Power distribution network power quality monitoring and comprehensive evaluation system and method

Abstract

The invention relates to the technical field of smart power grids, and discloses a power distribution network power quality monitoring and comprehensive evaluation system and method, and the method comprises the steps: collecting simulated power quality management data; processing according to the simulated electric energy quality management data to obtain simulated control parameters of the existing photovoltaic grid-connected inverter; generating simulation data; two-dimensional structure data is generated according to the simulation data, the two-dimensional structure data is input into a PVMFGCI strategy model, a strategy is output, and the strategy comprises the capacity of controllable source nodes needing to be additionally arranged by all dominant controlled nodes; executing the strategy to adjust the power grid; according to the method, the power grid treatment area can be intelligently divided, then the operation parameters of the photovoltaic grid-connected inverter are dynamically adjusted by applying the deep learning technology, and the power quality adjusting capacity is high.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and more specifically, to a distribution network power quality monitoring and comprehensive evaluation system and method. Background Art

[0002] With the widespread application of photovoltaic power generation systems in distribution networks, photovoltaic inverters have assumed more power quality management responsibilities. Traditional photovoltaic inverters mainly focus on achieving maximum power tracking and have limited ability to handle power quality issues. Summary of the invention

[0003] The present invention provides a distribution network power quality monitoring and comprehensive evaluation system and method to solve the technical problems in related technologies.

[0004] The present invention provides a method for monitoring and comprehensively evaluating power quality of a distribution network, comprising the following steps:

[0005] Step 100, collecting simulated power quality management data, the power quality management data including the operation data of the existing photovoltaic grid-connected inverter, the operation data of the entire power grid, and the operation data of each node of the power grid;

[0006] Step 200, obtaining simulated control parameters of an existing photovoltaic grid-connected inverter according to simulated power quality management data processing; the control parameters include harmonic management, reactive power compensation, and three-phase unbalance management coefficient of the photovoltaic grid-connected inverter;

[0007] Step 300, generating simulation data, wherein one simulation data includes simulated power quality management data collected at a fixed period within a set time period and corresponding simulated control parameters;

[0008] Step 400, generating two-dimensional structure data according to the simulation data, the two-dimensional structure data includes a data matrix and a relationship matrix, one unit of the data matrix represents one-dimensional structure data of an independent object, the independent object includes a controllable source node, a controlled node, and the entire power grid, and one unit of the data matrix only includes the simulation data of the independent object it represents;

[0009] The element in the i-th row and j-th column of the relationship matrix represents the association between the i-th unit of the data matrix and the independent object represented by the j-th unit. If there is an association, the value of this element of the relationship matrix is ​​1, otherwise it is 0;

[0010] A one-dimensional structured data includes n data items ordered by time, and the t-th data item represents the simulation data of the corresponding independent object collected at the t-th time;

[0011] Step 500, input the two-dimensional structure data into the PVMFGCI strategy model, the PVMFGCI strategy model includes a first intermediate layer, a second intermediate layer and a first output layer, the first intermediate layer inputs the one-dimensional structure data, outputs the first intermediate representation data to the second intermediate layer, the second intermediate layer also inputs the relationship matrix, outputs the second intermediate representation data to the first output layer, the first output layer outputs the strategy, the strategy includes the capacity of the controllable source nodes that need to be added to all the dominant controlled nodes;

[0012] Step 600, executing the strategy outputted in step 500 to adjust the power grid.

[0013] Furthermore, the strategy is represented as a vector, one component of which represents the per-unit value of the capacity of the controllable source node that needs to be added to a dominant controlled node.

[0014] Furthermore, the calculation formula of the first intermediate layer is as follows:

[0015] u ( t ) =σ(WuX ( t)+WuH( t -1 ) +bu)

[0016] r ( t ) =σ(WrX(t)+WrH( t -1)+br)

[0017] C(t)=tanh(WcX( t) +Wcr (t) ⊙ H( t - 1) +bc)

[0018] H(t ) =(1-u (t) )⊙ C( t) +u (t) ⊙ H( t - 1 )

[0019] Among them, W u , Wr, W c represents the weight parameter, bu, br, bc represent the bias parameter, ⊙ represents the dot product, u(t), r(t) and C(t) represent the first, second and third intermediate states respectively, where X(t) represents the tth data item of the one-dimensional structure data, H(t) and H(t- 1 ) represent the t-th and t-1-th intermediate representation data, n≥t≥1, n represents the total number of data items of the one-dimensional structure data, and when t=1, H(t- 1)=X(t), tanh is the hyperbolic tangent function, σ represents the sigmoid function;

[0020] The calculation formula for the second intermediate layer is as follows:

[0021]

[0022] Where kv represents the second intermediate representation data of the vth unit of the data matrix, M ( v ) is the set of cells associated with the vth cell of the data matrix, The nth first intermediate representation data output when the one-dimensional structure data of the uth unit of the data matrix is ​​input into the first intermediate layer, c v is a normalization constant, the default value is M ( v), σ is the sigmoid function, W k Represents the third weight parameter.

[0023] Furthermore, the calculation formula of the first output layer is as follows:

[0024]

[0025] Where y represents the first output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. ( al l ) represents the set of all cells of the data matrix, Wy represents the output weight parameter, and by represents the output bias parameter.

