Voltage optimization scheduling method for distribution and micro-distribution cooperation based on multiple sensitivity perception

Abstract

This invention discloses a distribution-microgrid collaborative voltage optimization scheduling method based on multi-sensitivity sensing. First, a multi-sensitivity real-time sensing model is established. When real-time measurements are incomplete, the offline-trained multi-sensitivity sensing model is used to sense real-time sensitivity online. Next, a model predictive control method is employed. Based on the multi-sensitivity model, prediction models for line loss, voltage, and real-time operating costs are established. These models jointly optimize the real-time measurable line losses and equipment real-time operating costs. Furthermore, the multi-sensitivity sensing model is used to correct line loss and voltage sensitivity in real-time, reducing control errors caused by power fluctuations. Finally, when the voltage of a real-time measurable node in the microgrid exceeds its limit, voltage autonomy is achieved based on the real-time fitted microgrid voltage sensitivity, realizing real-time, economical, and safe operation of the entire distribution-microgrid system.

Description

Technical Field

[0001] This invention relates to the field of coordinated optimization scheduling technology for distribution networks and microgrids, specifically a method for coordinated voltage optimization scheduling of distribution networks and microgrids based on multiple sensitivity sensing. Background Technology

[0002] The integration of numerous distributed resources within distribution networks and microgrids, coupled with the randomness and volatility of their real-time output, exacerbates the difficulty of matching them with load power. This leads to uneconomical operation and voltage fluctuations in the real-time operation of distribution networks and microgrids. Furthermore, during real-time optimization, the low level of intelligence and slow communication speeds of some nodes within my country's distribution networks and microgrids affect the completeness of real-time measurement data acquisition. This renders traditional calculation methods based on power flow physical models inapplicable. Therefore, there is an urgent need to research real-time collaborative optimization scheduling strategies for distribution networks and microgrids that consider the situation of missing measurements.

[0003] In related technologies, with the large-scale integration of distributed resources, the uncertainties in power system dispatching have increased significantly, posing a major challenge to the economic efficiency and security of power system operation. Therefore, power system optimal dispatching considering uncertainties has received considerable research and attention. Currently, power system optimal dispatching methods considering the uncertainties of distributed resources can be mainly divided into robust optimal dispatching methods and stochastic optimal dispatching methods.

[0004] Robust optimization scheduling methods describe random fluctuations using uncertain sets, thereby generating scheduling plans that meet the feasibility requirements of the worst-case scenario. Existing robust optimization scheduling methods can be divided into two-stage robust optimization methods and multi-stage robust optimization methods. Two-stage robust optimization methods include solving for a pre-scheduling plan for the predicted scenario and solving for a rescheduling plan for the worst-case scenario; multi-stage robust optimization problems couple multiple two-stage robust optimization problems in a nested manner.

[0005] Stochastic optimization scheduling is also one of the main methods for solving the optimization scheduling problem of uncertain power systems. As an important method for solving stochastic optimization problems, the scenario analysis-based stochastic optimization scheduling method transforms the stochastic optimization problem into a deterministic problem by generating and reducing scenarios. Applying this method to the optimal scheduling of master and distribution stations with a high proportion of distributed resource access can enhance the adaptability of the scheduling plan.

[0006] The shortcomings of existing technical solutions are: (1) Although the robust optimization scheduling algorithm can ensure the optimal scheduling under the worst scenario, it is not economical enough for other non-worst scenarios and is not the optimal solution. Therefore, the robust optimization scheduling has a large limitation in the main distribution system under high proportion of distributed resource access; (2) The stochastic optimization scheduling method is currently mainly applied to single voltage level power systems and is rarely used in the optimization scheduling of multi-level power systems. It is urgent to explore a distribution-micro stochastic optimization scheduling strategy suitable for distribution networks and microgrids with high proportion of distributed resource access, so as to ensure the safe and economical operation of distribution and microgrids under the increasing uncertainty. Summary of the Invention

[0007] The purpose of this invention is to provide a micro-cooperative voltage optimization scheduling method based on multiple sensitivity sensing to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a micro-cooperative voltage optimization scheduling method based on multiple sensitivity sensing, characterized by comprising the following steps:

[0009] Step 1: Establish a multi-sensitivity coefficient matrix, including voltage sensitivity, line loss sensitivity, and cost sensitivity; in the distribution network and microgrid system, write each regulating device node into the augmented matrix of the power flow equation, and simplify to obtain the power flow correction equation expressed in polar coordinates of node voltage:

[0010]

[0011] In the formula, ΔV is the Jacobian matrix; ΔV and Δθ are the voltage magnitude and phase angle correction of the node, respectively; ΔP and ΔQ are the changes in active and reactive power injection at each node.

[0012] Inverting the Jacobian matrix yields the relationship between ΔV, Δθ and ΔP, ΔQ. When the Newton-Raphson method converges, S... 21 S 22 These are the active and reactive voltage sensitivities at the nodes, respectively.

[0013]

[0014] In the formula, The block submatrices H, N, K, and L in the Jacobian matrix are respectively equal to

[0015] Step two involves establishing a multi-sensitivity real-time sensing model, including sensitivity fitting principle analysis, radial basis function neural network and its sensing architecture, and optimization of the RBF neural network using a social network search algorithm. An implicit functional relationship exists between voltage sensitivity and the voltage of each node and the injected power.

[0016]

[0017] In the formula, f U-P f U-Q These represent the functional relationships between active and reactive voltage sensitivity and voltage and injected power, respectively.

