Method for evaluating distributed photovoltaic bearing capacity of power distribution network based on data driving

Abstract

The invention relates to a power distribution network distributed photovoltaic bearing capacity assessment method based on data driving, and the method comprises the steps: employing a block attention-enhanced GRU neural network, and generating photovoltaic output time sequence prediction data of a target assessment region through spatial-temporal data; constructing a boundary condition model in combination with power grid topology and load characteristics, wherein the model comprises electric power and electric quantity balance, voltage, line capacity and harmonic current constraints; integrating related data to construct an evaluation database, and covering the steps of topology analysis, data mapping and the like; a multi-level evaluation result is obtained through database-driven safety, adequacy and economic index cooperative calculation; according to the evaluation result, a spatial distribution map fusing voltage grade checking, reverse power transmission early warning and bearing capacity partitioning is generated, and the regional bearing capacity grade is dynamically labeled, through the steps, a scientific and real-time decision basis is provided for planning and operation of the power distribution network, and the acceptance capacity and operation stability of the power distribution network for distributed photovoltaic power generation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network carrying capacity assessment, and particularly to a data-driven assessment method for the carrying capacity of distributed photovoltaic in the distribution network. Background Art

[0002] With the increasing global demand for clean energy, distributed photovoltaic, as a sustainable energy solution, has been more and more widely applied in the distribution network. The access of distributed photovoltaic to the distribution network not only helps to reduce the dependence on traditional fossil energy, reduce carbon emissions, but also improves the reliability and flexibility of energy supply. However, distributed photovoltaic has the characteristics of intermittency and volatility, and its large-scale access brings many challenges to the safe, stable and economic operation of the distribution network. Therefore, it is urgent to evaluate the carrying capacity of distributed photovoltaic in the distribution network.

[0003] For example, the Chinese invention patent with the publication number CN115774984A discloses a method for evaluating the carrying capacity of distributed power sources in a medium-voltage flexible interconnected line, including establishing a thermal stability evaluation model, a voltage deviation checking model, a harmonic checking model and a carrying capacity level division model for distributed power sources accessing the medium-voltage flexible interconnected line; aiming at the problem of evaluating the carrying capacity of distributed power sources in the medium-voltage flexible interconnected line, considering the influence of the medium-voltage flexible interconnected device on the power flow, voltage, harmonics, etc. of the medium-voltage line, modeling and correcting the thermal stability evaluation, voltage deviation checking, harmonic checking, and carrying capacity level division in the "Guidelines for Evaluating the Carrying Capacity of Distributed Power Sources Connected to the Grid", which is applicable to the evaluation of the carrying capacity of distributed power sources in the medium-voltage flexible interconnected line. The above patent is difficult to capture the complex change rules of photovoltaic output at different time and space scales, resulting in the inability to provide accurate reference for distribution network planning and operation, and the above patent mainly focuses on the medium-voltage flexible interconnected line, and the considered grid topology is relatively single. The actual distribution network has complex and diverse topological structures and load characteristics. Only evaluating the medium-voltage flexible interconnected line cannot fully reflect the actual operation status of the entire distribution network, and the generality and applicability of the evaluation results are limited. Therefore, there is an urgent need for a method for evaluating the carrying capacity of distributed photovoltaic in the distribution network that comprehensively considers the complex and diverse topological structures and load characteristics of the actual distribution network and fully captures the complex change rules of photovoltaic output at different time and space scales. Summary of the Invention

[0004] Based on the above technical problems, the present invention provides a data-driven assessment method for the carrying capacity of distributed photovoltaic in the distribution network. The method is specifically as follows:

[0005] Use a GRU neural network enhanced by block attention to generate time series prediction data of photovoltaic output in the target evaluation area;

[0006] Based on the above-mentioned photovoltaic output time series prediction data, combined with the power grid topology and load characteristics, construct a boundary condition model for evaluating the distributed photovoltaic carrying capacity;

[0007] Integrate the data related to the evaluation of the distributed photovoltaic carrying capacity, and construct an evaluation database that matches the boundary condition model;

[0008] Through the collaborative calculation of the safety, adequacy, and economy indicators driven by the evaluation database, obtain multi-level evaluation results;

[0009] According to the multi-level evaluation results, generate a spatial distribution map that integrates voltage level checking, reverse power transmission warning, and carrying capacity zoning, and dynamically mark the regional carrying capacity level in the spatial distribution map.