[0026] Further, the existence of an association between the controllable source node and the controlled node means that: the controllable source node is used to adjust the controlled node;

[0027] The existence of association between controllable source nodes means that: the controllable source nodes are located in the same comprehensive power quality management area;

[0028] The power grid as a whole is associated with all controllable source nodes and controlled nodes;

[0029] The existence of a correlation between the controlled nodes means that the controlled nodes are located in the same comprehensive power quality management area.

[0030] Furthermore, a deep learning model is used to obtain simulated control parameters of an existing photovoltaic grid-connected inverter, including the following steps:

[0031] The simulated power quality management data is processed into sequence data, where a sequence unit of the sequence data represents the simulated power quality management data collected once. The sequence data is input into a deep learning model, and the output represents the result of the simulated control parameters of the photovoltaic grid-connected inverter.

[0032] Furthermore, the calculation formula of the deep learning model is as follows:

[0033] h t =tanh(Whhh t -1+WxhX t +b h )

[0034] Where ht represents the t-th intermediate feature, ht-1 represents the t-1th first intermediate representation data, xt represents the t-th sequence unit of the sequence data, and Wh h and Wx h is the sum weight parameter, b h is the bias parameter, tanh is the hyperbolic tangent function;

[0035] OUT=σ(Wouthmax+bout)

[0036] Where OUT represents the first output vector, one of its component values ​​represents a control parameter of a simulation of a photovoltaic grid-connected inverter, hmax represents the last intermediate feature, Wou t is the out weight parameter, b out is the out bias parameter, and σ represents the sigmoid function.

[0037] Furthermore, the training of the deep learning model includes the following steps:

[0038] Step 101, initializing the parameters of the deep learning model;

[0039] Step 102: observe the simulated power quality management data St at time t, the strategy At executed at time t, and the simulated power quality management data S at time t+1. t+1 , the reward Rt obtained by executing the strategy At;

[0040] Step 103, then calculate the policy error:

[0041]

[0042] Where δt represents the policy error at time t, γ represents the discount coefficient, γ∈[0,1], Represents the deep learning model input St +1 The maximum probability value in the first output vector output when Represents the probability value corresponding to the strategy At in the first output vector output by the deep learning model when inputting St;

[0043] R t =w1·ΔTHD+W2·ΔVUF+W3·ΔQ-λ·nswitc h

[0044] Wherein, △THD is the reduction of total harmonic distortion rate from time t to time t+1;

[0045] ΔVUF is the reduction of the voltage unbalance rate from time t to time t+1;

[0046] △Q is the reduction in reactive power loss from time t to time t+1;

[0047] nswitc h The capacity ratio of the photovoltaic grid-connected inverter that changes its use from time t to time t+1;

[0048] W1, w2, and w3 are weight coefficients of the three power quality indicators, which are assigned different weights according to their importance. λ is the switching penalty coefficient, which controls the loss caused by the change of capacity usage of the photovoltaic grid-connected inverter;

[0049] Step 104, update the deep learning model, the updated formula is as follows:

[0050] V(St)←-V(St)+βδt

[0051] β∈[0,1], β represents the step size of deep learning, ↑ represents the transfer update;

[0052] Step 105, iterate steps 102-104 until the deep learning model converges or the number of iterations reaches a set value.

[0053] The present invention provides a distribution network power quality monitoring and comprehensive evaluation system, comprising:

[0054] The first module collects simulated power quality management data, which includes the operating data of the existing photovoltaic grid-connected inverter, the operating data of the entire power grid (load current, harmonic components, reactive components and unbalanced current components), and the operating data of each node in the power grid;

[0055] The second module obtains simulated control parameters of the existing photovoltaic grid-connected inverter according to the simulated power quality management data processing; the control parameters include harmonic management, reactive power compensation, and three-phase unbalance management coefficient of the photovoltaic grid-connected inverter;

[0056] The third module generates simulation data, where one simulation data includes simulated power quality management data collected at a fixed period within a set time period and corresponding simulated control parameters;

[0057] The fourth module generates two-dimensional structure data according to the simulation data. The two-dimensional structure data includes a data matrix and a relationship matrix. One unit of the data matrix represents one-dimensional structure data of an independent object. The independent object includes a controllable source node, a controlled node, and a power grid as a whole. One unit of the data matrix only contains simulation data of the independent object it represents.

[0058] The fifth module inputs the two-dimensional structure data into the PVMFGCI strategy model. The PVMFGCI strategy model includes a first intermediate layer, a second intermediate layer and a first output layer. The first intermediate layer inputs the one-dimensional structure data and outputs the first intermediate representation data to the second intermediate layer. The second intermediate layer also inputs the relationship matrix and outputs the second intermediate representation data to the first output layer. The first output layer outputs the strategy, which includes the capacity of the controllable source nodes that need to be added to all the dominant controlled nodes.

[0059] The present invention provides a computer storage medium for storing computer-readable instructions, which can execute the aforementioned distribution network power quality monitoring and comprehensive evaluation method when the computer-readable instructions are read.