[0018] Step 3: Real-time optimization and dispatching strategy for distribution networks based on model predictive control, including prediction models and feedback correction models; line loss, voltage, and cost prediction models are respectively:

[0019]

[0020] V(k+i|k)=V(k+i-1|k)+S U-P ΔP re +S U-Q ΔQ re ;

[0021] C re (k+i|k)=S c-P ΔP re +S c-PR ΔP rc +S c-QR ΔQ rc ;

[0022] In the formula, i represents the prediction step number, i = 1, 2, 3, ..., N p -1,N p V(k+i|k) represents the predicted voltage amplitude at time k+i from time k; P lossij (k+i|k) represents the line loss predicted at time k+i from time k; C re (k+i|k) represents the value of the operating cost function at time k, predicting the future operating cost at time k+i; ΔP re ΔQ re Let ΔP be the variable in the adjustable device to be optimized for control at time k+i-1. re =[ΔP PV,w,re ΔP EV,r,re ], ΔQ re =[ΔQ PV,w,re ΔQ SVC,n,re That is, the active power control variable of the regulating equipment includes the photovoltaic active power reduction ΔP. PV,w,re Active power regulation ΔP of electric vehicles EV,r,re The reactive power control variables of the regulating equipment include the photovoltaic reactive power regulation amount ΔQ. PV,w,re SVC reactive power regulation ΔQ SVC,n,re ;ΔP rc ΔQ rcThis is the corresponding real-time power correction amount for adjustable equipment, and its value is obtained by subtracting the power of the adjustable equipment at time k+i from the power of that period in the daily scheduling plan;

[0023] Step four: Based on the sensitivity-based real-time autonomous model of the microgrid, the objective function is evaluated using the power deviation penalty cost:

[0024]

[0025] In the formula, P tai,a,re Q tai,a,re These represent the real-time active and reactive power outputs of device a in the microgrid, respectively; P tai,a,in Q tai,a,in Reference values ​​for active and reactive power dispatch instructions within the day; S c-pra S c-qra These represent the active power and reactive power deviation penalty cost sensitivity of device a in the microgrid, respectively.

[0026] Step 5, the real-time collaborative optimization and scheduling process for distribution networks and microgrids, includes real-time optimization and scheduling of distribution networks based on model predictive control and real-time autonomous optimization of microgrids based on voltage sensitivity.

[0027] Furthermore, in step one, the active power loss of the line loss sensitivity is:

[0028] P lossij =(V i 2 +V j 2 -2V i V j cosθ ij )G ij ;

[0029] In the formula, P lossij Let be the line loss between node i and node j.

[0030] Furthermore, in step one, the cost sensitivity S c This includes revising the intraday scheduling plan based on real-time power demand and the actual power of distributed resources, as well as calculating the cost sensitivity of each adjustable device, including the adjustment cost of each device and the power adjustment penalty cost of each device, in order to optimize the objective function in real time.

[0031] Furthermore, in step one, based on voltage sensitivity, line loss sensitivity, and cost sensitivity, and combined with a multi-sensitivity real-time sensing model, a relationship is established between the real-time measurable line loss change, the real-time measurable node voltage change, the real-time operating cost of adjustable equipment, and the power change and correction of adjustable equipment, even when real-time measurements are incomplete.

[0032] Furthermore, in step two, the voltage sensitivity in the sensitivity fitting principle analysis is:

[0033]

[0034] In the formula, f U-P f U-Q These represent the functional relationships between active and reactive voltage sensitivity and voltage and injected power, respectively.

[0035] Line loss sensitivity can also be expressed as an implicit function relationship:

[0036]

[0037] Where, These represent the functional relationships between active / reactive line loss sensitivity and voltage and injected power, respectively.

[0038] Furthermore, in step three, based on the model predictive control method and combined with the multi-sensitivity sensing model, the line loss and real-time operating cost of the real-time measurable lines in the distribution network are calculated in a rolling manner based on the line loss sensitivity and cost sensitivity. During the calculation process, the real-time measurement nodes are checked for safety based on the voltage sensitivity, and the prediction model is corrected by feedback through the multi-sensitivity real-time sensing model and real-time electricity price.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) This invention provides a power distribution and micro-voltage optimization scheduling strategy under weak measurement conditions. A real-time sensing model with multiple sensitivities based on RBF neural network is established. The SNS algorithm is used to optimize the RBF neural network, so that accurate multiple sensitivities can still be obtained under the condition of incomplete measurement, thereby simplifying the real-time optimization model and getting rid of the limitations of the power flow physical model.

[0041] (2) A rolling optimization model was established with the goal of minimizing the real-time measurement of line loss and the real-time operating cost of adjustable equipment, and constrained by the real-time measurement of node voltage safety. The sensitivity was updated in real time through a multi-sensitivity sensing model to reduce control error and ensure the safe and economical operation of the distribution network.

[0042] (3) A real-time autonomous microgrid model with the goal of minimizing power correction was established. When the voltage of a measurable node in the microgrid exceeds its limit, voltage autonomy of the microgrid is achieved based on voltage sensitivity. Through spatiotemporal dual-scale coordination between distribution and microgrids, real-time optimized scheduling of distribution and microgrids that balances economy and safety is realized. Through the above strategies, rapid voltage regulation of distribution and microgrids can be achieved under weak measurement conditions. Attached Figure Description

[0043] Figure 1This is a diagram of the RBF neural network fitting architecture of the present invention;

[0044] Figure 2 This is a diagram illustrating the real-time optimization process of the power distribution network according to the present invention.

[0045] Figure 3 This is a schematic diagram of the model predictive control structure of the present invention;

[0046] Figure 4 This is a flowchart illustrating the real-time collaborative optimization solution process of the present invention. Detailed Implementation

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] Example:

[0049] Please see Figure 1-4 The present invention provides a technical solution: a micro-cooperative voltage optimization scheduling method based on multiple sensitivity sensing;

[0050] This invention is achieved using the following technical solution:

[0051] First, a multiple sensitivity coefficient matrix is ​​derived to simplify the traditional power flow model.