[0010] As a preferred implementation manner of the present invention, a GRU neural network enhanced by block attention is used to generate the photovoltaic output time series prediction data of the target evaluation area, specifically:

[0011] Obtain spatio-temporal data, including photovoltaic irradiance, ambient temperature, and cloud cover rate, input the spatio-temporal data into the GRU unit, and generate a hidden state sequence H = {h1, h2,..., h t ,…,h T} through the gating mechanism, where h t ∈ is the d-dimensional hidden vector at time t; is the number field; T is the length of the hidden state sequence;

[0012] Divide the hidden state sequence H into K = [T / w] blocks according to the time window length w, and calculate the short-range attention for the k-th block H k including:

[0013] Calculate the intra-block attention weight, which is expressed by the formula:

[0014]

[0015] Where:

[0016] Q k = H k W Q ;

[0017] K k = H k W K ;

[0018] In the formula, is the intra-block attention weight of the k-th block H k , Q k is the query matrix of the k-th block H k ; Kk is the key matrix of the k-th block H k ; W Q and W K are the weight matrices corresponding to the query matrix and the key matrix respectively,

[0019] Generate short-range context features based on the intra-block attention weights, which can be expressed by the formula:

[0020]

[0021] where:

[0022] V k = H k W V ;

[0023] In the formula, is the short-range context feature of the k-th block H k ; V k is the value matrix of the k-th block H k ; W V is the weight matrix corresponding to the value matrix,

[0024] Concatenate the short-range context features of each block in chronological order to Each time step t corresponds to a unique short-range feature

[0025] Perform average pooling on the hidden state sequence H with a window size of w and a step size of s, which can be expressed by the formula:

[0026]

[0027] In the formula, C long is the long-range feature of the hidden state sequence, and each time step t corresponds to a unique long-range feature

[0028] Fuse the short-range and long-range features for each time step t to obtain the fused feature, which can be expressed by the formula:

[0029]

[0030]

[0031] In the formula, α t is the dynamic gating weight; σ is the Sigmoid function; W α is the scalar weight matrix, b α is the initial value of the gating bias, is the fused feature,

[0032] The fused features are input into the fully connected layer to obtain the predicted data of the photovoltaic output time series, which is expressed by the formula as:

[0033]

[0034] In the formula, is the predicted data of the photovoltaic output time series at time step t, W f is the fully connected weight matrix, b f is the fully connected bias term,

[0035] As a preferred embodiment of the present invention, based on the predicted data of the photovoltaic output time series, combined with the power grid topology and load characteristics, a boundary condition model for evaluating the distributed photovoltaic carrying capacity is constructed, specifically as follows:

[0036] The distribution network corresponding to the target evaluation area is abstracted into a graph structure G=(N, E), where N is the set of nodes, with a total of n nodes, N={1, 2,..., n}; E is the set of branches, E={(i, j')|i, j'∈N}, and i and j' are the indices of different nodes;

[0037] Based on the graph structure, a branch parameter matrix is constructed, including the impedance matrix Z and the capacity matrix S max , which is expressed by the formula as:

[0038] Z ij' =R ij' +jX ij' ;

[0039]

[0040] In the formula, Z ij' is the line impedance between node i and node j'; is the set of n-order square matrices of complex numbers; R ij' is the line resistance between node i and node j'; Y ij' is the line admittance between node i and node j'; X ij' is the line reactance between node i and node j'; is the maximum apparent power that the branch between node i and node j' allows to transmit;

[0041] Based on the load of each node, the total load power is calculated, which is expressed by the formula as:

[0042]

[0043] Q load (t)=Pload (t)·tan(φ i );

[0044] Wherein:

[0045]

[0046] In the formula, P load (t) is the active load power at time step t; α' ik' is the proportion of the k'-th type of load on node i; L k' (t) is the normalized daily load of the k'-th type of load at time step t; μ k' is the load peak factor of the k'-th type of load; a k' , b' k' and λ k' are fitting parameters; ε(t) is the Gaussian white noise at time step t; Q load (t) is the reactive load power at time step t; φ i is the power factor angle on node i;

[0047] Based on the power grid topology and load characteristics, a boundary condition model is constructed. The boundary condition model includes power and energy balance constraints, voltage constraints, line capacity constraints, and harmonic circuit constraints. Wherein:

[0048] The power and energy balance constraints are expressed by the formula:

[0049] P PV (t) + P grid = P load (t);

[0050] Q PV (t) + Q grid = Q load (t);

[0051] Q PV (t) = P PV (t)·tan(φ i );

[0052] In the formula, P grid is the active power injected from the superior power grid; P PV (t) is the active power output of the distributed photovoltaic at time step t, that is, the value obtained from the photovoltaic output time series prediction data at the corresponding moment; Q grid is the reactive power injected from the superior power grid;; Q PV (t) is the reactive power output of the distributed photovoltaic at time step t;

[0053] The voltage constraints are expressed by the formula:

[0054]

[0055] In the formula, is the lower voltage limit of node i; is the upper voltage limit of node i; V' i (t) is the voltage amplitude of node i at time step t; V' nom is the rated voltage;

[0056] The line capacity constraint is expressed by the formula:

[0057]

[0058] In the formula, S ij' is the apparent power allowed to be transmitted by the branch between node i and node j';

[0059] The harmonic current constraint is expressed by the formula:

[0060]

[0061] In the formula, I h',PV is the h'th harmonic current; I h',limit is the allowable value of harmonic current.