[0060] The beneficial effects of the present invention are:

[0061] When a new photovoltaic access point is added to the power grid, the grid node layout will be changed, breaking the original grid operation balance. Generally, a period of continuous observation is required to determine whether to add information-controllable source nodes to adjust the grid parameters. Before the new photovoltaic access point is added to the power grid, the present invention uses a method of simulating the addition of a new photovoltaic access point to the power grid for simulated operation, and optimizes the existing photovoltaic grid-connected inverter, and then uses deep learning technology to provide a grid transformation strategy after actual grid connection. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is the structural block diagram of PVMFGCI of the present invention;

[0063] Figure 2 This is the distribution diagram of harmonic distortion rate before and after comprehensive treatment in different regions;

[0064] Figure 3 It is the three-phase imbalance distribution diagram before and after comprehensive governance in different regions;

[0065] Figure 4 It is a flow chart of the regional division strategy for comprehensive management of power quality in AC and DC distribution networks of the present invention;

[0066] Figure 5 It is a novel power distribution network topology structure of the present invention;

[0067] Figure 6 is the clustering validity function value of the present invention;

[0068] Figure 7 It is the regional division and dominant node diagram of the comprehensive power quality management of the present invention;

[0069] Figure 8 It is a flow chart of a method for monitoring and comprehensively evaluating power quality of a distribution network of the present invention. DETAILED DESCRIPTION

[0070] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0071] At least one embodiment of the present invention discloses a method for monitoring and comprehensively evaluating power quality of a distribution network. Figure 8 As shown, the following steps are included:

[0072] Step 100, collecting simulated power quality management data, the power quality management data includes the operating data of the existing photovoltaic grid-connected inverter, the operating data of the entire power grid (load current, harmonic components, reactive components and unbalanced current components), and the operating data of each node of the power grid;

[0073] Step 200, obtaining simulated control parameters of an existing photovoltaic grid-connected inverter according to simulated power quality management data processing; the control parameters include harmonic management, reactive power compensation, and three-phase unbalance management coefficient of the photovoltaic grid-connected inverter;

[0074] Step 300, generating simulation data, wherein one simulation data includes simulated power quality management data collected at a fixed period within a set time period and corresponding simulated control parameters;

[0075] Step 400, generating two-dimensional structure data according to the simulation data, the two-dimensional structure data includes a data matrix and a relationship matrix, one unit of the data matrix represents one-dimensional structure data of an independent object, the independent object includes a controllable source node, a controlled node, and the entire power grid, and one unit of the data matrix only includes the simulation data of the independent object it represents;

[0076] The element in the i-th row and j-th column of the relationship matrix represents the association between the i-th unit of the data matrix and the independent object represented by the j-th unit. If there is an association, the value of this element of the relationship matrix is ​​1, otherwise it is 0;

[0077] The association between the controllable source node and the controlled node means that: the controllable source node is used to adjust the controlled node;

[0078] The existence of association between controllable source nodes means that: the controllable source nodes are located in the same comprehensive power quality management area (see below);

[0079] The power grid as a whole is associated with all controllable source nodes and controlled nodes;

[0080] The relationship between the controlled nodes means that the controlled nodes are located in the same comprehensive power quality management area;

[0081] A one-dimensional structured data includes n data items ordered by time, and the t-th data item represents the simulation data of the corresponding independent object collected at the t-th time;

[0082] Step 500, input the two-dimensional structure data into the PVMFGCI strategy model. The PVMFGCI strategy model includes a first intermediate layer, a second intermediate layer and a first output layer. The first intermediate layer inputs the one-dimensional structure data and outputs the first intermediate representation data to the second intermediate layer. The second intermediate layer also inputs the relationship matrix and outputs the second intermediate representation data to the first output layer. The first output layer outputs the strategy, and the strategy includes the capacity of the controllable source nodes that need to be added to all the dominant controlled nodes.

[0083] In one embodiment of the present invention, the strategy is represented as a vector, one component of which represents the per-unit value of the capacity of the controllable source node that needs to be added to a dominant controlled node.

[0084] In one embodiment of the present invention, the calculation formula of the first intermediate layer is as follows:

[0085] u ( t)=σ(WuX ( t ) +WuH( t -1 ) +bu)

[0086] r ( t)=σ(WrX(t ) +WrH( t -1)+br)

[0087] C ( t)=tanh(WcX( t )+Wcr (t) ⊙ H( t -1)+bc)

[0088] H(t)=(1-u (t) )⊙ C(t)+u ( t ) ⊙ H(t - 1)

[0089] Among them, W u , Wr, W c represents the weight parameter, bu, br, bc represent the bias parameter, ⊙ represents the dot product, u(t), r(t) and C(t) represent the first, second and third intermediate states respectively, where X(t) represents the tth data item of the one-dimensional structure data, H(t) and H(t- 1 ) represent the t-th and t-1-th intermediate representation data, n≥t≥1, n represents the total number of data items of the one-dimensional structure data, t=1, H(t-1)=X(t), tanh is the hyperbolic tangent function, σ represents the sigmoid function;

[0090] The calculation formula for the second intermediate layer is as follows:

[0091]

[0092] Where kv represents the second intermediate representation data of the vth unit of the data matrix, M(v) is the set of units associated with the vth unit of the data matrix, The nth first intermediate representation data output when the one-dimensional structure data of the uth unit of the data matrix is ​​input into the first intermediate layer, c v is a normalization constant, the default value is M ( v), σ is the sigmoid function, W k Represents the third weight parameter.