[0052] Secondly, a multi-sensitivity real-time sensing model was established. In the case of incomplete real-time measurements, the real-time sensitivity was perceived online based on the offline-trained multi-sensitivity sensing model. Furthermore, a model predictive control method was adopted, and a prediction model for line loss, voltage, and real-time operating cost was established based on the multi-sensitivity model. Under the constraint of the reference value of the gate node voltage calculated by the micro-daily collaborative scheduling, the distributed photovoltaic, electric vehicle charging station, static var compensator and other objects were further controlled. The real-time measurable line loss and real-time operating cost of the equipment were jointly optimized, and the line loss and voltage sensitivity were corrected in real time based on the multi-sensitivity sensing model to reduce the control error caused by power fluctuations.

[0053] Finally, in the case of real-time measurement of node voltage exceeding limits within the microgrid, voltage autonomy is achieved based on the real-time fitted distribution area voltage sensitivity, realizing real-time economical and safe operation of the entire distribution microgrid system.

[0054] Multiple sensitivity coefficients

[0055] To establish real-time optimization models for distribution networks and microgrids and overcome the limitation that power flow physical models cannot be established when measurements are lacking, this section derives a multi-sensitivity coefficient matrix that includes voltage sensitivity, line loss sensitivity, and cost sensitivity. This simplifies the real-time optimization models for distribution networks and microgrids and further improves optimization efficiency.

[0056] Voltage sensitivity

[0057] To describe the voltage regulation effect of different regulation resources in distribution networks and microgrids, the voltage sensitivity matrix is ​​solved based on the operating parameters of distribution networks and microgrids.

[0058] In distribution networks and microgrid systems, each regulating device node is written into the augmented matrix of the power flow equations, and simplified to obtain the power flow correction equations expressed in polar coordinates of node voltages:

[0059]

[0060] In the formula, ΔV is the Jacobian matrix; ΔV and Δθ are the voltage magnitude and phase angle correction of the node, respectively; ΔP and ΔQ are the changes in active and reactive power injected into each node.

[0061] Inverting the Jacobian matrix yields the relationship between ΔV, Δθ and ΔP, ΔQ. When the Newton-Raphson method converges, S... 21 S 22 These are the active and reactive voltage sensitivities at the nodes, respectively.

[0062]

[0063] In the formula, The block submatrices H, N, K, and L in the Jacobian matrix are respectively equal to

[0064] In the two cases of i≠j and i=j and The element expression can be expressed as:

[0065]

[0066] In θ ij When the value is very small, cosθ can be approximated. ij =1, sinθ ij Since the value is 0, equation (1-3)-(1-4) can be simplified to:

[0067]

[0068] Line loss sensitivity

[0069] To describe the regulatory effect of adjustable resources on network losses within a distribution network, a network loss sensitivity matrix can be derived based on distribution network operating parameters. However, in real-time optimization, due to the lack of real-time measurements of the distribution network, it is difficult to obtain the real-time total network loss, resulting in the lack of initial values ​​for optimization methods based on network loss sensitivity, making them difficult to apply.

[0070] However, on some lines with high measurement configuration levels, the initial value of real-time line loss can be calculated using equation (1-9), thereby achieving line loss optimization for lines that can be measured in real time. Therefore, this embodiment proposes a line loss sensitivity derivation method, referring to the network loss sensitivity derivation process, to obtain the relationship between the change in loss of each line in the system and the adjustment amount of each regulating device, laying the foundation for real-time optimization of line losses that can be measured in real time.

[0071] First, the active power loss of line ij can be expressed as:

[0072] P lossij =(V i 2 +V j 2 -2V i V j cosθ ij )G ij (1-9)

[0073] In the formula, P lossij Let be the line loss between node i and node j.

[0074] Rewriting the above equation as total differentials for active and reactive power at each node, we get:

[0075]

[0076] Where, Let be the partial derivatives of the voltage phase angle / amplitude at node i with respect to the active / reactive power at each node, respectively. The solution method is as follows:

[0077] From equation (1-2), we can obtain that

[0078] At the same time, in equations (1-10)-(1-11) It can be represented as:

[0079]

[0080] In θ ij When the value is very small, cosθ can be approximated. ij =1, sinθ ij Since the value is 0, equation (1-9) can be simplified to:

[0081] Plossij =(V i 2 +V j 2 -2V i V j )G ij (1-16)

[0082] Furthermore, the line loss sensitivity calculation formulas (1-10)-(1-11) can be simplified to:

[0083]

[0084] Where, These are respectively equal to the active voltage sensitivity S at node i. Ui-P Reactive voltage sensitivity S Ui-Q . It can be represented as:

[0085]

[0086] Cost sensitivity

[0087] Since the real-time stage in this embodiment needs to modify the intraday scheduling plan based on the real-time power demand and the actual power of the distributed resources, this section establishes a real-time operating cost function for adjustable devices, which includes the adjustment cost of each device and the power adjustment penalty cost of each device. The cost sensitivity of each adjustable device is further derived to simplify the real-time optimization objective function.

[0088] For distributed photovoltaic (PV) systems, the real-time operating cost consists of the real-time distributed PV curtailment cost, which is composed of equations (1-21)-(1-22), and the real-time power regulation penalty cost, as shown in equations (1-21)-(1-23):

[0089]

[0090] In the formula, C PV,re Real-time operating costs of distributed photovoltaic systems; The real-time curtailment penalty coefficient is the value of the real-time electricity price; The real-time active power reduction of photovoltaic power at time t; P represents the real-time active and reactive power corrections for photovoltaic power relative to the intraday plan; PV,w,t,re Q PV,w,t,re Real-time active and reactive power output of photovoltaic power; P PV,w,t,in Q PV,w,t,in These are reference values ​​for the active and reactive power of photovoltaic systems during the day.

[0091] For electric vehicle charging stations, the real-time operating cost includes the real-time electric vehicle compensation cost and the real-time power regulation penalty cost, which are composed of equations (1-24)-(1-25), as shown in equations (1-24)-(1-25):

[0092]

[0093] In the formula, C EV,re Real-time operating costs of electric vehicle charging stations; The compensation coefficient is the real-time unit compensation power. Real-time power compensation for electric vehicle charging stations; P represents the real-time adjustment of electric vehicle charging stations relative to the intraday plan. EV,c,t,re Real-time power compensation for electric vehicle charging stations; P EV,c,t,in Reference value for daily compensation power of electric vehicle charging stations.

