RU2793231C1 - Method for intelligent control of voltage and reactive power of a power system - Google Patents

Abstract

FIELD: power industry.

SUBSTANCE: method includes installation of a centralized coordinating system (CCS), which contains a coordinating control system connected by communication channels with a database and an optimal control subsystem for voltage and reactive power of a power system (OCS), and at least one decentralized voltage and reactive power control system node (DVRPCS), connected to the CCS by communication channels for data transmission. The coordinating control system receives information about the state of the power system, forms a slice of data on it and transfers it to the OCS, receives a list of control actions and settings of local regulators of operating automation from the OCS and transfers them to the power system in accordance with the required time of their application. The OCS evaluates and processes the data slice, selects the composition of the control actions and settings, generates an output tag for each control action corresponding to the number of the power system node to which the control action is directed. Each decentralized DVRPCS calculates the control actions and settings of local regulators of regime automation in stand-alone operation in the event of the disappearance of all communication channels with the CCS.

EFFECT: minimization of losses in the power system, increased optimization of operating modes of the power system equipment with simultaneous increase in the overall fault tolerance of the power system.

5 cl, 3 dwg

Description

Technical field

The invention relates to the field of electrical energy distribution and is intended for use in the implementation of intelligent control of the operating modes of electric power systems.

State of the art

A known method of phase voltage control in an electrical system (Patent RU 2588058 dated 06/27/2016, IPC H02J 3/12), in which a phase voltage shift is created, characterized in that it is set by the required operating parameters of the electrical system, the mechanical moment of the absolute motion of the rotor of a synchronous electrical machine the systems are divided into relative and portable ones, and the portable moment is controlled by the specified voltage phase shift in accordance with the required operating parameters.

A system and method for controlling an electric power system are known (Patent RU 2518178 dated 06/10/2014, IPC H02J 3/00, H02J 13/00, H04L 12/00, H04L 12/54, G05B 19/00, G05B 19/042), containing to improve grid management of an electric power system, an integration framework for communicating a plurality of electric grid and power system components with a central electric power system control that controls the electric power system, said plurality of electric grid and power system components being configured to generate operational and event data in the electric power system containing:

a work bus configured to transmit operating data to a central control of the electric power system, the operating data comprising at least one real-time measurement of voltage, current, active power, or reactive power for at least a portion of the electric power system;

and an event bus configured to transmit the event data to the central control of the electric power system, wherein the event bus is separated from the operating bus, the event data is different from real-time measurements, and is obtained from said real-time measurements and contains at least one analytical determination regarding the operation of the electric power system based on said at least one real-time measurement, wherein the operating bus is configured to transmit operating data and not transmit event data, and the event bus is configured to transmit event data and not transmit work data.

An intelligent core of the system is known (Patent RU 2541911 dated 20.02.2015, IPC H02J 13/00), the control center of which, in order to increase reliability and speed in managing the power system, such as a power supply system, communicates with several systems for collecting and processing metering data and with several terminal systems. The power system control center contains a gateway layer and a kernel layer. The gateway layer includes multiple input connection procedures for communicating with each of the multiple source systems and multiple output connection procedures for communicating with each of the multiple target systems. The kernel layer contains multiple kernel adapters such that these kernel adapters perform one-to-one translation of communications from multiple command-generating accounting data collection and processing systems to multiple terminal systems.

A known method for controlling the modes of an electric power system (Patent RU 2750260 dated 06/25/2021, IPC H02J 3/06, G05B 13/02, H02J 3/24), aimed at controlling the modes of an electric power system (EPS), which contains energy clusters with groups of combined stations and substations (SS), grid control centers (NCC) and automated process control systems (APCS), as well as power lines between energy clusters.

The NCC of EPS energy clusters is equipped with devices for collecting and storing operational data coming from the SS from the process control systems within the corresponding energy clusters, and with software and hardware platforms with software agents of the multi-agent system (MAC). The NCCs are interconnected by a global computer network and exchange messages from the MAC over it, with the help of which they determine the parameters of distributed optimization, send commands from the NCC to implement the optimized mode back to the automated process control system of the substation within their energy clusters.

As software and hardware platforms, NCCs with MAC software agents use additional computing devices (TLDs) based on industrial servers.

