate change, the calculation method is as follows: Among them, E total,t is the total energy consumption at time t, and α t is the energy efficiency adjustment coefficient at time t, and P total is the total power output of the system; is the unit energy consumption ratio score; C umin is the minimum value of C u during this period, and C umax is the maximum value of C u during this period; Step 2.3.3: Calculate the operation efficiency index, which is weighted by and to obtain the score of this index: Among them, and are respectively set as the weights of the two indicators.

7. The distributed resource flexibility evaluation method based on a two-layer evaluation model according to claim 3, wherein Step 2.5 is specifically as follows: Step 2.5.1: Calculate the load regulation ability, that is, calculate the response time of the system to load changes considering factors such as the regulation ability of equipment and the degree of load fluctuation: where, ΔP load is the change in load, Δt is the response time, and σ load is the standard deviation of load fluctuations, and α l is the load fluctuation correction factor; is the load regulation ability score; L rmin is the minimum value of L during this period, and L r is the minimum value of L during this period, and L rmax is the maximum value of L during this period; r is the maximum value of L during this period; Step 2.5.2: Calculate the dynamic allocation ability of reserve capacity, and measure the dynamic allocation ability of reserve capacity considering factors such as equipment failures and the allocation of reserve capacity: Among them, P backup is the spare capacity, λ is the scheduling efficiency coefficient of the spare capacity, and Δt is the time required for deployment; is the score for the dynamic deployment ability of the spare capacity; C backupmin is the minimum value of C backup during this period, and C backupmax is the maximum value of C backup during this period; Step 2.5.3: Calculate the elastic capacity metric, which is weighted by and to obtain the score of this metric: Among them, and are respectively set as the weights of two indicators.

8. The distributed resource flexibility evaluation method based on a double-layer evaluation model according to claim 3, characterized in that Step 3 is specifically as follows: Step 3.1: Construct a connection layer according to the physical structure and line topology of the distributed energy system. Each building is regarded as a node in the connection layer. The power generation equipment, energy storage equipment, and loads in the system are associated according to the building location and corresponding equipment information. The connections in the connection layer represent the actual line connections between building nodes, intuitively showing the direct connection situation of each node in the system; Step 3.2: Extract the feature data of each node in the connection layer, and divide the feature data of each node n i into dynamic operation data, device health status data, and static inherent feature data according to the method in Step 1. Then, perform feature extraction on each type of data to form a feature matrix and fuse them. The fused feature matrix is used to extract the global feature vector through a Transformer; Step 3.3: Use the node resource flexibility evaluation framework in Step 2 to input the global feature vector of each node into the evaluation framework, calculate the node flexibility score of the node, and use it as the score of the node in the margin score mapping layer S = [S p S ef S h S el S r ; Step 3.4: For nodes that have mutual influence but no direct physical connection in the connection layer, the indirect influence propagation between nodes in the margin scoring mapping layer is reflected through five propagation matrices corresponding to five scoring indicators, quantifying how the change of the indirect node state propagates to the corresponding target node through the intermediate nodes in the system and causes an impact: Among them, is the propagation matrix, k = 1, 2, 3, 4, 5; S k” is an element in S, R is the influence matrix, and the five indicators respectively correspond to five kinds of S k' , and at the same time S k' ·R represents the interaction between the margin score and the influence matrix, d ij represents the shortest path between the target node n i and the indirect node n j , Δθ represents the voltage phase angle difference between the two nodes, ΔV represents the voltage amplitude difference between the two nodes, ΔP represents the active power difference between the two nodes, and ΔQ represents the reactive power difference between the two nodes; Step 3.5: When the system is updated, solve the updated propagation matrix P through the iterative propagation algorithm. Calculate the active power flow of each node through the topological data in the connection layer to form a power flow matrix, considering the multi-hop propagation effect of the system. The specific iterative propagation process is defined as: Among them, P (c) represents the propagation matrix of the c-th iteration, F ij is the active power flow from node i to node j, τ is the power flow influence coefficient, P (c-1) is the transpose of the propagation matrix, representing the reverse propagation of the influence; Step 3.6: Calculate five correlation coefficients and five coupling coefficients between the target node and the indirect node: Among them, is the correlation coefficient, is the coupling coefficient, represents node n i and node n j the propagation matrix of the interaction between them and the covariance of, and are respectively the standard deviations of the interaction matrices of the target node n i and the indirect node n j ; is the score of the target node n i and is the score of the indirect node n j ; Step 3.7: Determine the connection lines representing the influence relationship of the node margin score on each node in the margin mapping layer through the correlation coefficient and the coupling coefficient, and use the connection weight W to represent this influence: Among them, is the weight of the margin influence between the target node n i and the indirect node n j ; ε is the constraint influence coefficient. is the weight of the constraint influence between the target node n i and the indirect node n j ; Considering the margin and constraint influences corresponding to the five indicators for the target node n i and the node n j , the connection weight W of the target node n i and the node n j in the margin connection layer is obtained, which reflects the strength of the node connection in the margin layer; where γ g is a normalization coefficient. Step 3.8: According to the margin mapping layer connection weight W and the shortest path d in Step 3.2 ij calculate the power flow propagation influence coefficient. For any index, calculate the flexibility S of other nodes on the target node after being affected by the power flow propagation influence coefficient I : By calculating the direct and indirect node impacts on the flexibility of the target node, five flexibility indicators S of this node are finally obtained I , thus intuitively showing the flexibility distribution of distributed resources of this node within the entire system 9. The distributed resource flexibility evaluation method based on a two-layer evaluation model according to claim 1, characterized in that Step 3.1 is specifically as follows: Step 3.1.1: Define the connection layer G = (N, A, L, C), and use N = {n1, n2,..., n n} to represent the set of nodes in the system. For each node n i assign its relevant physical properties; Step 3.1.2: The attribute set A = {A1, A2,..., An} that constitutes the nodes, and each node n i 's attribute set Ai has physical parameters that describe performance and response capabilities according to different types of devices; Step 3.1.3: Use the set \(L = \{l ij , l ij' ,...\}\) to represent the energy transfer paths between system nodes in the connection layer and the physical connection relationships of interactions between different nodes. \(l ij \) or \(l ij' \) represents the physical connection between node \(n i \) and node \(n j \) or between node \(n i \) and node \(n j' \). Step 3.1.4: Assign actual physical characteristics to the physical connection l between each pair of nodes in the connection layer ij ∈(n i ,n j ) and define its physical constraint set C ij ={P max ,L loss}, where P max represents the maximum power transfer capacity of the connection line, and L loss represents the loss during the power transfer process.

Images