[0094] For SVC, its real-time operating cost is the real-time power regulation penalty cost, as shown in equations (1-26)-(1-27):

[0095] C SVC,re =cs SVC,f |ΔQ SVC,n,t,rc | (1-26)

[0096] ΔQ SVC,n,t,rc =Q SVC,n,t,re -Q SVC,n,t,in (1-27)

[0097] In the formula, C SVC,re For the real-time operating cost of SVC; ΔQ SVC,n,t,rc Q is the real-time correction value for SVC. SVC,n,t,re Provides real-time power to SVC; Q SVC,n,t,in This is the intraday power reference value for SVC.

[0098] The real-time operating cost function can be expressed as the sum of the real-time operating costs of distributed photovoltaics, electric vehicle charging stations, and SVCs, as shown in (1-28):

[0099]

[0100] In the formula: C re For real-time operating costs.

[0101] Equation (1-28) is applied to the power adjustment of each device. By taking the derivative, we can obtain the sensitivity of the real-time operating cost function to the adjustment amount of different adjustable devices:

[0102]

[0103] In the formula, S cPV-pw S represents the adjustment cost sensitivity of the w-th photovoltaic cell; cEV-pc Let be the adjustment cost sensitivity of the c-th electric vehicle charging station.

[0104] Equation (1-28) is applied to the power correction amount of each device. ΔQ SVC,n,t,rc Taking the derivative, we can obtain the sensitivities of the real-time operating cost function for different adjustable device correction amounts as shown in equations (1-31)-(1-33):

[0105]

[0106] In the formula, S cPV-prw S represents the sensitivity of the penalty cost for the w-th photovoltaic active power deviation; cPV-qrw S represents the sensitivity of the penalty cost for the w-th photovoltaic reactive power deviation; cEV-prc The sensitivity of the active power deviation penalty cost for the c-th electric vehicle charging station; S cSVC-qrn The reactive power deviation penalty cost sensitivity for the nth SVC.

[0107] Therefore, cost sensitivity S can be obtained. c As shown in equation (1-34):

[0108]

[0109] In the formula, S C-P To adjust the cost sensitivity of the adjustable device in real time; S C-PR S C-QR These represent the sensitivity of the penalty cost for the active and reactive power deviation of adjustable equipment; ΔP re ΔQ re ΔP represents the real-time active and reactive power regulation of the adjustable equipment. rc ΔQ rc To adjust the active and reactive power of the adjustable equipment in real time, the real-time operating cost can be obtained by multiplying the corresponding cost sensitivity with the adjustment and correction amount of the adjustable equipment during the calculation.

[0110] Definition of multiple sensitivity coefficient matrix

[0111] Based on the aforementioned voltage sensitivity, line loss sensitivity, and cost sensitivity, a multi-sensitivity coefficient matrix consisting of two parts, S1 and S2, can be obtained. Combined with the multi-sensitivity real-time sensing model, even with incomplete real-time measurements, the relationships between real-time measurable line loss changes, real-time measurable node voltage changes, real-time operating costs of adjustable equipment, and power changes and corrections of adjustable equipment can be established. Expressing the optimization model uniformly through multi-sensitivity not only reduces the scale of the optimization problem but also lowers its complexity. The power flow and real-time operating costs obtained after adjusting equipment power changes can be written in the form of equation (1-35) based on the multi-sensitivity coefficient matrix:

[0112]

[0113] Therefore, the multiple sensitivity coefficient matrix is... and When establishing the optimization model, the traditional power flow optimization model is simplified based on the multiple sensitivity coefficient matrix: the line loss sensitivity and cost sensitivity form the objective function, and the voltage sensitivity constructs the safety check. The three assist each other to simplify the traditional optimal power flow calculation and realize real-time and fast optimization under the condition of incomplete measurement.

[0114] Real-time sensing model with multiple sensitivity based on radial basis function neural network

[0115] Because sensitivity parameters change with system operating points, errors may occur in voltage and line loss sensitivity, especially when power changes are large, further increasing control errors. However, real-time sensitivity updates based on Jacobian matrix inversion are limited when real-time measurements are missing in distribution networks and low-voltage distribution substations. Therefore, this embodiment proposes a multi-sensitivity real-time sensing model based on deep learning technology. When real-time measurements are missing, it senses the real-time voltage and line loss sensitivity of adjustable equipment nodes, reducing control errors in real-time measurable line losses and optimizing voltage distribution at real-time measurable nodes through real-time sensitivity updates.

[0116] Sensitivity Fitting Principle Analysis

[0117] From the combined analysis of equations (1-2)-(1-8), it can be seen that there is an implicit functional relationship between voltage sensitivity and the voltage of each node and the injected power. Therefore, the solution for voltage sensitivity can be expressed by the relationship (1-36):

[0118]

[0119] In the formula, f U-P f U-Q These represent the functional relationships between active and reactive voltage sensitivity and voltage and injected power, respectively.

[0120] Similarly, from equations (1-16)-(1-20) and the analysis of the relationship between line loss sensitivity and the operating point of the distribution network, it can be seen that line loss sensitivity can also be expressed as the above implicit functional relationship, as shown in equation (1-37):

[0121]

[0122] Where, These represent the functional relationships between active / reactive line loss sensitivity and voltage and injected power, respectively.

[0123] Therefore, when the sensitivity relationship of the distribution network and distribution area cannot be solved by inverting the Jacobian matrix due to incomplete real-time measurements, this embodiment first calculates the corresponding historical voltage / line loss power sensitivity based on the physical model under different historical operating conditions, according to the historical power data and voltage data of the real-time measurable nodes. Then, the mapping relationship between power, voltage data and voltage / line loss sensitivity is fitted by training a neural network to establish a real-time sensitivity sensing model under incomplete measurements, thereby laying the model foundation for the real-time sensitivity update of the distribution network and microgrid.