At the beginning of the process of managing EPS modes in each TLD, MAC agents are launched to collect information and calculate the optimal operation mode of the energy cluster controlled from this TLD, optimize the mode inside their energy cluster by MAC agents and determine the boundary parameters of voltage and active power to create a common mode with other energy clusters on based on the interests of the power cluster owner.

Each TLD detects TLDs of adjacent energy clusters connected by common power lines with the energy cluster of this TLD, and the control of the EPS operation mode is started by all TLDs simultaneously and is carried out in parallel mode.

From each TLD, via the global computer network, TLDs of adjacent energy clusters are sent to the voltage values and parameters of active power flows along power lines between energy clusters obtained in the course of calculating the optimal mode of each energy cluster.

Each TLD receives messages from TLDs of adjacent power clusters with restrictions imposed on the voltage and parameters of active power flow along power lines and presented for each TLD in the form of an objective function (TF) of losses, which ensures the maintenance of the levels of boundary voltages of power lines and active power flows. Taking into account these restrictions, as well as criteria determined by business interests, they adjust the optimization of the mode of their energy cluster, stop the interaction of TLDs with each other if the levels of boundary voltages of power lines of adjacent energy clusters and the parameters of active power flows coincide, and also taking into account the interests of all subjects of the optimization process .

The disadvantages of the above technical solutions are a large number of nodes that regulate the control process, as well as a large number of nodes that exchange information about the introduced restrictions or control actions, which, in turn, leads to a decrease in the overall reliability of the system.

Disclosure of the essence of the invention

The technical problem is to create a method for intelligent control of the voltage and reactive power of the power system to minimize losses in this power system.

EFFECT: minimizing losses in the power system, increasing the consistency of optimizing the operating modes of the power system equipment with a simultaneous increase in the overall fault tolerance of the power system.

The technical problem is solved, and the technical result is achieved due to the fact that they use the method of intelligent control of the voltage and reactive power of the power system, which includes the installation of a centralized coordinating system (CCS) for collecting and processing data from the nodes of the power system, as well as the formation and transmission of control actions to the nodes a power system containing a coordinating control system with a database connected by communication channels and an optimal control subsystem (OPS) for voltage and reactive power of the power system, and at least one decentralized nodal voltage and reactive power control system (USUNRM) connected to the MSC by communication channels for data transmission , while the coordinating control system is performed with the ability to obtain information about the state of the power system, form a data slice on it, transfer the data slice to the POU, obtain a list of control actions and settings of local regulators of regime automation from the POU and transfer control actions and settings to the power system in accordance with the required time of their application, the POU is performed with the ability to evaluate and process the data slice generated by the coordinating control system, select the composition of control actions and settings based on the data slice, form an output tag in the database for each control action corresponding to the number of the power system node to which this control action, and each decentralized USUNRM is performed with the ability to calculate the control actions and settings of local regulators of regime automation for optimal control of the voltage and reactive power of the power system nodes in real time by transmitting requests and control actions to the power system nodes, as well as autonomous operation in the event of the disappearance of all communication channels with the CCS.

To ensure the uninterrupted operation of the power system, all communication channels for data transmission can be duplicated.

The CCS can also include a simulation subsystem (IP), including server hardware and designed to simulate a real dynamic power system to form a preliminary database and work out the general interaction of the nodes of the intelligent voltage and reactive power control system of the power system. The CCS is intended to simulate a real dynamic power system, to replace the OIC, to form a preliminary database and work out the general interaction of the elements of intelligent control of voltage and reactive power of the power system.

The MSC and/or USUNRM preferably include a workstation configured to visualize and control the workflow by the user.

Brief description of the drawings

Further, the invention is explained in detail on the example of its implementation with reference to the attached figures.

In FIG. 1 shows a general block diagram of the voltage and reactive power control process of the power system.

201 - collection of data on the current state of the system;

202 - forming a slice of data in the node/branch model;

203 - state estimation;

204 - mode prediction;

205 - formation of restrictions;

206 - dynamic optimization taking into account restrictions;

207 - control of optimization results;

208 - issuance of control actions.

In FIG. 2 shows a block diagram of one cycle of operation of the subsystem for optimal control of voltage and reactive power of the power system.