[0062] As a preferred embodiment of the present invention, the data related to the evaluation of the distributed photovoltaic carrying capacity includes power grid basic data, equipment nameplate parameters, operation measured data, spatio-temporal data, and original safety boundary parameters, where:

[0063] The power grid basic data includes the primary wiring diagram and equivalent impedance diagram of the target distribution network and the measured short-circuit capacity table of each voltage level bus; the equipment nameplate parameters include the rated capacity of the transformer, the measured values of line resistance and reactance, the adjustable range of the power factor of the photovoltaic inverter, and the rated breaking current of the circuit breaker; the operation measured data includes the historical daily load curve on the load side, the sequential measured values of bus voltage and current, and the grid-connected capacity and actual output sampling data of the distributed photovoltaic site; the original safety boundary parameters include voltage deviation limits, current-carrying capacity limits of lines and transformers, and allowable harmonic current values.

[0064] As a preferred embodiment of the present invention, the data related to the evaluation of the distributed photovoltaic carrying capacity is integrated to construct an evaluation database matching the boundary condition model, including topological analysis, data mapping, constraint coding, and feature association, where:

[0065] The topological analysis specifically generates an adjacency matrix based on the graph structure and stores the node-branch connection relationship; the data mapping specifically calibrates the photovoltaic output time series prediction data and the active power output of the distributed photovoltaic according to the time stamp to construct a time series key-value pair; the constraint encoding specifically converts the boundary condition model into a database query rule; the feature association specifically performs an association index on the integrated features and the active and reactive powers injected from the superior power grid in the power and energy balance constraint.

[0066] As a preferred embodiment of the present invention, through the collaborative calculation of the security, adequacy, and economy indicators driven by the evaluation database, a multi-level evaluation result is obtained, where:

[0067] The security indicators include the voltage violation probability and the line overload risk, and are expressed by the formula:

[0068]

[0069] In the formula, VR is the voltage violation probability; λ' ij' (t) is the branch load rate;

[0070] The adequacy indicators include the photovoltaic penetration rate and the harmonic carrying margin, and are expressed by the formula:

[0071]

[0072] In the formula, η(t) is the photovoltaic penetration rate; M' is the harmonic carrying margin;

[0073] The economy indicators include the network loss cost and the return on investment, and are expressed by the formula:

[0074]

[0075] Cost = P loss × electricity price;

[0076]

[0077] In the formula, P loss is the total network loss; Cost is the network loss cost; ROI is the return on investment.

[0078] As a preferred embodiment of the present invention, according to the multi-level evaluation result, a spatial distribution map integrating voltage level checking, reverse power transmission warning, and bearing capacity zoning is generated, and the regional bearing capacity level is dynamically marked in the spatial distribution map, where:

[0079] Generating the integrated voltage level checking according to the security indicator evaluation result specifically is:

[0080] Count the nodes with voltage out-of-limit, map the graph structure corresponding to the nodes with voltage out-of-limit to the GIS platform, and perform heat map rendering according to the evaluation results and preset rendering rules. The specific rendering rules are as follows:

[0081] When and render as a red block;

[0082] When render as a yellow block;

[0083] When and render as a green block;

[0084] In the formula, is the voltage out-of-limit probability threshold; △V is the safety margin parameter;

[0085] Generating a reverse power transmission warning according to the evaluation result of the adequacy index is specifically: when η(t) > 100%, mark the reverse power flow path in the corresponding topology diagram of the GIS platform and push a warning message to the dispatching system;

[0086] Generating a bearing capacity partition according to the evaluation results of the safety index and the economic index and dynamically marking the regional bearing capacity level is specifically:

[0087] When or λ' ij' (t) > λ' limit , set it as the red level, λ' limit is the line load rate exceeding the thermal stability limit of the equipment;

[0088] When and η avg (t) > η base (t), set it as the yellow level, C low and C high are the lower limit and upper limit of the network loss cost respectively; η avg (t) is the average PV penetration rate; η base (t) is the PV penetration rate reference value;

[0089] When and λ' ij' (t) < λ' safe and Cost ≤ C opt , set it as the green level, λ' safe is the safe branch load rate, C opt is the optimal network loss threshold.

[0090] As a preferred embodiment of the present invention, the method further includes real-time updating of the photovoltaic output time series prediction data through an API interface, and refreshing each region in the spatial distribution map within a preset time period.

[0091] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a data-driven method for evaluating the distributed photovoltaic carrying capacity of a distribution network as described in any embodiment of the present invention.

[0092] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a data-driven method for evaluating the distributed photovoltaic carrying capacity of a distribution network as described in any embodiment of the present invention.

[0093] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0094] 1. The present invention provides a data-driven method for evaluating the distributed photovoltaic carrying capacity of a distribution network. By obtaining spatio-temporal data and inputting it into a GRU unit to generate a hidden state sequence, calculating short-range attention in blocks, combining long-range features and fusing them, and finally inputting them into a fully connected layer to output photovoltaic output time series prediction data, the accuracy of photovoltaic output time series prediction is significantly improved, and the time dependence and local features of the data can be captured more accurately, providing a reliable data basis for subsequent evaluation of distributed photovoltaic carrying capacity.

