CN114421470B - Intelligent real-time operation control method for flexible diamond type power distribution system - Google Patents

Abstract

The invention discloses an intelligent real-time operation control method for a flexible diamond type power distribution system, which comprises the steps of solving a plurality of typical working conditions by adopting a simulated annealing algorithm with global optimization capability in one stage, constructing a training set to provide data support for an artificial neural network, and carrying out network training in an off-line manner and rapidly solving on line when the working conditions with the solution are received by adopting the artificial neural network in the second stage so as to provide the optimal operation state of a corresponding flexible interconnection device. Aiming at the real-time operation optimization problem of the flexible interconnection device of the flexible diamond type power distribution network, the invention provides a two-stage optimization method based on a simulated annealing algorithm and an artificial neural network by taking the minimum network loss as an optimization target, solves a typical working condition by the simulated annealing algorithm, constructs a data set, and greatly shortens the calculation time by utilizing the offline training and online calculation of the artificial neural network, thereby meeting the requirement of real-time optimization.

Description

Intelligent real-time operation control method for flexible diamond type power distribution system

Technical Field

The invention relates to an intelligent real-time operation control method for a flexible diamond type power distribution system, which is used in the field of intelligent control of a power grid.

Background

The distribution network is at the hub location between the transmission network and the load, and takes on the important tasks of receiving, distributing and supplying electric energy. The traditional power distribution network adopts a radial framework and has the characteristic of unidirectional tide. In recent years, the energy crisis puts higher demands on reducing emission and improving the utilization of renewable energy, the permeability of the distributed power supply is gradually improved, but the intermittence and the uncertainty of the height of the distributed power supply cause the fluctuation of the running state of the power distribution network, and cause the problems of voltage regulation and the like. On the other hand, a large number of novel load and energy storage systems such as electric vehicles are connected, and the tide in the power distribution network is more complicated.

Distribution networks are facing new pressures and challenges in terms of reliability, flexibility, while the means of regulation of traditional primary equipment have been difficult to deal with: the regulation capability and precision of reactive compensation capacitor bank switching and on-load tap regulation of the voltage regulating transformer are limited; the network reconstruction method based on the section switch and the interconnection switch has the problems of limitation of switching-on impact current, switching action time, reconstruction times, incapability of continuous adjustment and the like. Therefore, the traditional power distribution network is gradually changing to a new form of intelligent power distribution network and active power distribution network.

The Flexible Interconnection Device (FID) becomes a core device for power transmission and conversion by virtue of the characteristics of strong adjusting capability, high response speed and the like. The flexible interconnection device FID is realized based on power electronic devices, is used for replacing a contact switch in a power distribution network, and can independently regulate and control active power and reactive power in real time and continuously. Replacing the TS with FID, the main functions and advantages include: 1) Reactive support is provided, the voltage level of a feeder line is improved, and voltage overrun is avoided; 2) The DG consumption capacity of the power distribution network is improved; 3) Adjusting the power flow distribution and balancing the feeder line load; 4) Network loss is reduced, and operation economy is improved; 5) And the load transfer is realized during the fault, and the power supply reliability is ensured.

In order to make the FID fully exert the above advantages in the distribution network, several solutions are proposed for the operational control problem of the FID. One scheme provides an operation optimization model of the power distribution network containing the FID, and the operation optimization model is compared with network reconstruction, so that the effects of the FID in the aspects of reducing network loss, improving voltage and the like are verified; in the second scheme, the coexistence of FID and TS and the coexistence of FID and a reactive compensation device are considered, a model in which two adjusting means participate in optimization is constructed, and a hybrid optimization algorithm based on a simulated annealing algorithm and a cone optimization method, a genetic algorithm and a primal-dual interior point method is adopted for solving; the third approach investigated the FID planning problem and also investigated its operational control as a sub-problem. But the device losses of the FID itself are either ignored or not considered to be complete, only the losses generated by transmitting the active power are calculated, not the losses generated by supplying the reactive power; meanwhile, the optimization speed of the operation is discussed less, and extra means for accelerating the calculation are not considered, so that the requirement of real-time optimization is met.

Disclosure of Invention

The invention aims to overcome the defects of the prior art, provides an intelligent real-time operation control method for a flexible diamond type power distribution system, aims at minimizing the total loss (including the loss of an FID (field induced device)) of a power distribution network, provides a simulated annealing-artificial neural network two-stage method for rapidly solving, and can meet the requirement of real-time optimization.