301 - collection of data on the current state of the system;

302 - logical control of input data;

303 - forming a slice of data in the node/branch model;

304 - static state evaluation;

305 - dynamic state estimation;

306 compares the objective function (FIG. 3) with the critical value;

307 - preference for dynamic state estimation;

308 - generalization of the results of the state evaluation;

309 - forecasting according to load schedules;

310 - forecasting using machine learning;

311 - generalized forecast;

312 - calculation of reserves for regime restrictions and stability;

313 - mode classification;

314 - choice of restrictions;

315 - dynamic optimization subject to constraints;

316 - check if emergency restrictions are violated;

317 - check whether warning restrictions are violated;

318 - emergency messages;

319 - warning messages;

320 - issuance of control actions.

In FIG. 3 shows an example of the objective function ƒ _i (X _i ).

Implementation of the invention

The generalized structure of the power system voltage and reactive power intelligent control method is presented as a flowchart of the power system voltage and reactive power control process in FIG. 1. The method includes the use of a centralized coordinating system (CCS), consisting of collecting data on the current state of the system 201 (Fig. 1), including the formation of a data slice in the node/branch model 202 (Fig. 1), state estimation 203 (Fig. 1 ), mode prediction 204 (FIG. 1), constraint generation 205 (FIG. 1), data processing from power system nodes using dynamic constraint optimization 206 (FIG. 1), optimization results monitoring 207 (FIG. 1), and also the issuance of control actions 208 (Fig. 1), which generates and transmits control actions to the nodes of the power system, containing a coordinating control system connected by communication channels with a database and an optimal control subsystem (POA) (Fig. 2) voltage and reactive power of the power system. The method also includes the use of at least one decentralized nodal voltage and reactive power control system (USUNRM) connected to the MSC by communication channels for data transmission. The method includes server equipment configured to receive information about the state of the power system, form a data slice on it, transfer the data slice to the POU, obtain a list of control actions and settings of local regulators of regime automation from the POU and transfer control actions and settings to the power system in accordance with with the required time of their application. The PO also includes server hardware configured to evaluate and process the data slice generated by the coordinating control system, selection based on the collection of data on the current state of the system 301 (Fig. 2) using logical control of input data 302 (Fig. 2) of the composition of control actions and settings, the formation of a data slice in the node/branch model 303 (Fig. 2), that is, the formation in the database for each control action of an output tag corresponding to the number of the power system node to which this control action is directed. Based on the data slice, the POA performs both static estimation of the power system state 304 (FIG. 2) and dynamic estimation of the power system state 305 (FIG. 2), then compares the objective function (FIG. 3) with the critical value 306 (FIG. 2) and, if the objective function (Fig. 3) is less than the critical value, then preference is given to dynamic state estimation 307 (Fig. 2), if the objective function (Fig. 3) is greater than the critical value, then generalization of the results of state estimation 308 (Fig. 2) is used. ), and then perform both load curve prediction 309 (FIG. 2) and machine learning prediction 310 (FIG. 2), then generalized forecast 311 (FIG. 2) is obtained, and margins are calculated by mode constraints. and stability 312 (FIG. 2), mode classification 313 (FIG. 2), and constraint selection 314 (FIG. 2). Based on the results obtained, dynamic optimization is performed taking into account restrictions 315 (Fig. 2) and checking for violation of emergency restrictions 316 (Fig. 2), if emergency restrictions are not violated, check whether warning restrictions are violated 317 (Fig. 2), if emergency restrictions are violated, then the POA issues alarm messages 318 (Fig. 2), and if the warning restrictions are violated, then the POA issues warning messages 319 (Fig. 2), then based on the warning messages 319 (Fig. 2) or the absence of violations of the warning restrictions, the POA generates the issuance control actions 320 (FIG. 2).

Each decentralized USUNRM includes server hardware and is configured to calculate the control actions and settings of local regulators of regime automation for optimal control of the voltage and reactive power of power system nodes in real time by transmitting requests and control actions to the power system nodes, as well as autonomous operation in the event of the disappearance of all channels connection with the CCC. The DCS can also include a simulation subsystem (IP), including server hardware and designed to simulate a real dynamic power system to form a preliminary database and work out the general interaction of the nodes of the intelligent control system for voltage and reactive power of the power system.

The MSC and/or USUNRM preferably include a workstation configured to visualize and control the workflow by the user.