[0093] The calculation formula of the first output layer is as follows:

[0094]

[0095] Where y represents the first output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. ( a 11 ) represents the set of all cells of the data matrix, W y represents the output weight parameter, and by represents the output bias parameter;

[0096] Step 600, executing the strategy outputted in step 500 to adjust the power grid.

[0097] It should be noted that the training data of the PVMFGCI strategy model comes from the simulation system. The simulation system simulates the rail transit information that needs to be scheduled and simulates the earthquake source to obtain the training data in the simulation environment.

[0098] The training objective of the PVMFGCI strategy model is to minimize the total harmonic distortion, voltage unbalance and reactive power loss.

[0099] In one embodiment of the present invention, the distribution network is divided into several comprehensive power quality management areas with high coupling between nodes. Due to the strong coupling of nodes in the area, there must be a dominant management node. When the power quality pollution of the node is managed, the power quality pollution of other nodes in the same area will also be greatly improved, thereby achieving a larger regional management effect with a smaller management capacity;

[0100] The present invention proposes a regional division strategy for comprehensive power quality management of AC and DC distribution networks. The process is as follows: Figure 4 As shown, the specific steps are as follows:

[0101] (1) A comprehensive voltage sensitivity model is established based on harmonic sensitivity, reactive power sensitivity, and three-phase unbalance sensitivity. The distributed photovoltaic management resources in the distribution network are considered and the photovoltaic control areas are divided to take into account the three main power quality indicators.

[0102] (2) The DDTW-FCM clustering partitioning algorithm is used to process time series data to obtain the number of clusters with the largest clustering validity function and the nodes contained in each cluster, thereby forming a power quality regional division, and the node closest to the cluster center in each region is defined as the regional governance dominant node.

[0103] (3) Based on the spatial relationship between the two types of areas, the following method is adopted to define the comprehensive power quality management area: if the PV control area is located inside the power quality area, the comprehensive management area is determined as the scope of the PV control area; if the PV control area overlaps with the power quality area, the PV control area and the power quality area corresponding to the dominant node it covers are integrated into the comprehensive power quality management area.

[0104] Such a definition method helps to effectively integrate the management resources of distributed photovoltaics, ensure the coordination of photovoltaic control to improve the quality of power, and facilitate the better implementation of comprehensive management measures.

[0105] In order to verify the rationality of the partition management method in this paper, Figure 5The new distribution network of is taken as the reference model, which is transformed from the IEEE14-node network. The new distribution network consists of a 16-node 220V AC subnet and a 5-node ±375V DC subnet. The VSC realizes the AC / DC interconnection network in the distribution network. The VSC capacity is set to 200kV·A. PV is connected to the distribution network station through AC three-phase and DC. The system capacitor and ideal transformer are not considered to absorb harmonics. The distribution system only uses the remaining capacity of the photovoltaic grid-connected inverter for comprehensive power quality management. The photovoltaic remaining capacity at the grid-connected node is set to 18kVA, and the solar cell array output is the maximum power matching with a maximum power of 15kW. After setting the full capacity of the photovoltaic grid-connected inverter for grid connection and harmonic control, there is still a certain margin left. This margin is neither used for active grid connection nor for power quality management. In the example solution, it is considered that the available capacity is 17kVA. Combined with the typical active output curve of photovoltaic power generation on a typical day, the designed power quality management optimization strategy is verified by example.

[0106] According to my country's harmonic standards, this chapter sets the upper limit of the harmonic voltage distortion rate THD. N The harmonic distribution of the system nodes used in the calculation example is shown in the following table, considering the harmonic current frequencies of 5, 7, 11, and 13. Nodes 3, 7, 13, and 15 are connected to 20kW unbalanced loads respectively. These settings help to consider the impact of harmonic currents in the analysis, especially when nodes 3, 7, 13, and 15 are connected to unbalanced loads, and effectively monitor and control the power quality.

[0107] Table 1 Node harmonic pollution data

[0108]

[0109] Based on the comprehensive voltage sensitivity model, the AC subgrid can be divided into photovoltaic control areas that take into account the three power quality indicators. The four photovoltaics have their own control areas respectively. The specific controllable source nodes and their controlled nodes are shown in Table 2.

[0110] Through this division, each photovoltaic device has a clear control area, allowing the system to be more targeted when considering multiple power quality indicators. This fine regional division helps to improve the effect of power quality management, allowing each photovoltaic device to more effectively adjust the power quality parameters within its control area.

[0111] Table 2 PV control areas and their nodes

[0112]

[0113] From the information in the table, we can see that the four controllable source nodes correspond to different photovoltaic control areas. Since their power quality compensation is in the form of current, this results in each control area being mainly located at the nodes on the same branch. Under the photovoltaic power quality management function, the power quality in the photovoltaic control area will be effectively improved. In essence, harmonic compensation, reactive power compensation and unbalance compensation for the system are all management of the current waveform. Through unified compensation, the system current becomes symmetrical and presents a sinusoidal waveform. This process helps to improve the power quality of the system and make it meet the harmonic standards and other related power quality requirements.