One technical scheme for achieving the above purpose is as follows: an intelligent real-time operation control method for a flexible diamond power distribution system comprises the following steps

Step 1, selecting a plurality of typical working conditions;

step 2, solving by using a simulated annealing algorithm to obtain the optimal operation state of the flexible interconnection device under each working condition;

step 3, taking the working condition and the optimal running state of the corresponding flexible interconnection device as a training set of the artificial neural network for training;

and 4, quickly giving the optimal running state of the flexible interconnection device when the trained artificial neural network receives a new working condition.

Further, step 2 specifically includes:

step 2.1, setting a hyper-parameter of a simulated annealing algorithm;

2.2, selecting a typical working condition to initialize parameters of the simulated annealing algorithm;

step 2.3, randomly generating an initial state to obtain the corresponding system energy;

step 2.4, randomly generating a new state to obtain the corresponding system energy;

step 2.5, judging whether the energy of the system in the new state is accepted according to the Metropolis criterion, if not, returning to the step 2.4, and if so, entering the step 2.6;

step 2.6, transferring the system state and reducing the temperature;

step 2.7, judging whether the ending temperature is reached, otherwise returning to step 2.4, and if so, entering step 2.8;

step 2.8, obtaining the optimal power combination of the flexible interconnection device under the typical working condition, and putting the optimal power combination into an artificial neural network training set;

and 2.9, judging whether all the typical working conditions are traversed, otherwise, returning to the step 2.2, and if so, entering the step 3.

Further, step 3 specifically includes:

step 3.1, setting the hyper-parameters of the artificial neural network, and reading a training set;

step 3.2, forward propagation is carried out, and a loss function is calculated;

step 3.3, updating parameters of the artificial neural network by a back propagation and gradient descent method;

and 3.4, judging whether the training times are reached, otherwise returning to the step 3.2, and if so, finishing the artificial neural network training.

The invention has the beneficial effects that:

1. according to the reasonable operation control mechanism of the existing power distribution network flexible interconnection device, the device loss of the flexible interconnection device is ignored or not considered completely, only the loss generated by transmitting active power is calculated, but the loss generated by providing reactive power is not calculated, meanwhile, the operation optimization speed is discussed less, extra means for accelerating calculation are not considered, so that the requirement of real-time optimization is met, and the problems existing in the two-stage method of simulated annealing and artificial neural network are considered and solved quickly, so that the requirement of real-time optimization can be met.

2. According to the simulated annealing-artificial neural network two-stage optimization method, the node voltage can be effectively improved, the node voltage is prevented from exceeding the limit, and the network loss is reduced before and after the access optimization is performed through the flexible interconnection device.

Drawings

FIG. 1 is a schematic diagram of an algorithm flow of an intelligent real-time operation control method for a flexible diamond power distribution system according to the present invention;

FIG. 2 is a schematic diagram of a topology structure of a BTB VSC-based FID for research and analysis according to the present invention;

FIG. 3 is a schematic diagram of an artificial neural network of an intelligent real-time operation control method for a flexible diamond-type power distribution system according to the present invention;

FIG. 4 is a schematic diagram of an exemplary 9-node algorithm for verifying the algorithm of the present invention using a 9-node system.

Detailed Description

In order to better understand the technical solution of the present invention, the following detailed description is made by specific examples:

the invention discloses an intelligent real-time operation control method for a flexible diamond type power distribution system, which consists of a Simulated Annealing (SA) and Artificial Neural Network (ANN) two-stage optimization method.

The simulated annealing is a random search meta-heuristic algorithm provided by referring to the annealing process of the solid matter, starting from a set initial temperature, a process that a simulation system gradually approaches to an equilibrium state at the current temperature is simulated, and the temperature of the simulation system is gradually reduced to reach a final equilibrium state.

In the middle annealing process, when the system is transferred from the current state i to the new state j, whether the accepting state j is the new state is determined according to the Metropolis criterion: if E _j ＜E _i The acceptance state j is a new state; otherwise, the following probability acceptance state j is taken as a new state:

E _i ，E _j the system energy corresponding to the states i and j respectively, and t is the current temperature of the system. The criterion enables the system to receive a new state with higher energy than the existing state with higher probability at high temperature, thereby effectively avoiding falling into local optimum; followed byAs the temperature decreases, the probability of accepting new states with higher energy than the current state is also lower, thereby speeding up the convergence to the global optimum.