Data transmission is carried out via any communication channels known to a person skilled in the art, for example, Ethernet communication channels. To ensure the smooth operation of the system, all communication channels for data transmission can be duplicated.

The optimal control subsystem (OCS) of the voltage and reactive power of the power system is designed to calculate the control actions and settings of local regulators of regime automation for optimal control of the voltage and reactive power of the power system in real time based on telemetry data and cyclic solution of optimization problems, taking into account the specified constraints and optimal criteria ( admissible) control.

The subsystem for optimal control of voltage and reactive power of the power system includes:

• Integration platform - data collection and transmission system (DSTS), providing:

- obtaining telemetry data, telesignals (TI, TS) about the current state of the power system according to the standard protocol IEC 60870-5-104;

- execution of optimal control algorithms;

- transfer of control actions to a centralized coordinating system of regime automation, providing optimal control in accordance with a given objective function;

- an archival database for storing: archives of data on the state of the power system, operating data of the subsystem for optimal control of voltage and reactive power.

- a message subsystem that provides the issuance of messages to the user and logging of diagnostic data on the operation of the system;

• Basic settlement platform;

• Data preparation system for calculation of regimes;

• Estimation of the state (calculation of modes according to telemetry data);

• Calculation of steady modes;

• Optimization and entry into the allowable area;

• Machine learning system;

• Display system.

Optimal control algorithms consist in the cyclic execution of the following tasks:

• Collection of data on the current state of the system 201 (Fig. 1) and 301 (Fig. 2), such as:

- telemetry;

- states of switching devices;

- operational status of local control devices;

- current values of settings of local control devices (voltage settings, positions of energized regulators (OLTC), values of inductances of controlled shunt reactors (CSR));

• State Estimation 203 (FIG. 1);

• Static state evaluation 304 (FIG. 2);

• Dynamic state estimation 305 (FIG. 2) based on a modification of the Kalman filter;

• Mode Prediction 204 (FIG. 1);

• Formation of constraints 205 (FIG. 1);

• Dynamic optimization of the mode, taking into account the restrictions 206 (Fig. 1) - minimization of active losses in the electrical network, taking into account:

- regime restrictions on voltage and reactive power;

- restrictions on the margins of static stability obtained at the previous steps of the iterative process;

- "costs" of control actions;

• Formation of control actions based on optimization results.

The output information about the control actions contains the required time for their processing, as well as the time of their action, which follows from the forecasting depth.

Data slicing in node/branch model 202 (FIG. 1) and 303 (FIG. 2) is a process of preparing measurements suitable for mode modeling and optimization and consists of:

• Obtaining current teleinformation data and network status;

• Obtaining archival teleinformation data and network status (for the learning process and the initial launch of forecasting);

• Logical control of received data integrity;

• Formation of a slice in the node/branch model using a topological processor.

This requires binding telemetry to a single-line electrical network diagram.

A teleinformation slice is a set of measurements and TV signals about the state of the network, taken over a certain time. Time simultaneity is determined by the timestamps present in each analog or switch state measurement. The timestamp is usually transmitted from the final digital measuring device (digital converter). If the time stamp is not fixed on the final measuring device, the time stamp is assigned on the SSPD or on the local control device. The Operational Information Complex (OIC) must provide TI and TS data with time stamps of the end measurement devices.

The final slices of the TI and TS data are formed in the subsystem of the optimal voltage and reactive power control. At the same time, there is a measurement error in time in the TI and TS data slices. This is due to delays in the data transmission path of the TI and TS to the upper level, determined by the speed of digital devices for collecting and transmitting data, data transmission channels, and the sporadic nature of data transmission.

The TI and TS data slices also include additional information entered by the dispatcher through the OIC, such as the mode of operation of local automation devices.

Composition of TI and TS data slices:

• Telemetering of active and reactive (Р, Q) power on power transmission lines, including lines extending to the consumer (supplying the load).

• Voltage measurements on the buses of stations and substations.

• States of switching devices.

• Mode of operation of local control devices:

- manual regulation according to the stage number;

- manual regulation by the value of transverse conductivity;

- automatic voltage regulation;

- not at work.

• Planned time for the control device to be in a given mode of operation.

• Positions of devices of control devices of control devices (number of the current OLTC tap, number of the current control stage of compensating devices in case of step control).

• Set value for the transverse conductance of the compensating devices (for continuous regulation).