[0095] 2. The present invention provides a data-driven method for evaluating the distributed photovoltaic carrying capacity of a distribution network. The distribution network is abstracted into a graph structure, a branch parameter matrix is constructed, and the total load power is calculated to form a boundary condition model containing various constraints; various relevant data are integrated, and an evaluation database is constructed through topological analysis, data mapping, constraint coding, and feature association, realizing a comprehensive and accurate description of the operation of the distribution network. The evaluation database effectively integrates and manages relevant data, improves data utilization efficiency, and provides strong support for the accurate evaluation of distributed photovoltaic carrying capacity.

[0096] 3. The present invention provides a data-driven method for evaluating the distributed photovoltaic carrying capacity of a distribution network. According to the multi-level evaluation results, a spatial distribution map integrating voltage level checking, reverse power transmission warning, and carrying capacity zoning is generated, and the regional carrying capacity level is dynamically marked. The evaluation results are presented in an intuitive visualization method, which is convenient for grid operation personnel and decision-makers to quickly understand the distributed photovoltaic carrying capacity of each region of the distribution network and timely discover potential problems; at the same time, the data is updated in real time and the map is refreshed through an API interface. The real-time update ensures the timeliness and adaptability of the evaluation results and can better cope with the dynamic changes of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 It is the flowchart of the method of the embodiment of the present invention. Specific embodiments

[0098] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0099] Embodiment 1:

[0100] Embodiment 1 of the present invention discloses a method for evaluating the distributed photovoltaic carrying capacity of a distribution network based on data driving. The method includes:

[0101] S1. Using a GRU neural network enhanced by block attention to generate time series prediction data of photovoltaic output in the target evaluation area;

[0102] Since the neural network that relies on the gating mechanism to regulate the information flow has limited ability to perceive local fine-grained features, and in time series modeling scenarios such as power grid load forecasting, the short-term fluctuation characteristics and long-term evolution laws of key nodes are equally important. Therefore, in this embodiment, the GRU neural network constructs a hybrid neural network architecture with dual feature perception capabilities by parallelly integrating short-range and long-range attention mechanisms, and improves the model prediction accuracy through the collaborative optimization of local feature enhancement and global dependence modeling. Specifically:

[0103] S11. Obtain spatio-temporal data, including photovoltaic irradiance, ambient temperature, and cloud cover rate, input the spatio-temporal data into the GRU unit, and generate a hidden state sequence H = {h1, h2,..., h t ,..., h T} through the gating mechanism, where is the d-dimensional hidden vector at time t, and the hidden state dimension d controls the capacity of the neural network. The higher the dimension, the stronger the feature expression ability; is the number field; T is the total length of the hidden state sequence;

[0104] S12. Short-range attention (block alignment):

[0105] Divide the hidden state sequence H into K = [T / w] blocks according to the time window length w (w determines the time window range of short-range attention and needs to match the photovoltaic fluctuation period. In this embodiment, w = 2 m , and m is the time difference between sunrise and sunset), and calculate the short-range attention for the kth block H k , including:

[0106] Calculate the attention weight within the block, expressed as:

[0107]

[0108] in:

[0109] Q k =H k W Q ;

[0110] K k =H k W K ;

[0111] In the formula, For the kth block H k The intra-block attention weights, Q k For the kth block H k The query matrix K k For the kth block H k The bond matrix of Q and W K are the weight matrices corresponding to the query matrix and the key matrix respectively,

[0112] The short-range context features are generated based on the attention weight within the block, which can be expressed as:

[0113]

[0114] in:

[0115]

[0116] In the formula, For the kth block H k Short-range context features of V k For the kth block H k The value matrix of V is the value matrix corresponding to the weight matrix,

[0117] The short-range context features of each block are concatenated in time order as Each time step t corresponds to a unique short-range feature

[0118] S13, long-range attention (sliding pooling alignment):

[0119] The hidden state sequence H is subjected to average pooling with a window size of w and a step size of s (s controls the smoothness of long-range features. The smaller the step size, the more details are retained. In this embodiment, s = [w / 2]), which can be expressed as:

[0120]

[0121] where C long is the long-range feature of the hidden state sequence, and each time step t corresponds to a unique long-range feature

[0122] In this embodiment, the alignment rule corresponding to the long-range attention is that the center point of the pooling window is aligned with the time step t, and zero padding is used at the edges to ensure the temporal length of C long , that is, the length T of the hidden state sequence;

[0123] S14. Dynamic feature fusion (time step alignment):

[0124] Fuse the short-range and long-range features for each time step t to obtain the fused feature, which is expressed by the formula as:

[0125]

[0126] where α t is the dynamic gating weight, which reflects the importance of local fluctuations and global trends in real time; σ is the Sigmoid function; W α is the scalar weight matrix, b α is the initial value of the gating bias, used to initially balance the short-range / long-range weights, and is set to 0.5 in this embodiment; is the fused feature,