[0114] The power quality time series data of 16 nodes were reduced in dimension and sorted, and the DDTW-FCM method was used to partition the power quality. The point closest to the cluster center in each partition was set as the dominant node in the partition. The clustering effectiveness and partition results were shown in the following table and the following table. Through this analysis, the distribution of power quality data can be clearly seen, and the dominant nodes in each partition are also marked in the table. This study helps to better understand the spatiotemporal distribution of power quality characteristics and provides important guidance for the subsequent comprehensive management of power quality.

[0115] Through the analysis of the clustering validity function, it can be seen that the regional division effect is best when 4 regions are selected. At this time, the power quality coupling relationship between regions is the weakest, while the coupling relationship within the region is the strongest. According to the data in the table, it can be determined that the dominant control nodes of regions I, II, IV, and V are 3, 7, 15, and 11 respectively. Considering the distributed photovoltaic management resources, the power quality zoning can be further determined according to the regional division strategy for comprehensive power quality management of the new distribution network.

[0116] Table 3 Power quality area division results

[0117]

[0118] Based on the spatial relationship between the two types of areas, the power quality comprehensive management area is defined, and the AC subnet of the AC / DC distribution network can be divided into four power quality comprehensive management areas, as shown below. This definition method helps to effectively integrate the management resources of distributed photovoltaics, ensure that photovoltaic control coordinates to improve power quality, and is conducive to better implementation of comprehensive management measures.

[0119] In my country's low-voltage distribution network, the three-phase four-wire power supply mode is widely used. The four-bridge-arm photovoltaic grid-connected inverter with active filtering function mainly includes grid-connected power generation, active filtering, maximum power point tracking (MPPT), command current calculation and tracking control and other technologies. The structure block diagram of the photovoltaic multi-function grid-connected inverter (PVMFGCI) is shown below.

[0120] The main circuit of the system consists of a photovoltaic array, a three-phase four-arm converter, a filter inductor, a parallel load, a three-phase four-wire power grid, etc. The control loop consists of a signal detection unit, a maximum power point tracking (MPPT) control unit, a command current calculation unit, and a main control unit. This structure ensures the stable operation of the photovoltaic grid-connected inverter in the low-voltage distribution network, and at the same time has multiple key functions such as active filtering, which provides a strong guarantee for the reliability and performance of the system.

[0121] The control loop of the photovoltaic grid-connected inverter monitors the load current of the distribution network in real time and extracts harmonic, reactive and unbalanced current components. By generating a compensation current reference value and combining it with the active current value of photovoltaic power generation, the final grid-connected command current is formed to ensure the unified regulation of grid-connected power generation and active filtering functions. This study focuses on the combination of photovoltaic grid-connected and parallel APF, aiming to better compensate for power problems such as imbalance in the grid while delivering active power to the grid. Therefore, MPPT is not described in detail. Through this integrated control strategy, the system can more effectively coordinate photovoltaic power generation and active filtering functions, improve power quality, and provide reliable support for the stable operation of the power grid.

[0122] The total capacity of the photovoltaic grid-connected inverter is S N , the capacity of the grid-connected inverter occupied by DG is S DG , the remaining capacity of the grid-connected inverter is S RE .

[0123]

[0124] Where U is the distribution network voltage, and I1 is the inverter output active current.

[0125] The relationship between the harmonic control and reactive power compensation capacity of the photovoltaic grid-connected inverter and its remaining capacity can be expressed as:

[0126]

[0127] Where: S h , S q , S2 are the harmonic control, reactive power compensation and three-phase unbalance control capacity of the DG grid-connected inverter respectively

[0128] By rationally configuring the location and capacity of distributed generation (DG) grid-connected inverters in the distribution network, the allocation of the remaining capacity of the inverter can be optimized, thereby achieving differentiated management of power quality in different regions. By optimizing the compensation strategy, the management effect of PV grid-connected inverters in terms of harmonics, reactive power, and three-phase imbalance can be made more economical and reasonable, achieving effective allocation of management resources

[50] -

[51] .

[0129] In order to establish a mathematical model for PVMFGCI capacity allocation and to find the optimal compensation coefficient of its remaining capacity for harmonics, reactive power and three-phase imbalance, we introduce the following mathematical model to represent the coefficients of these three aspects:

[0130]

[0131]

[0132] In the formula, I1, I h ,I q , I0 are the grid-side fundamental current, harmonic current, reactive current and three-phase unbalanced current respectively.

[0133] Assume that the harmonic control, reactive power compensation, and three-phase unbalance control coefficients of the remaining capacity of PVMFGCI are ε h , ε q , ε2, the power quality coefficient after compensation is

[0134]

[0135] In the formula, l h0 , l q0 , l 20 and l h , l q , l2 are the harmonic, reactive power and three-phase unbalance coefficient before and after compensation respectively.

[0136] Solve PVMFGCI to establish a comprehensive power quality evaluation model under multiple operating scenarios

[0137]

[0138] In the formula, I p1 ,I ph ,I pq ,I p2 They are PVMFGCI output active current, harmonic control, reactive power compensation and three-phase unbalanced current control.

[0139] In view of the simplicity and feasibility of the Lagrange multiplier method, this paper adopts this method to solve the optimal coefficient problem of comprehensive power quality management of PVMFGCI remaining capacity.