• The voltage setting value of the local regulation devices.

Obtaining current slices of TI and TS data

The receipt of the current data of TI and TS and the state of the equipment to the optimal control subsystem is carried out using the IEC 60870-5-104 telecommunication protocol. In this case, the SSPD acts as a data server (Slave), and the optimal control subsystem acts as a client (Master).

The formation and control of the completeness of the cut of the data of TI and TS is carried out on the side of the subsystem of the optimal control of voltage and reactive power.

Obtaining archival slices of TI and TS data

Obtaining a slice of TI and TS data and the state of switching devices is carried out through an SQL query to the archive database of the SSPD.

Archival data slices are formed on the side of the SSPD. The composition of the slices of the original data is similar to the composition of the data received on-line, except that the data is supplemented with a slice timestamp. The slice timestamp is assigned the same for all slice parameters and may differ from the timestamps received from the devices. So, for example, if a slice is formed in time t, but some data did not change (for example, the states of switching devices) during time θ, then the timestamp received from the end device for this data will be t-θ. The slice for this data must have a slice timestamp t. In this case, the device timestamp must also be present in the slice as a separate field.

The resulting data slice is checked for compliance with the following conditions:

- If both the power flow and the state of the switching device are measured at the connection, then their compliance is checked.

- If there are measurements of power flows at all connections outgoing from the busbars, then the sum of all flows should not exceed the allowable unbalance specified by the setting.

- Sequential slicing controls data variability. Between two consecutive slices within the same object, there must be at least one change in dimensions or states.

- The presence of signs of unreliability in the source data.

In order to update the initial data for assessing the state, it is necessary to compare the data of TI and TS with the calculation model of the SOA. Such a comparison is provided by "binding" the data of TI and TS to the calculation model (scheme).

The binding is carried out from the graphical scheme presented in the POU, and is stored accordingly in the graphical scheme. This approach allows not to increase the size of the calculation model to switching circuits. Thus, as a result of “binding”, an element of the graphic scheme is associated with one TS and several (according to the number of types) TIs. For example, TC of the circuit breaker state, flow P through the circuit breaker, flow Q through the circuit breaker. As a TS and TI, a formula specified in the POU can act. Binding is carried out by a unique data identifier of TI and TS.

The transformation of TI and TS to the level of the model is carried out by the topological processor.

The node/branches name of the power system model is the equivalent circuit of the electric power system or network, which is equivalent to a given electrical circuit and adequately reflects the processes taking place in it. In the equivalent circuit, real network elements (physical devices) are replaced by idealized elements (resistances, capacitances, inductances, ideal transformers, driving currents and powers).

Estimating the state 203 (FIG. 1) is to obtain such a steady state, which would be closest to the available measurements. The result of the state estimation task is:

- Calculation of all parameters of the current mode (active and reactive power of the load and generation, modules and phases of nodal voltages, active and reactive power flows) according to the TI and TS data on the position of the switching equipment;

- Calculation of retrospective modes based on the archive of telemetry in SSPD;

- Validation of all parameters used in the calculation modes;

- Preparation of the initial model for other tasks of the complex (steady state, optimization, etc.).

The solution of the problem of state estimation 203 (Fig. 1) is carried out by two methods: static and dynamic. The problem of static state estimation is solved independently for each moment of time and makes it possible to obtain a balanced steady state for each measurement slice. The task of dynamic state estimation updates its state when moving from one slice to another, thereby taking into account the change in the EPS operation mode in time. These two state estimation methods complement each other, mutually verifying the estimation results.

If the results of dynamic state estimation are better than the results of static state estimation, then short-term forecasting of the regime is carried out by methods of dynamic state estimation.

The results of short-term mode prediction are used to estimate the state at the next time steps. This allows state evaluations to be performed in the event of missing or delayed acquisition of part of the measurements. Predictive measurements are used with much greater variance than measured ones. The magnitude of the change in the variance of the predicted measurements is adjusted during commissioning of the system.

The static state estimation 304 (FIG. 2) uses a weighted least squares method. Today, this approach is the classical formulation of the state estimation problem.

In the static state estimation (OS) block, there is a function of preliminary rejection of erroneous measurements. The user can perform the calculation with or without such a function. The program rejects only voltages in nodes and overflows in branches, while loads and generations are not affected by the rejection algorithm.