[0127] Furthermore, through the dynamic gating weight α t realize the adaptive coupling of spatio-temporal features, automatically adjust the contribution degree of short / long-range attention according to the features at the current moment, when α t →1, more dependent on short-range attention (capturing transient fluctuations, such as cloud occlusion), when α t →0, more dependent on long-range attention (capturing the overall trend, such as day-night cycle);

[0128] S15. Photovoltaic output time series prediction:

[0129] Input the fused feature into the fully connected layer to obtain the photovoltaic output time series prediction data, which is expressed by the formula as:

[0130]

[0131] where is the photovoltaic output time series prediction data, W f is the fully connected weight matrix, b f is the fully connected bias term,

[0132] S16. Further, in this embodiment, the loss function L of the neural network adopts a piecewise weighted mean square error to enhance the prediction accuracy of the sunrise / sunset period, which is expressed by the formula as follows:

[0133]

[0134] In the formula, γ t is the piecewise weight factor, which is used to dynamically adjust the loss weight according to the business importance of the time period and enhance the prediction accuracy of the key period (such as sunrise / sunset when the photovoltaic output changes rapidly). In this embodiment, the 24-hour system is adopted, and the sunrise period is from 6:00 to 8:00, and the sunset period is from 16:00 to 18:00; β t is the volatility-sensitive weight, which is used to adaptively adjust the weight according to the historical output fluctuation amplitude and enhance the sensitivity to the mutation period; △P t is the output change amount between adjacent time steps t and t-1, which reflects the output fluctuation intensity; max|△P τ | is the maximum absolute value of the output change in the whole sequence, which is used for normalization; τ is the temporary variable for traversing the time step; P t is the true data of the photovoltaic output time series;

[0135] S2. Based on the predicted data of the photovoltaic output time series, combined with the power grid topology and load characteristics, construct a boundary condition model for evaluating the distributed photovoltaic carrying capacity;

[0136] S21. Power grid topology:

[0137] Abstract the distribution network corresponding to the target evaluation area into a graph structure G=(N, E), where N is the set of nodes, with a total of n nodes, N={1, 2,..., n}; E is the set of branches, E={(i, j')|i, j'∈N}, and i and j' are the indexes of different nodes;

[0138] Based on the graph structure, construct a branch parameter matrix, including an impedance matrix Z and a capacity matrix S max , which is expressed by the formula as follows:

[0139] Z ij' =R ij' +jX ij' ;

[0140]

[0141]

[0142] In the formula, Z ij' is the line impedance between node i and node j'; is the set of n - order square matrices of complex numbers; R ij' is the line resistance between node i and node j'; Y ij' is the line admittance between node i and node j'; X ij' is the line reactance between node i and node j'; is the maximum apparent power that the branch between node i and node j' allows to transmit;

[0143] S22. Load characteristics:

[0144] The total load power is calculated based on the loads of each node and is expressed by the formula:

[0145]

[0146] Q load (t) = P load (t)·tan(φ i );

[0147] Where:

[0148]

[0149] In the formula, P load (t) is the active load power at time step t; α' ik' is the proportion of the k'-th type of load on node i; L k' (t) is the normalized daily load of the k'-th type of load at time step t; μ k' is the load peak factor of the k'-th type of load; a k' , b' k' and λ k' are fitting parameters; ε(t) is the Gaussian white noise at time step t; Q load (t) is the reactive load power at time step t; φ i is the power factor angle on node i;

[0150] S23. Based on the power grid topology and load characteristics, a boundary condition model is constructed. The boundary condition model includes power and energy balance constraints, voltage constraints, line capacity constraints, and harmonic circuit constraints. Where:

[0151] The power and energy balance constraints are expressed by the formula:

[0152] P PV (t)+P grid = P load (t);

[0153] Q PV (t)+Q grid = Q load (t);

[0154] Q PV Q(t) = P PV (t)·tan(φ i );

[0155] Wherein, P grid is the active power injected from the superior power grid; P PV (t) is the active power output of the distributed PV at time step t, that is, the value corresponding to the time obtained from the PV output time series prediction data; Q grid is the reactive power injected from the superior power grid; Q PV (t) is the reactive power output of the distributed PV at time step t;

[0156] The voltage constraint is expressed by the formula:

[0157]

[0158] Wherein, is the lower voltage limit of node i; is the upper voltage limit of node i, set based on the allowable deviation range of the supply voltage specified by the national standard (such as GB / T 12325); V' i (t) is the voltage amplitude of node i at time step t; V' nom is the rated voltage;

[0159] The line capacity constraint is expressed by the formula:

[0160]

[0161] Wherein, S ij' is the apparent power that the branch between node i and node j' is allowed to transmit;

[0162] The harmonic current constraint is expressed by the formula:

[0163]

[0164] Wherein, I h',PV is the h'-th harmonic current; I h',limit is the allowable value of the harmonic current;

[0165] S3. Integrate the relevant data for the assessment of the distributed PV carrying capacity, and construct an assessment database that matches the boundary condition model;

[0166] S31. The relevant data for the assessment of the distributed PV carrying capacity includes grid basic data, equipment nameplate parameters, operation measured data, spatio-temporal data, and original safety boundary parameters, where:

[0167] The basic power grid data includes the primary wiring diagram of the target distribution network, the equivalent impedance diagram, and the measured short-circuit capacity table of busbars at each voltage level; the equipment nameplate parameters include the rated capacity of the transformer, the measured values of the line resistance and reactance, the adjustable range of the power factor of the photovoltaic inverter, and the rated breaking current of the circuit breaker; the measured operation data includes the historical daily load curve on the load side, the time-series measured values of the bus voltage and current, and the grid-connected capacity and actual output sampling data of the distributed photovoltaic sites; the original safety boundary parameters include the voltage deviation limit, the current-carrying capacity limit of the line and transformer, and the allowable harmonic current value.

[0168] S32. Construct an evaluation database that matches the boundary condition model through topology analysis, data mapping, constraint encoding, and feature association, where:

[0169] The topology analysis is specifically to generate an adjacency matrix based on the graph structure and store the node-branch connection relationship; the data mapping is specifically to calibrate the time-series prediction data of photovoltaic output with the active power output of the distributed photovoltaic according to the time stamp and construct a time-series key-value pair; the constraint encoding is specifically to convert the boundary condition model into a database query rule; the feature association is specifically to perform an associative index on the fusion feature and the active power and reactive power injected from the superior power grid in the power and energy balance constraint.

[0170] S4. Drive the collaborative calculation of the security, adequacy, and economy indicators through the evaluation database to obtain multi-level evaluation results.

[0171] S41. The security indicators include the voltage over-limit probability and the line overload risk, and are expressed by the formula:

[0172]

[0173] In the formula, VR is the voltage over-limit probability; λ' ij' (t) is the branch load rate;

[0174] S42. The adequacy indicators include the photovoltaic penetration rate and the harmonic carrying margin, and are expressed by the formula:

[0175]

[0176] In the formula, η(t) is the photovoltaic penetration rate; M' is the harmonic carrying margin;

[0177] S43. The economy indicators include the network loss cost and the return on investment, and are expressed by the formula:

[0178]

[0179] Cost = P loss × electricity price;

[0180]

[0181] In the formula, P loss is the total network loss; Cost is the network loss cost; ROI is the return on investment;

[0182] S5. Generate a spatial distribution map that integrates voltage level verification, reverse power transmission warning, and bearing capacity zoning based on the multi-level evaluation results, and dynamically mark the regional bearing capacity level in the spatial distribution map;

[0183] S51. Generating the integrated voltage level verification according to the safety index evaluation result is specifically as follows:

[0184] Count the voltage over-limit nodes, map the graph structure corresponding to the voltage over-limit nodes to the GIS platform, and perform heat map rendering according to the evaluation results and preset rendering rules. The specific rendering rules are as follows:

[0185] When and render as a red block;

[0186] When render as a yellow block;

[0187] When and render as a green block;

[0188] In the formula, is the voltage over-limit probability threshold; △V is the safety margin parameter;

[0189] S52. Generating the reverse power transmission warning according to the adequacy index evaluation result is specifically as follows: When η(t) > 100%, mark the reverse power flow path in the corresponding topology map on the GIS platform, and push a warning message to the dispatching system. Preferably, mark the reverse power flow path with a purple flashing arrow (not red, yellow, or green) to avoid color conflicts with voltage / carrying capacity;

[0190] S53. Generating the bearing capacity zoning and dynamically marking the regional bearing capacity level according to the safety index and economic index evaluation results is specifically as follows:

[0191] When or λ' ij' (t) > λ' limit , set it as the red level, and access is prohibited. λ' limit is the line load rate exceeding the thermal stability limit of the equipment;

[0192] When and η avg (t) > η base (t), set it as the yellow level, and it needs to be modified before access. C low and Chigh are the lower limit and upper limit of the network loss cost respectively; η avg (t) is the average PV penetration rate; η base (t) is the reference value of the PV penetration rate;

[0193] When and λ' ij' (t) < λ' safe and Cost ≤ C opt it is set as the green level and recommended for connection. λ' safe is the safety branch load rate, and C opt is the optimal network loss threshold, which is obtained by analyzing the inflection point of the network loss cost - PV capacity curve (second derivative method);

[0194] To avoid overlapping with the color blocks of S51, the red level is filled with diagonal lines, the yellow level is marked with a dotted box, and the green level is represented by a solid box + internal dot matrix density;

[0195] S6. Real - time update the PV output time - series prediction data through the API interface, and refresh each area in the spatial distribution map within a preset time period.

[0196] Embodiment 2:

[0197] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a data - driven method for evaluating the distributed PV carrying capacity of a distribution network as described in any embodiment of the present invention.

[0198] Embodiment 3:

[0199] This embodiment provides a computer - readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a data - driven method for evaluating the distributed PV carrying capacity of a distribution network as described in any embodiment of the present invention.

[0200] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of the present invention by the same token.