[0124] Radial basis neural networks and their perceptual architecture

[0125] Radial Basis Function Neural Network (RBF) is a commonly used three-layer feedforward network. It can construct different network topologies for different targets, has strong learning capabilities for data sample features, and can quickly solve multi-dimensional space interpolation and extrapolation. It can approximate arbitrary nonlinear functions and does not suffer from local optima. It is widely used in function fitting, image processing, pattern recognition, and other fields. A real-time sensitivity sensing architecture based on RBF neural networks is shown below. Figure 1 As shown.

[0126] This embodiment uses an RBF neural network to construct an "offline training, online sensing" architecture. During offline training, historical active and reactive power and voltage of real-time measurable nodes at different operating points are used as inputs, and voltage sensitivity and line loss sensitivity of each node are used as outputs. This trains a fitting model of real-time measurable node measurement data and voltage / line loss sensitivity. During online sensing, the corresponding real-time measurable node data is input, and the real-time voltage / line loss sensitivity of adjustable device nodes is output. This allows for real-time updates and corrections to the sensitivity of the distribution network and microgrid after changes in operating points, even when real-time measurements are incomplete. This architecture can accurately obtain precise sensitivity even with large power variations, improving the accuracy of control model parameters and demonstrating stronger robustness compared to traditional measurement-driven methods.

[0127] Optimization of Social Network Search Algorithm for RBF Neural Network

[0128] In RBF neural network training, network creation is essentially a process of continuous trial and error. By precisely adjusting the number of neurons in the intermediate layers, the network continuously approximates the function to be fitted. The expansion rate of the radial basis functions is a crucial indicator affecting the smoothness of function fitting; both excessively high and excessively low expansion rates will lead to a decrease in network fitting performance. This embodiment employs a Social Network Search (SNS) algorithm to optimize the expansion rate, thereby improving the training effect of the RBF neural network.

[0129] (1) Social network search algorithm

[0130] The social network search algorithm is a novel intelligent optimization algorithm proposed by Siamak Talatahari in 2021. This algorithm optimizes by simulating how users express emotions in social networks, with each emotion acting as an optimization operator. The corresponding model is typically built using the following methods:

[0131] 1) Imitation

[0132] Emotional mimicry can be modeled using formula (1-38):

[0133] x m,new =x m +rand(-1,1)×rand(0,1)×(x m -x z (1-38)

[0134] Where x m 、x z These refer to the opinions of the m-th and z-th users selected randomly, respectively; x m,new This represents the new feedback from users after the optimization.

[0135] 2) Communication

[0136] The communication of emotions is shown in equation (1-39):

[0137] x m,new =x l +rand(0,1)×sign(f m -f z )×(x m -x z (1-39)

[0138] Where x l The l-th randomly selected conversation partner; the Sign function compares f m f z Determine x l The direction.

[0139] 3) Debate

[0140] The debate, by selecting a random number of comments or group members, yields formula (1-40):

[0141]

[0142] Where x e It is the e-th user selected randomly; N e The size of the comment or group is a random number between 1 and the total number of users.

[0143] 4) Innovation

[0144] Innovation causes an overall update by updating a certain dimension, as shown in equation (1-41):

[0145]

[0146] In the formula, d refers to the d-th dimension randomly selected in the search space; ub d and lb d These are the upper and lower limits corresponding to the d-th dimension.

[0147] (2) Optimize the process

[0148] The optimization aims to minimize the average percentage error of the test set, as shown in equation (1-42):

[0149]

[0150] In the formula, y RBF,f y is the f-th fitted output value of the RBF neural network; test,f These are the true values ​​in the test set.

[0151] The steps for optimizing the expansion speed of the RBF neural network using a social network search algorithm are described in Table 1:

[0152] Table 1 Optimization Steps

[0153]

[0154]

[0155] Real-time Optimization Dispatch Strategy for Distribution Networks Based on Model Predictive Control

[0156] Because the real-time output of distributed resources is volatile and random, higher requirements are placed on the economic and safe operation of the real-time distribution network. Therefore, this embodiment establishes a real-time optimization scheduling model for the distribution network based on model predictive control to adjust the adjustable equipment in the distribution network in real time.

[0157] This embodiment is based on a model predictive control method, combined with a multi-sensitivity sensing model. It performs rolling optimization of line losses and real-time operating costs of measurable lines in the distribution network based on line loss sensitivity and cost sensitivity. During the optimization process, safety checks are performed on measurable nodes based on voltage sensitivity. Feedback corrections are made to the predictive model using a multi-sensitivity real-time sensing model and real-time electricity prices, overcoming the limitations of traditional physical models and improving control accuracy. The real-time optimization process of the distribution network is as follows: Figure 2 As shown.

[0158] Model predictive control principle

[0159] Model predictive control (MDI) is a closed-loop optimization control strategy based on the finite time domain. It boasts advantages such as simple modeling, strong ability to handle uncertainties, and effective handling of system disturbances. Currently, MDI is widely used in industrial control fields such as aerospace and chemical engineering.

[0160] Model predictive control consists of three parts: predictive model, rolling optimization, and feedback correction. Figure 3 The diagram illustrates its structure. In this algorithm, the prediction model predicts the system's output for future stages based on current and past input / output quantities and changes in control quantities. The rolling optimization model constructs an objective function and related constraints based on the deviation between the expected and predicted output values, optimizing the optimal control sequence for the controlled object within a finite time domain. In rolling optimization, only the optimization result of the first time period in the optimal control sequence is executed, thereby improving the real-time performance and accuracy of system control. The feedback correction model collects real-time system output values ​​through secondary equipment at each sampling time and feeds these real-time output values ​​back to the prediction model, improving the prediction accuracy of the prediction model.