The OS program automatically compiles groups of TPs, in which, with greater or lesser accuracy, it is possible to check the correctness of the TP balance based on Kirchhoff's laws or to link the TPs based on the laws of loop stresses. For example, it is possible to check the coincidence of the algebraic sum of measurements of power flows in all branches suitable for the node and the measured load of the node; checking the coincidence of measurements of power flows at the beginning and at the end of the branch with an approximate allowance for losses and capacitive generation of the branch; checking the coincidence of measurements of flows in parallel branches, etc. This approach is known as the control equation method.

Further, depending on the given accuracy of the TI, on the magnitude of the imbalance and on the given imbalance tolerance threshold, all TIs included in this balance group can be recognized as reliable or doubtful (and intermediate gradations between reliability and doubtfulness can also be introduced). In case of doubt, all measurements included in such a group will be used in calculations with reduced accuracy or completely excluded from calculations.

However, the determination of the reliability or doubtfulness of TI in this way cannot be fully guaranteed. For example, two erroneous TIs with approximately equal in magnitude but opposite in sign error can fall into the balance group, which will ensure that the balance is maintained and erroneous TIs will be recognized as reliable. Or, due to switching errors, the balance in the TI group can be grossly disturbed and all TIs of this group will be recognized as unreliable.

It should also be noted that with the currently existing volumes of TI in power systems, in most cases, it is possible to determine only the fact of the doubtfulness of the TI group. And there is very little opportunity to identify specifically that erroneous TI, because of which the rest of this group and possibly correct TIs will be identified as dubious. As a result, a large number of sufficiently accurate TIs, several times greater than the number of erroneous TIs, will have to be recognized as unreliable and excluded from the calculation.

Next, an attempt is made to “reinstate” at least a part of the falsely rejected, but sufficiently accurate, TIs, as well as to additionally check the TIs that were recognized as reliable. For this purpose, the calculation is carried out according to the main OS algorithm. Doubtful TIs that agree well enough with their calculated values are again recognized as reliable, and TIs that grossly disagree with their calculated values are recognized as erroneous. Further, taking into account the identified erroneous TIs, the algorithm for preliminary rejection of "bad" TIs again works and the calculation is again performed according to the main OS algorithm. And so several iterations are performed. As a result, the number of falsely rejected TIs is reduced and the reliability of rejecting really erroneous TIs is increased.

The static error rejection algorithm is a preliminary rejection algorithm. Further, it is supplemented by a dynamic algorithm for rejecting erroneous measurements, based on the results of dynamic state estimation.

As measurements, the OS program provides for setting voltages at network nodes, active and reactive power of loads and generations at nodes, active and reactive power of flows in circuit branches. Parameter values that do not come directly from measuring devices are usually referred to as pseudo-measurements. Each such measurement is specified by three parameters: the sign of the measurement, the measured value, and the accuracy.

Signs of measurements are set at the stage of commissioning of the system. In the future, they can be adjusted during its operation.

The measurement variance is set as the measurement accuracy. It should be taken into account that the measurement accuracy obtained from TM devices is determined by the total error of the entire measuring path.

For greater clarity, standard deviations can be given as accuracy. This will only affect the value of the objective function. At the same time, it is important that for all measurements one approach to specifying the measurement accuracy (either dispersion or standard deviation) is used.

The relative deviation θ _{i of} the calculated value of the parameter i from its measured value is calculated by the formula:

- измеренные параметры режима,

- measured mode parameters,

v(x) - calculated mode parameters from the values of the system state vector x,

σ _i - measurement variance.

Relative deviation is a dimensionless relative value. For loads and generations of nodes, θ _i with the "+" sign means the presence of excess power in the node in comparison with the measured or statistically calculated value, with the "-" sign - insufficient.

The practical utility of the relative deviation is that it characterizes the degree, the roughness of the violation of the measured value, regardless of the absolute value of this violation and the given accuracy, i.e. the value of the relative deviation is an invariant indicator with respect to the indicated values. For a group of connected nodes, for individual sections of the network diagram, θ _i shows, as a rule, a tendency for a group of nodes to excess or lack of power, due to measurements of power flows in the branches along the boundaries of this group of nodes and their total load and generation.