The artificial neural network is a machine learning model which is inspired by biology and neuroscience, and simulates a biological neural network by artificial neurons and a connected topological structure thereof, and has strong fitting capability on a nonlinear function. The artificial neurons constitute an input layer, an output layer, and zero to several hidden layers therebetween. The training process can be divided into forward propagation and backward propagation processes. In the forward propagation process, information propagation from each neural layer to the next neural layer is equivalent to affine transformation; meanwhile, in order to enhance the representation and learning capability of the network and ensure the fitting effect of the nonlinear function, a nonlinear function is required to be introduced behind each hidden layer, and the function is called as an activation function; at the output level, a loss function is introduced to evaluate the difference between the network output and the reference value. And in the back propagation process, starting from the loss function of the output layer, calculating error terms of each layer, and updating the weight and the bias parameters of affine transformation between the nerve layers according to a gradient descent method. And finally obtaining a group of network parameters by alternately carrying out the two processes until the set training times are reached, and finishing the artificial neural network training. When the trained artificial neural network is applied, only one forward propagation process needs to be carried out, so that the calculation speed is high.

The invention provides a two-stage optimization method based on Simulated Annealing (SA) and Artificial Neural Network (ANN): selecting a plurality of typical working conditions, respectively solving by using a simulated annealing algorithm to obtain the optimal running state of the FID under each working condition, and then taking the working conditions and the corresponding optimal running state of the FID as a training set of the artificial neural network, wherein the network training process can be carried out off-line; when the trained artificial neural network receives a new working condition, the corresponding FID optimal operation state can be given at a very high speed, and the calculation time is greatly shortened so as to meet the requirement of real-time optimization.

Referring to fig. 1, the intelligent real-time operation control method for a flexible diamond power distribution system of the present invention specifically includes the following steps:

step 1, inputting various network parameters and selecting a plurality of typical working conditions.

And 2, solving (SA) by using a simulated annealing algorithm to obtain the optimal operation state of the flexible interconnection device under each working condition.

The Solving (SA) using the simulated annealing algorithm specifically includes:

and 2.1, setting a hyper-parameter of the simulated annealing algorithm.

And 2.2, selecting a typical working condition to carry out parameter initialization on the simulated annealing algorithm.

And 2.3, randomly generating an initial state to obtain the corresponding system energy.

And 2.4, randomly generating a new state and solving the energy of the corresponding system.

And 2.5, judging whether the energy of the system in the new state is accepted according to the Metropolis criterion, otherwise returning to the step 2.4, and if so, entering the step 2.6.

And 2.6, transferring the system state and reducing the temperature.

And 2.7, judging whether the ending temperature is reached, otherwise returning to the step 2.4, and if so, entering a step 2.8.

And 2.8, obtaining the optimal power combination of the flexible interconnection device under the typical working condition, and putting the optimal power combination into an artificial neural network training set.

And 2.9, judging whether all typical working conditions are traversed, otherwise returning to the step 2.2, and if so, entering the step 3.

Step 3, training the working conditions and the corresponding optimal running state of the flexible interconnection device as a training set of an Artificial Neural Network (ANN), and specifically comprising the following steps:

step 3.1, setting the hyper-parameters of the artificial neural network, and reading a training set;

step 3.2, forward propagation is carried out, and a loss function is calculated;

step 3.3, updating parameters of the artificial neural network by a back propagation and gradient descent method;

and 3.4, judging whether the training times are reached, otherwise returning to the step 3.2, and if so, finishing the artificial neural network training.

And 4, the trained artificial neural network receives and reads the new working condition to be solved, and the Artificial Neural Network (ANN) is called to quickly give the optimal running state of the flexible interconnection device.

The intelligent real-time operation control method of the flexible diamond type power distribution system is implemented by taking a FID structure based on BTB VSC as an example, and is shown in FIG. 2.

The device has multiple control modes, and when in normal operation, PQ-V is often adopted _dc And Q control, namely one converter controls transmitted active power, the other converter controls direct-current side voltage, and in addition, the two converters can independently control reactive power injected into the network. Thus for an FID connected between nodes i and j, there are 3 controllable variables: active power P output by a converter _FID，i Reactive power Q output by two converters respectively _FID，i And Q _FID，j . The operational boundary conditions for the FID include the following constraints:

1) FID capacity constraint:

2) FID active power balance constraint:

P _FID，i +P _FID，j +P _{FID，loss，ij} ＝0

3) FID reactive power constraint:

|Q _FID，i |≤μS _FID，ij

|Q _FID，j |≤μS _FID，ij

in the above formulas, S _FID，ij Capacity of the FID connected between nodes i and j; p is _FID，i ，P _FID，j ， Q _FID，I ，Q _FID，j Respectively outputting active power and reactive power of the FID, and taking the injected power grid as positive power; p _{FID，loss，ij} Device losses for the FID; c. C _FID，loss Is the loss factor of the FID; μ is the reactive power limiting coefficient of the FID.