Claims

1. A data-driven distributed photovoltaic carrying capacity assessment method for distribution networks, characterized in that: The method is specifically as follows: The GRU neural network with block attention enhancement is used to generate the photovoltaic output time series prediction data of the target assessment area; Based on the photovoltaic output time series prediction data, combined with the grid topology and load characteristics, a boundary condition model for distributed photovoltaic carrying capacity assessment is constructed; Integrate the relevant data of distributed photovoltaic carrying capacity assessment and construct an assessment database matching the boundary condition model; The evaluation database drives the collaborative calculation of safety, adequacy and economy indicators to obtain multi-level evaluation results; According to the multi-level evaluation results, a spatial distribution map integrating voltage level verification, reverse power transmission warning and bearing capacity zoning is generated, and the regional bearing capacity levels are dynamically marked in the spatial distribution map.

2. According to claim 1, a data-driven method for evaluating the distributed photovoltaic carrying capacity of a distribution network is characterized in that: The GRU neural network with block attention enhancement is used to generate the photovoltaic output time series prediction data of the target evaluation area, specifically: Acquire spatiotemporal data, including photovoltaic irradiance, ambient temperature, and cloud coverage, input the spatiotemporal data into the GRU unit, and generate a hidden state sequence H = {h1,h2,…,h t ,…,h T }, where is the d-dimensional latent vector at time t; is the number field; T is the length of the hidden state sequence; Divide the hidden state sequence H into K = [T / w] blocks according to the time window length w, and for the kth block H k Compute short-range attention, including: Calculate the attention weight within the block, expressed as: in: Q k =H k W Q ; K k =H k W K ; In the formula, For the kth block H k The intra-block attention weights, Q k For the kth block H k The query matrix K k For the kth block H k The bond matrix of Q and W K are the weight matrices corresponding to the query matrix and the key matrix respectively, The short-range context features are generated based on the attention weight within the block, which can be expressed as: in: V k =H k W V ; In the formula, For the kth block H k Short-range context features of V k For the kth block H k The value matrix of V is the value matrix corresponding to the weight matrix, The short-range context features of each block are concatenated in time order as Each time step t corresponds to a unique short-range feature The hidden state sequence H is averagely pooled with a window size of w and a step size of s, which can be expressed as: In the formula, C long is the long-range feature of the hidden state sequence, and each time step t corresponds to a unique long-range feature For each time step t, the short-range and long-range features are fused to obtain the fused features, which can be expressed as: In the formula, α t is the dynamic gating weight; σ is the Sigmoid function; W α is the scalar weight matrix, b α is the initial value of the gate bias, To fusion features, The fused features are input into the fully connected layer to obtain the photovoltaic output time series prediction data, which is expressed as follows: In the formula, is the photovoltaic output time series prediction data at time step t, W f is the fully connected weight matrix, b f is the fully connected bias term, 3. A data-driven distributed photovoltaic carrying capacity assessment method for distribution network according to claim 2, characterized in that: Based on the photovoltaic output time series prediction data, combined with the grid topology and load characteristics, a boundary condition model for distributed photovoltaic carrying capacity assessment is constructed, specifically: The distribution network corresponding to the target evaluation area is abstracted into a graph structure G = (N, E), where N is a node set, with a total of n nodes, N = {1, 2, ..., n}; E is a branch set, E = {(i, j') | i, j'∈N}, i and j' are indexes of different nodes; Construct branch parameter matrix based on graph structure, including impedance matrix Z and capacity matrix S max , expressed as: In the formula, Z ij' is the line impedance between node i and node j'; is a set of n-order square matrices of complex numbers; R ij' is the line resistance between node i and node j'; Y ij' is the line admittance between node i and node j'; X ij' is the line reactance between node i and node j'; is the maximum apparent power allowed to be transmitted by the branch between node i and node j'; The total load power is calculated based on the load of each node and expressed as: Q load (t)=P load (t)·and(ϕ i ): in: Where P load (t) is the active load power at time step t; α' ik' is the proportion of the k'th type of load on node i; L k' (t) is the normalized daily load of the k'th type of load at time step t; μ k' is the load peak factor of the k'th type of load; a k' , b' k' and λ k' is the fitting parameter; ε(t) is the Gaussian white noise at time step t; Q load (t) is the reactive load power at time step t; φ i is the power factor angle at node i; Based on the grid topology and load characteristics, a boundary condition model is constructed, which includes power balance constraints, voltage constraints, line capacity constraints and harmonic circuit constraints, where: The power balance constraint is expressed as: P PV (t)+P grid =P load (t); Q PV (t)+Q grid =Q load (t); Q PV (t)=P PV (t)·and(ϕ i ): Where P grid is the active power injected from the upper power grid; P PV (t) is the active power output of distributed photovoltaic power generation at time step t, that is, the value at the corresponding moment is obtained from the photovoltaic power generation time series prediction data; Q grid is the reactive power injected from the upper power grid; Q PV (t) is the reactive power output of distributed photovoltaic power generation at time step t; The voltage constraint is expressed as: In the formula, is the voltage lower limit of node i; is the voltage upper limit of node i; V' i (t) is the voltage amplitude of node i at time step t; V' nom is the rated voltage; The line capacity constraint is expressed in the formula: In the formula, S ij' is the apparent power allowed to be transmitted by the branch between node i and node j'; The harmonic current constraint is expressed as: In the formula, I h',PV is the h'th harmonic current; I h',limit is the allowable value of harmonic current.