For greater information content and clarity, the magnitude of the tendency to excess or lack of power is also calculated for nodes in which there is no load or generation and is also displayed in the form of a relative deviation.

An important value in the analysis of state estimation is the total relative deviation, calculated as the sum of all measurements according to the following formula:

- измеренные параметры режима,

- measured mode parameters,

v(x) - calculated mode parameters from the values of the system state vector x,

σ _i - measurement variance.

Nodes that do not have a measurement do not participate in the calculation of the total deviation (although the relative deviation is calculated for them).

Also, rejected measurements, measurements of balancing nodes, measurements in disconnected lines do not participate in the calculation of the total deviation.

Dynamic State Estimation 305 (FIG. 2)

The problem of dynamic state estimation is solved using the Kalman filter. To predict weakly variable components of the system state vector x for a short period of time, a dynamic OS based on the modification of the Kalman filter is used. The lead time can be from several seconds (the interval between the acquisition of measurement data (slice) at time k and at time k+1) to 1 min.

The Kalman filter is a classic method for dynamic state estimation.

Identification of erroneous measurements in the process of dynamic state estimation is performed as follows. At the first stage, measurements are identified whose values have changed relative to the current estimate or the measured value at the previous time step by more than a certain threshold calculated in proportion to the measurement variance:

- измеренные параметры режима,

- measured mode parameters,

v(x) - calculated mode parameters from the values of the system state vector x,

d - threshold value of the mode parameter.

Among the strongly changed measurements, those that also deviate from the predicted value of this measurement are identified:

The block of dynamic state estimation does not require the input of additional initial data in excess of those specified for static state estimation. However, in order for the dynamic state estimation to start producing adequate results, it takes time to accumulate statistics. If there is an archive database, it is possible to accumulate statistics using retrospective information from it.

To control the operation of the state estimation algorithm, the issuance of custom messages is provided. For a generalized control over the operation of the state estimation unit, the values of the total relative error (RS) of the static and dynamic state estimation algorithms are used.

Mode prediction 204 (FIG. 1), load curve prediction 309 (FIG. 2), machine learning prediction 310 (FIG. 2), generalized prediction 311 (FIG. 2):

Mode prediction is performed by a combined algorithm using load curve prediction 309 (FIG. 2) and using machine learning 310 (FIG. 2) using artificial neural networks.

The input values of the prediction algorithm for each considered moment of time are the measured voltage values at the nodes of the electrical network, the active and reactive power flows in the branches, and the measurements of active and reactive power injections at the nodes (load and generation). Also, in the presence of vector measurements, measured stress angles may be present in the source data. The vector of input measurements can be sorted in accordance with the optimal order of nodes from the point of view of convergence, or be unordered. The archival database of initial data for forecasting the regime is filled in by the POU.

In addition to the archived measurement data converted into a node/branch model, the operation of the prediction algorithm requires information about the measurement variances. This information is prepared in the process of commissioning of the state evaluation unit.

To control the operation of the prediction algorithm, the following information and diagnostic messages are provided:

- Predictive mode (in the diagram and in tables).

- Values of relative deviations of measurements (on the scheme and in tables).

- Values of dispersions of measurements (on the scheme and in tables).

- The generalized indicator of the quality of the updated forecast (total root-mean-square deviation).

- In case of long-term deterioration of the quality of the updated forecast, an emergency message is issued.

- Formation of restrictions (205)

- Regime restrictions are the input data for the optimal voltage and reactive power control algorithm. The regime restrictions are:

- Restrictions on the level of voltage in the nodes of the electrical network (minimum and maximum);

- Restrictions on the injection of reactive power in the nodes of the electrical network, which are generators and compensating devices;

- Current limits in the branches of the node/branches model, corresponding to the current limits for high-voltage lines and transformers;

- Restrictions on power flow in the branches of the node / branch model, corresponding to the power restrictions of high-voltage lines and transformers.

Regime restrictions are set during the commissioning of the system, and are also adjusted during its operation by operational personnel.

It is possible to refine the restrictions in the direction of greater reliability by an automatic algorithm for analyzing reserves for static stability and regime restrictions by a weighting algorithm.

An important feature of regime restrictions is their compliance with the EPS operation mode (normal, repair, emergency, post-accident, forced, etc.).