[0140] Let λ be the Lagrange multiplier, and construct the Lagrange function as:

[0141]

[0142] By using the Lagrange multiplier method, we can solve:

[0143]

[0144] The comprehensive power quality management strategy for the remaining capacity of PVMFGCI is as follows:

[0145] (1) Calculate the remaining capacity S of the grid-connected inverter RE , set the remaining capacity threshold Q N 0.

[0146] (2) If S RE ≥≥QN0, and satisfies

[0147]

[0148] Then full compensation is performed directly, that is, εh=εq=ε0=1; otherwise, the compensation coefficients are calculated according to formula (10) before compensation is performed.

[0149] (3) If S RE ≤Q N0 , then no compensation is required, that is, εh=εq=ε0=0.

[0150] In some other embodiments of the present invention, a deep learning model is used to obtain simulated control parameters of an existing photovoltaic grid-connected inverter, including the following steps:

[0151] The simulated power quality management data is processed into sequence data, where a sequence unit of the sequence data represents the simulated power quality management data collected once. The sequence data is input into a deep learning model, and the output represents the result of the simulated control parameters of the photovoltaic grid-connected inverter.

[0152] The calculation formula of the deep learning model is as follows:

[0153] h t =tanh(W hh h t -1+WxhX t +b h )

[0154] Where ht represents the t-th intermediate feature, ht-1 represents the t-1th first intermediate representation data, xt represents the t-th sequence unit of the sequence data, and W hh and Wx h is the sum weight parameter, b his the bias parameter, tanh is the hyperbolic tangent function;

[0155] OUT=σ(Wouthma x +bout)

[0156] Where OUT represents the first output vector, a component value of which represents a control parameter of a simulation of a photovoltaic grid-connected inverter, hmax represents the last intermediate characteristic, Wo u t is the out weight parameter, b out is the out bias parameter, σ represents the sigmoid function;

[0157] The training of a deep learning model consists of the following steps:

[0158] Step 101, initializing the parameters of the deep learning model;

[0159] Step 102: observe the simulated power quality management data St at time t, the strategy At executed at time t, and the simulated power quality management data St at time t+1. +1 , the reward Rt obtained by executing the strategy At;

[0160] Step 103, then calculate the policy error:

[0161]

[0162] Where δt represents the policy error at time t, γ represents the discount coefficient, γ∈[0,1], Represents the deep learning model input S t+ The maximum probability value in the first output vector when 1 is output, V (S t) represents the probability value corresponding to the strategy At in the first output vector output when the deep learning model inputs St;

[0163] R t =W1·ΔTHD+W2·ΔVUF+W3·ΔQ-λ·nswitc h

[0164] Wherein, ΔTHD is the reduction of total harmonic distortion from time t to time t+1;

[0165] ΔVUF is the reduction of the voltage unbalance rate from time t to time t+1;

[0166] △Q is the reduction in reactive power loss from time t to time t+1;

[0167] nswitc h The capacity ratio of the photovoltaic grid-connected inverter that changes its use from time t to time t+1;

[0168] w1, w2, and w3 are weight coefficients of the three power quality indicators, which are assigned different weights according to their importance. λ is the switching penalty coefficient, which controls the loss caused by the change of capacity usage of the photovoltaic grid-connected inverter;

[0169] Step 104, update the deep learning model, the updated formula is as follows:

[0170] V( S t)←-V( S t)+βδt

[0171] β∈[0,1], β represents the step size of deep learning, ↑ represents the transfer update;

[0172] Step 105, iterate steps 102-104 until the deep learning model converges or the number of iterations reaches a set value.

[0173] Simulation verification of governance optimization results:

[0174] In order to solve the optimal compensation coefficient of harmonics, reactive power and three-phase imbalance for the power quality optimization strategy of the adjustable capacity of the photovoltaic grid-connected system, the corresponding mathematical modeling was carried out, and the capacity allocation and compensation coefficient were obtained, as shown in the following table.

[0175] Table 4 Treatment optimization results and compensation coefficients

[0176]

[0177] According to the data in Table 1, this project selected the maximum output power of the photovoltaic system at 12:00 as 15kW. Through calculation, it is known that the power quality compensation capacity does not exceed the limit of the remaining capacity. This shows that the remaining capacity of the distributed photovoltaic system fully meets the needs of power quality management of the distribution network, so that the active power grid connection of the photovoltaic system can be fully realized without reduction. Before and after the comprehensive power quality management, the changes in the harmonics and three-phase imbalance indicators of each node in the AC subnet of the AC / DC distribution network are shown in the figure. Figure 2-3 shown.

[0178] After comprehensive power quality management, the harmonic level of each node in the AC subnet was less than 2%, and the three-phase imbalance was significantly reduced. In addition, the ripple factor of the DC bus dropped from 1.75% before management to 0.5% after management. This shows that by implementing power quality management measures in the AC subnet, the harmonic ripple introduced by the AC subnet was successfully reduced, the power quality of the DC subnet was further improved, and its stable operation was ensured.