The invention takes the minimum network loss (including FID device loss) as an objective function, and the optimized objective mathematical expression is as follows:

wherein: p _loss The total loss of the network; n is the number of network nodes; p is _inj，i Injecting active power into the node i; p _G，i The active power of the power supply connected for the node i takes the injection power grid as positive; p _LD，i The load active power connected for node i is taken positive from the grid.

The following constraints exist when the power distribution network operates:

1) And (3) system flow constraint:

2) Node voltage constraint:

V _i，min ≤V _i ≤V _i，max

3) And (3) branch current constraint:

in the above formulas, N (i) is a set of nodes adjacent to the node i; p _inj，i ，Q _inj，i Injecting active and reactive power on node i; p _G，i ，Q _G，i And P _FID，i ，Q _FID，i The active power and the reactive power of a power supply and an FID which are respectively connected with the node i are both positive by being injected into a power grid; p _LD，i ，Q _LD，i The active power and the reactive power of the load connected with the node i are obtained from the power grid as positive power; v _i Is the voltage amplitude of node i; v _i，max ，V _i，min The upper limit and the lower limit of the voltage amplitude of the node i are set; i is _ij The current amplitude of branch ij; i is _ij，max The upper limit of the current amplitude of branch ij.

The content of the FID optimization operation model of the power distribution network takes the minimum total network loss as the constraint condition of the objective function, and the constraint condition comprises FID operation boundary constraint, system power flow constraint and system safety operation constraint.

The invention provides an intelligent real-time operation control method for a flexible diamond type power distribution system, and the large-scale nonlinear optimization problem that FID operation contains a plurality of continuous variables is considered.

Firstly, selecting a plurality of typical working conditions, respectively solving by using a simulated annealing algorithm to obtain the optimal operation state of the FID under each working condition, and then taking the working conditions and the corresponding optimal operation state of the FID as a training set of the artificial neural network, wherein the network training process can be carried out off-line; when the trained artificial neural network receives a new working condition, the corresponding FID optimal operation state can be provided at a very high speed.

Simulated annealing algorithm for solving FID operation optimization problem discussed herein, combining 3 controlled variables (P) of FID _FID，i ，Q _FID，i ，Q _FID，j ) As a system state, solving the system power flow by a Newton-Raphson method in a corresponding system running state, and calculating the network loss P calculated according to the FID optimization target mathematical expression _loss The minimum net loss and its corresponding FID controlled variable combination, hereinafter referred to as the optimum power combination, are determined as the system energy according to the simulated annealing process described above.

The artificial neural network is applied to the FID operation optimization problem, the operation state of the power distribution network is used as network input, the corresponding FID optimal power combination is used as network output, and the network structure is shown in figure 3. Specifically, the input layer is a working condition (k) _G ，k _LD ) Wherein: k is a radical of _G The ratio of the output of each power source to its capacity, k _LD Is the ratio of the power of each load to its rated power(ii) a The output layer is the optimal power combination (P) of the FID under the working condition _FID，i ，Q _FID，i ，Q _FID，j ). The activation function is selected from a ReLU function, the loss function is selected from a Mean Squared Error (MSE) function having the following equation:

in the above formula: x is each component of the hidden layer; y is the vector representation of the output layer, k is its dimension, y _i And y _i ^* The components and their reference values, respectively.

The application of the above structures and methods are further described below with reference to specific simulation examples.

In connection with the above embodiments, the following computer environment that performs the optimized operation control is employed: intel (R) Core (TM) i5-3320M CPU (master frequency 2.60 GHz), a memory 8G and Windows 10 64 bit professional operating system are operated, and the programming environment for realizing the optimization algorithm is Python 3.8. The relevant parameters are shown in tables 1, 2, 3 and 4.

TABLE 1 line parameters

TABLE 2 Power Capacity

/>

TABLE 3 load rating

TABLE 4 FID parameters

The implementation scheme is as follows: the adjusting effect of the invention is verified.