4. A data-driven distributed photovoltaic carrying capacity assessment method for distribution network according to claim 3, characterized in that: The distributed photovoltaic carrying capacity assessment related data include basic grid data, equipment nameplate parameters, operational measured data, spatiotemporal data and original safety boundary parameters, among which: The basic data of the power grid include the primary wiring diagram and equivalent impedance diagram of the target distribution network and the measured table of bus short-circuit capacity of each voltage level; the equipment nameplate parameters include the rated capacity of the transformer, the measured values ​​of line resistance and reactance, the adjustable range of the power factor of the photovoltaic inverter and the rated breaking current of the circuit breaker; the measured operation data include the historical daily load curve of the load side, the time series measured values ​​of the bus voltage and current, and the grid-connected capacity and actual output sampling data of the distributed photovoltaic site; the original parameters of the safety boundary include the voltage deviation limit, the current carrying capacity limit of the line and transformer, and the allowable value of the harmonic current.

5. A data-driven distributed photovoltaic carrying capacity assessment method for distribution network according to claim 4, characterized in that: Integrate the relevant data of distributed photovoltaic carrying capacity assessment and build an assessment database matching the boundary condition model, including topology analysis, data mapping, constraint coding and feature association, where: The topology analysis is specifically to generate an adjacency matrix based on the graph structure and store the node-branch connection relationship; the data mapping is specifically to calibrate the photovoltaic output timing prediction data with the active output of distributed photovoltaics according to timestamps to construct timing key-value pairs; the constraint encoding is specifically to convert the boundary condition model into a database query rule; the feature association is specifically to associate and index the active power and reactive power injected from the upper power grid in the fusion feature and the power balance constraint.

6. A data-driven method for evaluating the carrying capacity of distributed photovoltaic power distribution networks according to claim 5, characterized in that: The evaluation database drives the collaborative calculation of safety, adequacy and economy indicators to obtain multi-level evaluation results, where: The safety index includes voltage over-limit probability and line overload risk, which can be expressed as: Where, VR is the voltage over-limit probability; λ' ij' (t) is the branch load rate; The adequacy index includes photovoltaic penetration and harmonic carrying margin, which can be expressed as: Where η(t) is the photovoltaic penetration rate; M' is the harmonic carrying margin; The economic indicators include network loss cost and return on investment, which can be expressed as follows: Cost = P loss ×Electricity price; Where P loss is the total network loss; Cost is the network loss cost; ROI is the return on investment.

7. A data-driven method for evaluating the carrying capacity of distributed photovoltaic power distribution networks according to claim 6, characterized in that: According to the multi-level evaluation results, a spatial distribution map integrating voltage level verification, reverse power transmission warning and bearing capacity zoning is generated, and the regional bearing capacity level is dynamically marked in the spatial distribution map, wherein: The fusion voltage level check generated based on the safety index evaluation results is as follows: The voltage exceeding limit nodes are counted, and the graph structure corresponding to the voltage exceeding limit nodes is mapped to the GIS platform. The heat map is rendered according to the evaluation results and the preset rendering rules. The specific rendering rules are as follows: when and When , it is rendered as a red block; when When , it is rendered as a yellow block; when and When , it is rendered as a green block; In the formula, is the voltage over-limit probability threshold; △V is the safety margin parameter; The reverse power transmission warning is generated according to the results of the adequacy index evaluation. Specifically, when η(t)>100%, the reverse power flow path is marked in the corresponding topological map of the GIS platform, and the warning information is pushed to the dispatching system; Based on the evaluation results of safety and economic indicators, the bearing capacity zones are generated and the regional bearing capacity levels are dynamically marked as follows: when or λ' ij' (t)>λ' limit , set to red level, λ' limit The line load rate exceeds the thermal stability limit of the equipment; When Cost∈[C low ,C high ] and η avg (t)>η base (t), set to yellow level, C low and C high are the lower limit and upper limit of network loss cost respectively; η avg (t) is the average photovoltaic penetration rate; η base (t) is the photovoltaic penetration benchmark value; when And λ' ij' (t)<λ' safe and Cost≤C opt When it is set to green level, λ' safe is the safety branch load rate, C opt is the optimal network loss threshold.

8. A data-driven distributed photovoltaic carrying capacity assessment method for distribution network according to claim 7, characterized in that: The method also includes updating photovoltaic output timing prediction data in real time through an API interface, and refreshing each area in the spatial distribution map within a preset time period.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for evaluating the distributed photovoltaic carrying capacity of a distribution network based on data-driven is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a data-driven distributed photovoltaic carrying capacity assessment method for a distribution network is implemented.

Images