[0161] Predictive Model

[0162] A predictive model uses historical information and future inputs of the controlled object to predict the future output of the system. Model predictive control does not impose strict restrictions on the form of the predictive model; therefore, any model that can predict the future state of the system can be used as a predictive model. This embodiment establishes predictive models for line loss, voltage, and real-time operating costs based on a multiple sensitivity coefficient matrix.

[0163] The prediction models for line loss, voltage, and cost are shown in equations (1-43)-(1-45), respectively:

[0164]

[0165] V(k+i|k)=V(k+i-1|k)+S U-P ΔP re +S U-Q ΔQre (1-44)

[0166] C re (k+i|k)=S c-P ΔP re +S c-PR ΔP rc +S c-QR ΔQ rc (1-45)

[0167] In the formula, i represents the prediction step number, i = 1, 2, 3, ..., N p -1,N p V(k+i|k) represents the predicted voltage amplitude at time k+i from time k; P lossij (k+i|k) represents the line loss predicted at time k+i from time k; C re (k+i|k) represents the value of the operating cost function at time k, predicting the future operating cost at time k+i; ΔP re ΔQ re Let ΔP be the variable in the adjustable device to be optimized for control at time k+i-1. re =[ΔP PV,w,re ΔP EV,r,re ], ΔQ re =[ΔQ PV,w,re ΔQ SVC,n,re That is, the active power control variable of the regulating equipment includes the photovoltaic active power reduction ΔP. PV,w,re Active power regulation ΔP of electric vehicles EV,r,re The reactive power control variables of the regulating equipment include the photovoltaic reactive power regulation amount ΔQ. PV,w,re SVC reactive power regulation ΔQ SVC,n,re ;ΔP rc ΔQ rc The corresponding real-time power correction for adjustable equipment is obtained by subtracting the power of the adjustable equipment at time k+i from the power of that period in the daily scheduling plan, as shown in equations (1-46)-(1-47):

[0168]

[0169] In the formula, P(k+i|k) and Q(k+i|k) are the adjustable equipment output predicted at time k+i in the future; P(k+i) and Q(k+i) are the adjustable equipment output at time k+i in the intraday scheduling plan.

[0170] Rolling optimization model

[0171] (1) Objective function

[0172] In the rolling optimization process, this embodiment establishes a real-time optimization model with the objectives of minimizing the line loss of real-time measurable lines and minimizing the real-time operating cost of adjustable equipment. The model uses a 3-minute time scale to continuously optimize and solve the sequence of control variables for future finite time periods, helping to smooth out real-time fluctuations in distributed resources and further ensuring the safe and economical operation of the distribution network, as shown in equation (1-48).

[0173]

[0174] In the formula, P lossij C can be calculated from equation (1-43); re The result can be obtained from equation (1-45); o is the oth real-time measurable line, O is the set of real-time measurable lines; tou is the real-time electricity price;

[0175] This embodiment measures the voltage optimization effect by voltage deviation, as shown in equation (1-49):

[0176]

[0177] In the formula, V bais This is the voltage deviation value; V i V i,ref Here are the node voltages and their reference values.

[0178] (2) Constraints

[0179] The rolling optimization model must satisfy the operational constraints of distributed photovoltaic systems, electric vehicle charging systems, and SVC systems. Simultaneously, real-time measurable nodes must meet voltage safety constraints.

[0180] Feedback correction model

[0181] To further reduce the impact of randomness and volatility on real-time optimization scheduling, the real-time measurement data of the new round of system is used as the initial state for the new round of rolling optimization, thus forming a closed-loop feedback.

[0182] X0(k)=X real (k-1) (1-50)

[0183] In the formula, X0(k) is the initial state value of the system during rolling optimization at time k; X real (k) represents the actual voltage measurement value of the nodes that can be measured in real time and the actual line loss value of the lines that can be measured in real time at time k-1.

[0184] Meanwhile, to mitigate the adverse impact of sensitivity errors caused by changes in the system's operating point on dispatch reliability, a sensitivity error correction feedback mechanism was added, forming a closed-loop feedback mechanism together with the aforementioned closed loop. This new closed-loop feedback mechanism, by invoking a multi-sensitivity sensing model, receives real-time input of the actual measured values ​​of the system's real-time measurable nodes and outputs the corrected adjustable device node voltage / line loss sensitivity. Given the relatively small power supply range of the distribution network, and the fact that most of these networks belong to the same management and operation agency within this specific power supply range, a unified real-time electricity price is used to update the real-time adjustment cost sensitivity portion of the cost sensitivity, thereby enhancing the economy and reliability of the system's real-time dispatch.

[0185] Sensitivity-based real-time autonomous model of microgrids

[0186] Given the incomplete real-time measurement capabilities within microgrids, it is impossible to assess their bidirectional coordination capabilities in real time and calculate the dispatchable range of microgrid resources mapped to the common coupling point between the microgrid and the distribution network. This embodiment, during real-time distribution-microgrid coordinated optimization, issues the microgrid multi-resource dispatch commands calculated from intraday distribution-microgrid coordinated optimization. In real-time operation, mean interpolation is used to smoothly transition between intraday dispatch commands, thereby leveraging the real-time coordination effect of the microgrid on the distribution network. If a node voltage exceedance event occurs within the time interval between dispatch command issuances, a rapid real-time autonomous optimization response is triggered, achieving voltage autonomy within the microgrid. This ensures the safe operation of the microgrid while reducing the impact of power fluctuations within the microgrid on the distribution network.

[0187] The real-time autonomous optimization model for microgrids assumes that the voltage of real-time measurable nodes within the microgrid meets safety constraints. It aims to minimize the output of controllable equipment and optimize the output of multiple resources within the microgrid. The objective function is evaluated using a power deviation penalty cost, as shown in equation (1-51).