The algorithm was verified using a 9-node system, the structure of which is shown in fig. 4, with the line parameters shown in table 1. The node 1 is connected with a power grid, the nodes 8 and 9 are respectively connected with a power supply capable of providing sufficient reactive support, the nodes 2, 4 and 6 are connected with a load, and the active capacity and the load rated power of the power supply are shown in tables 2 and 3. The FID is connected between the nodes 5, 6 and the relevant parameters are shown in table 4.

Under the condition that the power supply outputs power by capacity and the load is connected according to rated power, partial node voltage and network loss conditions before and after the FID connection nodes 5 and 6 participate in operation optimization are compared, and the FID operation optimization result is obtained and is shown in table 5.

TABLE 5

The implementation scheme is as follows: the accuracy and the rapidity of the invention are verified.

And (3) obtaining 36 typical working conditions by carrying out value selection on kG and kLD from 0.0 to 1.0 respectively by taking 0.2 as a step length, and calculating the FID optimal power combination (PFID, 5, QFID, 6) under each working condition by adopting a simulated annealing algorithm. The neural network hidden layer dimension is set to 50. And (3) respectively taking each working condition (kG, kLD) and the corresponding FID optimal power combination (PFID, 5, QFID, 6) as the input and the output of the neural network, constructing a training set, and training a group of neural network parameters.

The accuracy of the algorithm is verified, a plurality of new working conditions (kG, kLD) are randomly generated to serve as a test set, the simulated annealing algorithm and the trained neural network are used for calculation respectively, and the results of the simulated annealing algorithm and the trained neural network are compared with the operation time, as shown in table 6.

TABLE 6

According to the comparison, the calculation result of the trained artificial neural network under the new randomly given working condition is almost consistent with the result obtained by the simulated annealing algorithm, the calculation time is extremely short, and the requirement of real-time optimization can be met.

The checking result shows that the simulation annealing-artificial neural network two-stage method is based on the simulation annealing-artificial neural network two-stage method, under the condition that the FID real-time operation optimization problem of the power distribution network is solved by using the simulation annealing algorithm and the minimum network loss is taken as the target, the typical working condition is solved, the data set is constructed, the artificial neural network offline training and online calculation are utilized, the calculation time is greatly shortened, the real-time optimization requirement is met, meanwhile, the reasonable operation control of the FID of the power distribution network can play the functions of regulating the system load flow, improving the voltage level, reducing the network loss and the like, and the method is greatly beneficial to the reliability, flexibility and economical operation of the power distribution network.

It should be understood by those skilled in the art that the above embodiments are only for illustrating the present invention and are not to be used as a limitation of the present invention, and that changes and modifications to the above described embodiments are within the scope of the claims of the present invention as long as they are within the spirit and scope of the present invention.

Claims (1)

1. An intelligent real-time operation control method for a flexible diamond power distribution system is characterized by comprising the following steps:

step 1, selecting a plurality of typical working conditions;

step 2, solving by using a simulated annealing algorithm to obtain the optimal operation state of the flexible interconnection device under each working condition;

step 3, training the working condition and the corresponding optimal running state of the flexible interconnection device as a training set of the artificial neural network;

step 4, when the trained artificial neural network receives a new working condition, the optimal operation state of the flexible interconnection device is quickly given,

the step 2 specifically comprises the following steps:

step 2.1, setting a hyper-parameter of a simulated annealing algorithm;

2.2, selecting a typical working condition to initialize parameters of the simulated annealing algorithm;

step 2.3, randomly generating an initial state to obtain the corresponding system energy;

step 2.4, randomly generating a new state, and solving the energy of a corresponding system;

step 2.5, judging whether the energy of the system in the new state is accepted according to the Metropolis criterion, if not, returning to the step 2.4, and if so, entering the step 2.6;

step 2.6, transferring the system state and reducing the temperature;

step 2.7, judging whether the ending temperature is reached, if not, returning to the step 2.4, and if so, entering the step 2.8;

step 2.8, obtaining the optimal power combination of the flexible interconnection device under the typical working condition, and putting the optimal power combination into an artificial neural network training set;

step 2.9, judging whether all typical working conditions are traversed, if not, returning to the step 2.2, if so, entering the step 3,

the step 3 specifically comprises the following steps:

step 3.1, setting the hyper-parameters of the artificial neural network, and reading a training set;

step 3.2, forward propagation is carried out, and a loss function is calculated;

step 3.3, updating parameters of the artificial neural network by a back propagation and gradient descent method;

and 3.4, judging whether the training times are reached, otherwise returning to the step 3.2, and if so, finishing the artificial neural network training.

Images