[0179] At least one embodiment of the present invention provides a distribution network power quality monitoring and comprehensive evaluation system, which is used to execute the above-mentioned distribution network power quality monitoring and comprehensive evaluation method, including:

[0180] The first module collects simulated power quality management data, which includes the operating data of the existing photovoltaic grid-connected inverter, the operating data of the entire power grid (load current, harmonic components, reactive components and unbalanced current components), and the operating data of each node in the power grid;

[0181] The second module obtains simulated control parameters of the existing photovoltaic grid-connected inverter according to the simulated power quality management data processing; the control parameters include harmonic management, reactive power compensation, and three-phase unbalance management coefficient of the photovoltaic grid-connected inverter;

[0182] The third module generates simulation data, where one simulation data includes simulated power quality management data collected at a fixed period within a set time period and corresponding simulated control parameters;

[0183] The fourth module generates two-dimensional structure data according to the simulation data. The two-dimensional structure data includes a data matrix and a relationship matrix. One unit of the data matrix represents one-dimensional structure data of an independent object. The independent object includes a controllable source node, a controlled node, and a power grid as a whole. One unit of the data matrix only contains simulation data of the independent object it represents.

[0184] The fifth module inputs the two-dimensional structure data into the PVMFGCI strategy model. The PVMFGCI strategy model includes a first intermediate layer, a second intermediate layer and a first output layer. The first intermediate layer inputs the one-dimensional structure data and outputs the first intermediate representation data to the second intermediate layer. The second intermediate layer also inputs the relationship matrix and outputs the second intermediate representation data to the first output layer. The first output layer outputs the strategy, and the strategy includes the capacity of the controllable source nodes that need to be added to all the dominant controlled nodes.

[0185] Step 600, executing the strategy outputted in step 500 to adjust the power grid.

[0186] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for monitoring and comprehensively evaluating power quality in a distribution network, characterized in that: The following steps are involved: Step 100, collecting simulated power quality management data, the power quality management data including the operation data of the existing photovoltaic grid-connected inverter, the operation data of the entire power grid, and the operation data of each node of the power grid; Step 200, obtaining simulated control parameters of an existing photovoltaic grid-connected inverter according to simulated power quality management data processing; the control parameters include harmonic management, reactive power compensation, and three-phase unbalance management coefficient of the photovoltaic grid-connected inverter; Step 300, generating simulation data, wherein one simulation data includes simulated power quality management data collected at a fixed period within a set time period and corresponding simulated control parameters; Step 400, generating two-dimensional structure data according to the simulation data, the two-dimensional structure data includes a data matrix and a relationship matrix, one unit of the data matrix represents one-dimensional structure data of an independent object, the independent object includes a controllable source node, a controlled node, and the entire power grid, and one unit of the data matrix only includes the simulation data of the independent object it represents; The element in the i-th row and j-th column of the relationship matrix represents the association between the i-th unit of the data matrix and the independent object represented by the j-th unit. If there is an association, the value of this element of the relationship matrix is ​​1, otherwise it is 0; A one-dimensional structured data includes n data items ordered by time, and the t-th data item represents the simulation data of the corresponding independent object collected at the t-th time; Step 500, input the two-dimensional structure data into the PVMFGCI strategy model, the PVMFGCI strategy model includes a first intermediate layer, a second intermediate layer and a first output layer, the first intermediate layer inputs the one-dimensional structure data, outputs the first intermediate representation data to the second intermediate layer, the second intermediate layer also inputs the relationship matrix, outputs the second intermediate representation data to the first output layer, the first output layer outputs the strategy, the strategy includes the capacity of the controllable source nodes that need to be added to all the dominant controlled nodes; Step 600, executing the strategy outputted in step 500 to adjust the power grid.

2. A method for monitoring and comprehensively evaluating power quality in a distribution network according to claim 1, characterized in that: The strategy is represented as a vector, one component of which represents the per-unit value of the capacity of the controllable source node that needs to be added to a dominant controlled node.

3. A method for monitoring and comprehensively evaluating power quality in a distribution network according to claim 1, characterized in that: The calculation formula for the first intermediate layer is as follows: u (t) =σ(W u X (t) +W u H (t-1) +b u ) r (t) =σ(W r X (t) +W r H (t-1) +b r ) C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c ) H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1) Among them, W u , W r , W c represents the weight parameter, b u , b r , b c represents the bias parameter, ⊙ represents the dot product, u (t) 、r (t) and C (t) Represent the first, second, and third intermediate states respectively, where X (t) Represents the tth data item of one-dimensional structured data, H (t) and H (t-1) They represent the t-th and t-1-th intermediate representation data respectively, n≥t≥1, n represents the total number of data items of the one-dimensional structure data, and when t=1, H (t-1) =X (t) , tanh is the hyperbolic tangent function, σ represents the sigmoid function; The calculation formula for the second intermediate layer is as follows: where k v The second intermediate representation data of the vth cell of the data matrix, M (v) is the set of cells associated with the vth cell of the data matrix, The nth first intermediate representation data output when the one-dimensional structure data of the uth unit of the data matrix is ​​input into the first intermediate layer, c v is a normalization constant, the default value is M (v) The total number of units in , σ is the sigmoid function, W k Represents the third weight parameter.