[0188]

[0189] In the formula, P tai,a,re Q tai,a,re These represent the real-time active and reactive power outputs of device a in the microgrid, respectively; P tai,a,in Q tai,a,in Reference values ​​for active and reactive power dispatch instructions within the day; S c-pra S c-qra These are the active power and reactive power deviation penalty cost sensitivities of device a in the microgrid, respectively. Their values ​​can be obtained by substituting the power deviation penalty cost sensitivity formula in section 5.2.3 cost sensitivity into the calculation of the microgrid device.

[0190] The voltage after regulation within the microgrid is calculated based on the real-time sensed voltage sensitivity, and the calculation formula is shown in equation (1-52):

[0191]

[0192] In the formula, U tai,s,t For a microgrid that is autonomous in real time at time t, the voltage at node s can be measured in real time; U tai,s,t,0 ΔP represents the voltage at node s of the microgrid before regulation at time t; tai,a,re ΔQ tai,a,re Let be the amount of adjustment change for device a; Let t be the active and reactive voltage sensitivity of the control equipment to node s, and its value can be obtained based on the sensitivity sensing model in the previous section.

[0193] Micro-real-time collaborative optimization scheduling process

[0194] The real-time collaborative optimization scheduling process for distribution networks mainly includes real-time optimization scheduling of distribution networks based on model predictive control and real-time autonomous optimization of microgrids based on voltage sensitivity.

[0195] First, real-time optimization of the distribution network based on model predictive control (MMCC) is performed. Under the constraint of intraday threshold node voltage reference values, rolling optimization is conducted on a 3-minute timescale to obtain the real-time optimized dispatch strategy for the distribution network. Second, real-time autonomous optimization of the microgrid based on voltage sensitivity is implemented. Within the time interval of dispatch command issuance, the microgrid internal node voltage over-limit events are triggered, and the multi-resource optimization dispatch commands within the microgrid are corrected based on real-time sensed voltage sensitivity. Through spatiotemporal dual coordination between distribution and microgrids, the real-time safe and economical operation of distribution stations is ensured, further reducing the impact of real-time power fluctuations of distributed resources on the distribution network. The real-time collaborative optimization dispatch process between distribution and microgrids is as follows: Figure 4 As shown, the specific steps are as follows:

[0196] Step 1: Initialize the parameters of each control device in the system, import the day-ahead OLTC and CB device scheduling results and the intraday distribution gateway node voltage into the distribution network real-time optimization model as constraints, and use the intraday continuous device scheduling instructions as references.

[0197] Step 2: Based on the current operating status of the distribution network, multiple sensitivities are perceived using a multi-sensitivity sensing model; using the current real-time measurable line loss and real-time measurable node voltage as initial values, voltage, line loss, and cost prediction models are established according to equations (1-43)-(1-45).

[0198] Step 3: With the goal of minimizing the real-time line loss of the measurable line and the real-time operating cost of the adjustable equipment, and with the constraint that the real-time measurable node voltage meets the safe operation requirements, establish a rolling optimization model to solve for the optimal control sequence of the adjustable equipment for Np time periods.

[0199] Step 4: Issue the first instruction in the optimal control sequence of the adjustable device.

[0200] Step 5: Issue intraday reference instructions for the scheduling of multiple resources within the microgrid, and smoothly transition between intraday instructions at a real-time scale.

[0201] Step 6: If an internal node voltage over-limit event occurs in the microgrid within the time interval of the dispatch command, the sensitivity-based real-time autonomous microgrid model is triggered, and the real-time voltage autonomy of the microgrid is performed according to equations (1-51)-(1-52).

[0202] Step 7: The real-time measurement data of the measurement system is used as the initial state for a new round of rolling optimization. The real-time voltage / line loss sensitivity is corrected based on the feedback of the multi-sensitivity sensing model, the cost sensitivity is updated based on the real-time electricity price, and the process returns to Step 2 to execute a new round of optimization.

[0203] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing voltage scheduling based on multi-sensitivity sensing and micro-cooperative voltage, characterized in that, Includes the following steps: Step 1: Establish a multi-sensitivity coefficient matrix, including voltage sensitivity, line loss sensitivity, and cost sensitivity; in the distribution network and microgrid system, write each regulating device node into the augmented matrix of the power flow equation, and simplify to obtain the power flow correction equation expressed in polar coordinates of node voltage: Where, ΔV is the Jacobian matrix; ΔV and Δθ are the voltage magnitude and phase angle correction of the node, respectively; ΔP and ΔQ are the changes in active and reactive power injection at each node. Inverting the Jacobian matrix yields the relationship between ΔV, Δθ and ΔP, ΔQ. When the Newton-Raphson method converges, S... 21 S 22 These are the active and reactive voltage sensitivities at the nodes, respectively. In the formula, The block submatrices H, N, K, and L in the Jacobian matrix are respectively equal to Step two involves establishing a multi-sensitivity real-time sensing model, including sensitivity fitting principle analysis, radial basis function neural network and its sensing architecture, and optimization of the RBF neural network using a social network search algorithm. An implicit functional relationship exists between voltage sensitivity and the voltage of each node and the injected power. In the formula, f U-P f U-Q These represent the functional relationships between active and reactive voltage sensitivity and voltage and injected power, respectively. Step 3: Real-time optimization and dispatching strategy for distribution networks based on model predictive control, including prediction models and feedback correction models; line loss, voltage, and cost prediction models are respectively: V(k+i|k)=V(k+i-1|k)+S U-P ΔP re +S U-Q ΔQ re ; C re (k+i|k)=S c-P ΔP re +S c-PR ΔP rc +S c-QR ΔQ rc ; In the formula, i represents the prediction step number, i = 1, 2, 3, ..., N p -1,N p V(k+i|k) represents the predicted voltage amplitude at time k+i from time k; P lossij (k+i|k) represents the line loss predicted at time k+i from time k; C re (k+i|k) represents the value of the operating cost function at time k, predicting the future operating cost at time k+i; ΔP re ΔQ re Let ΔP be the variable in the adjustable device to be optimized for control at time k+i-1. re =[ΔP PV,w,re ΔP EV,r,re ], ΔQ re =[ΔQ PV,w,re ΔQ SVC,n,re That is, the active power control variable of the regulating equipment includes the photovoltaic active power reduction ΔP. PV,w,re Active power regulation ΔP of electric vehicles EV,r,re The reactive power control variables of the regulating equipment include the photovoltaic reactive power regulation amount ΔQ. PV,w,re SVC reactive power regulation ΔQ SVC,n,re ;ΔP rc ΔQ rc This is the corresponding real-time power correction amount for adjustable equipment, and its value is obtained by subtracting the power of the adjustable equipment at time k+i from the power of that period in the daily scheduling plan; Step four: Based on the sensitivity-based real-time autonomous model of the microgrid, the objective function is evaluated using the power deviation penalty cost: In the formula, P tai,a,re Q tai,a,re These represent the real-time active and reactive power outputs of device a in the microgrid, respectively; P tai,a,in Q tai,a,in Reference values ​​for active and reactive power dispatch instructions within the day; S c-pra S c-qra These represent the active power and reactive power deviation penalty cost sensitivity of device a in the microgrid, respectively. Step 5, the real-time collaborative optimization and scheduling process for distribution networks and microgrids, includes real-time optimization and scheduling of distribution networks based on model predictive control and real-time autonomous optimization of microgrids based on voltage sensitivity.