4. A method for monitoring and comprehensively evaluating power quality of a distribution network according to claim 3, characterized in that: The calculation formula of the first output layer is as follows: Where y represents the first output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. (all) represents the set of all cells of the data matrix, W y represents the output weight parameter, b y Represents the output bias parameter.

5. A method for monitoring and comprehensively evaluating power quality in a distribution network according to claim 1, characterized in that: The association between the controllable source node and the controlled node means that: the controllable source node is used to adjust the controlled node; The existence of association between controllable source nodes means that: the controllable source nodes are located in the same comprehensive power quality management area; The power grid as a whole is associated with all controllable source nodes and controlled nodes; The existence of a correlation between the controlled nodes means that the controlled nodes are located in the same comprehensive power quality management area.

6. A method for monitoring and comprehensively evaluating power quality in a distribution network according to claim 1, characterized in that: A deep learning model is used to obtain the simulated control parameters of an existing photovoltaic grid-connected inverter, including the following steps: The simulated power quality management data is processed into sequence data, where a sequence unit of the sequence data represents the simulated power quality management data collected once. The sequence data is input into a deep learning model, and the output represents the result of the simulated control parameters of the photovoltaic grid-connected inverter.

7. A method for monitoring and comprehensively evaluating power quality in a distribution network according to claim 6, characterized in that: The calculation formula of the deep learning model is as follows: h t =tanh(W hh h t-1 +W xh x t +b h ) where h t represents the tth intermediate feature, h t-1 represents the t-1th first intermediate representation data, x t represents the tth sequence unit of the sequence data, W hh and W xh is the sum weight parameter, b h is the bias parameter, tanh is the hyperbolic tangent function; OUT=σ(W out h max +b out ) Where OUT represents the first output vector, a component value of which represents a control parameter of a simulation of a photovoltaic grid-connected inverter, h max represents the last intermediate feature, W out is the out weight parameter, b out is the out bias parameter, and σ represents the sigmoid function.

8. A method for monitoring and comprehensively evaluating power quality in a distribution network according to claim 7, characterized in that: The training of a deep learning model consists of the following steps: Step 101, initializing the parameters of the deep learning model; Step 102: Observe the simulated power quality management data S at time t t , strategy A executed at time t t , the simulated power quality management data S at time t+1 t+1 , execute strategy A t Rewards received R t ; Step 103, then calculate the policy error: where δ t represents the policy error at time t, γ represents the discount coefficient, γ∈[0,1], Represents the deep learning model input S t+1 The maximum probability value in the first output vector output when Represents the deep learning model input S t The first output vector outputted when t is t corresponds to strategy A t The probability value of R t =w1·ΔTHD+w2·ΔVUF+w3·ΔQ-λ·n switch Wherein, ΔTHD is the reduction of total harmonic distortion from time t to time t+1; ΔVUF is the reduction of the voltage unbalance rate from time t to time t+1; ΔQ is the reduction in reactive power loss from time t to time t+1; n switch The capacity ratio of the photovoltaic grid-connected inverter that changes its use from time t to time t+1; w1, w2, and w3 are weight coefficients of the three power quality indicators, which are assigned different weights according to their importance. λ is the switching penalty coefficient, which controls the loss caused by the change of capacity usage of the photovoltaic grid-connected inverter; Step 104, update the deep learning model, the updated formula is as follows: β∈[0,1], β represents the step size of deep learning, ← represents the transfer update; Step 105, iterate steps 102-104 until the deep learning model converges or the number of iterations reaches a set value.

9. A distribution network power quality monitoring and comprehensive evaluation system, characterized in that: It is used to execute a distribution network power quality monitoring and comprehensive evaluation method as described in any one of claims 1 to 8, comprising: The first module collects simulated power quality management data, which includes the operating data of the existing photovoltaic grid-connected inverter, the operating data of the entire power grid (load current, harmonic components, reactive components and unbalanced current components), and the operating data of each node in the power grid; The second module obtains simulated control parameters of the existing photovoltaic grid-connected inverter according to the simulated power quality management data processing; the control parameters include harmonic management, reactive power compensation, and three-phase unbalance management coefficient of the photovoltaic grid-connected inverter; The third module generates simulation data, where one simulation data includes simulated power quality management data collected at a fixed period within a set time period and corresponding simulated control parameters; The fourth module generates two-dimensional structure data according to the simulation data. The two-dimensional structure data includes a data matrix and a relationship matrix. One unit of the data matrix represents one-dimensional structure data of an independent object. The independent object includes a controllable source node, a controlled node, and a power grid as a whole. One unit of the data matrix only contains simulation data of the independent object it represents. The fifth module inputs the two-dimensional structure data into the PVMFGCI strategy model. The PVMFGCI strategy model includes a first intermediate layer, a second intermediate layer and a first output layer. The first intermediate layer inputs the one-dimensional structure data and outputs the first intermediate representation data to the second intermediate layer. The second intermediate layer also inputs the relationship matrix and outputs the second intermediate representation data to the first output layer. The first output layer outputs the strategy, which includes the capacity of the controllable source nodes that need to be added to all the dominant controlled nodes.

10. A computer storage medium, characterized in that: It is used to store computer-readable instructions, and when the computer-readable instructions are read, a distribution network power quality monitoring and comprehensive evaluation method as described in any one of claims 1-8 can be executed.

Images