2. The method for optimizing voltage scheduling based on multiple sensitivity sensing according to claim 1, characterized in that: In step one, the active power loss for line loss sensitivity is: P lossij =(V i 2 +V j 2 -2V i V j cosθ ij )G ij ; In the formula, P lossij Let be the line loss between node i and node j.

3. The method for optimizing voltage scheduling based on multiple sensitivity sensing according to claim 1, characterized in that: In step one, cost sensitivity S c This includes revising the intraday scheduling plan based on real-time power demand and the actual power of distributed resources, as well as calculating the cost sensitivity of each adjustable device, including the adjustment cost of each device and the power adjustment penalty cost of each device.

4. The micro-cooperative voltage optimization scheduling method based on multiple sensitivity sensing according to claim 1, characterized in that: In step one, based on voltage sensitivity, line loss sensitivity, and cost sensitivity, and combined with a multi-sensitivity real-time sensing model, the relationship between the real-time measurable line loss change, the real-time measurable node voltage change, the real-time operating cost of adjustable equipment, and the power change and correction of adjustable equipment is established when real-time measurements are incomplete.

5. The method for optimizing voltage scheduling based on multiple sensitivity sensing according to claim 1, characterized in that: In step two, the voltage sensitivity in the sensitivity fitting principle analysis is: In the formula, f U-P f U-Q These represent the functional relationships between active and reactive voltage sensitivity and voltage and injected power, respectively. Line loss sensitivity can also be expressed as an implicit function relationship: Where, These represent the functional relationships between active / reactive line loss sensitivity and voltage and injected power, respectively.

6. The method for optimizing voltage scheduling based on multiple sensitivity sensing according to claim 1, characterized in that: In step three, based on the model predictive control method and combined with the multi-sensitivity sensing model, the line loss and real-time operating cost of the distribution network's real-time measured lines are calculated in a rolling manner based on line loss sensitivity and cost sensitivity. During the calculation process, the real-time measurement nodes are checked for safety based on voltage sensitivity, and the prediction model is corrected by feedback through the multi-sensitivity real-time sensing model and real-time electricity price.

7. The method for optimizing voltage scheduling based on multiple sensitivity sensing according to claim 1, characterized in that: In step three, the closed-loop feedback of the feedback correction model is: X0(k)=X real (k-1); In the formula, X0(k) is the initial state value of the system during rolling optimization at time k; X real (k) represents the actual voltage measurement value of the nodes that can be measured in real time and the actual line loss value of the lines that can be measured in real time at time k-1.

8. The method for optimizing voltage scheduling based on multiple sensitivity sensing according to claim 1, characterized in that: In step four, the voltage after regulation within the microgrid is calculated based on the real-time sensed voltage sensitivity: Where U tai,s,t For a microgrid that is autonomous in real time at time t, the voltage at node s can be measured in real time; U tai,s,t,0 ΔP represents the voltage at node s of the microgrid before regulation at time t; tai,a,re ΔQ tai,a,re Let be the amount of adjustment change for device a; Let be the active and reactive voltage sensitivity of the control equipment at time t to node s.

9. The micro-cooperative voltage optimization scheduling method based on multiple sensitivity sensing according to claim 1, characterized in that: Step five involves the following specific steps: S1: Initialize the parameters of each control device in the system; S2: Based on the current operating status of the distribution network, multiple sensitivities are perceived using a multi-sensitivity sensing model; voltage, line loss and cost prediction models are established using the current real-time measurable line loss and real-time measurable node voltage as initial values. S3: With the goal of minimizing the real-time measurable line loss and the real-time operating cost of adjustable equipment, and with the constraint that the real-time measurable node voltage meets the safe operation requirements, a rolling optimization model is established to solve the optimal control sequence of adjustable equipment for Np time periods. S4: Issue the first instruction in the optimal control sequence for the adjustable device; S5: Issue intraday reference instructions for multi-resource scheduling within the microgrid, and smoothly transition between intraday instructions at a real-time scale; S6: If an internal node voltage over-limit event occurs in the microgrid within the time interval of the dispatch command, the sensitivity-based real-time autonomous microgrid model is triggered, and real-time voltage autonomy of the microgrid is performed. S7: The real-time measurement data of the measurement system serves as the initial state for a new round of rolling optimization. The real-time voltage / line loss sensitivity is corrected based on feedback from the multi-sensitivity sensing model, the cost sensitivity is updated based on the real-time electricity price, and the system returns to S2 to execute a new round of optimization.

Images