Static stability is understood as the possibility of synchronous operation of synchronous electric generators and synchronous motors of EPS with slow load changes. The task of calculating static stability is estimated on the basis of cyclic calculations of steady-state modes (SD). The criterion for violation of static stability is the lack of convergence of the SD after the gradual weighting of the regime. To find the limiting mode by the weighting method, a weighting trajectory is specified, i.e. a list of nodes in which the generation and load are changed and the initial change step for each node. Since the main objective of the method is to obtain the optimal mode, the trajectory of weighting can be obtained automatically based on the difference between the optimal and initial mode. Moving further along this trajectory by the weighting method, one obtains the limiting regime, both in terms of static stability, and in terms of regime or technological restrictions.

The classification of the EPS operation mode can be performed using formal logical rules, both by analyzing single modes and by analyzing sequential modes using the finite automaton method. Formal rules for the classification of modes are set during the commissioning of the optimal control subsystem.

Greater versatility in classification is achieved by using machine learning methods such as decision trees and a more generalized version of decision trees using the random forest method. An important advantage of decision trees is that they not only refer a mode to one class or another, but also allow one to substantiate the mode classification numerically. A decision tree is a tree whose leaves correspond to the values of the objective function, and in the remaining nodes there are transition conditions that determine which of the edges of the node to pass the algorithm.

Dynamic optimization subject to constraints 206 (FIG. 1) and 315 (FIG. 2) is used to obtain optimal control actions for a given depth using a dynamic optimization algorithm.

The initial data for dynamic optimization are:

- Current rated mode.

- Forecast as a set of predicted modes.

- List of possible control actions (HC).

- HC states.

The predicted states of the system for the considered time range are the set:

X _i is the state vector of the system at each moment of time.

The proposed solution takes into account the cost of control actions, which depends not only on the state vector of the system, but also on time. As a result of optimization, the entire considered time range is divided into several subranges, each of which contains the optimal solution for this subrange. The complete objective function of the optimization problem can be written as:

- вектор оптимальных управляющих параметров на поддиапазоне;

- vector of optimal control parameters on the subrange;

b _p and e _p - indices of the beginning and end of the subrange in the time slice;

C is a function of the cost of impacts, depending on the cost of each specific impact, taking into account the time specified by the index i;

ξ ₀ is taken equal to the initial value of the control parameters X ₀ .

ƒ _i - objective function of each subproblem of optimization, including:

- loss of active power;

- deviations from the allowable voltage ranges;

- Violations of regime restrictions.

The cost of managing this or that equipment depends on factors such as:

- residual resource of the equipment;

- priority use of hydrocarbons;

- the minimum allowable time between switching by the same device.

This formulation of the dynamic optimization problem allows one to obtain the optimal value for the entire period under consideration. The search for the global optimum over the entire forecast range is carried out taking into account the cost of the HC, which is taken into account in the general objective function.

The objective function ƒ _i (X _i ) used in static optimization is assumed to be convex. A non-convex objective function in the problem of mode optimization can be replaced by its dual convex function.

The set of local optimization solutions minƒ _i (X _i ) limit from below the solutions of the general optimization problem with an objective function over the entire time range. On the example of a convex function in the context of one control parameter for two points of the time range: if x _d is the value of the control parameter for the optimal point of the entire presented range, then the value of the objective function at this point is obviously higher than the value of the objective function at the local optimum points (x _l , x ₂ ).

Using this property, the search areas for optimal solutions for the entire time range are determined and maximally limited. Each static optimization calculation defines the dimensions that affect the objective function:

X _ξ - the value of the control parameters;

x - values of the system state vector.

This area is a multidimensional cube with a number of dimensions greater by one than the number of parameters that have changed in the process of optimizing single points in time and limited by the limits in which these parameters have changed. In practice, such a region turns out to be small, since for regimes close in time, as a rule, close optimal control actions are obtained.

An annealing simulation algorithm is used to search for global optima and their number over the entire time range. This is all the more justified for the case of discrete control actions.

The initial approximation for the annealing simulation algorithm is the solutions of the local optimization problem included in the solution search area. The annealing simulation algorithm solves the following optimization problem:

ξ _d - vector of optimal control parameters on the subrange;

b _p and e _p - indices of the beginning and end of the subband in the time slice;

C - impact cost function depending on the cost of each specific impact, taking into account the time specified by the index i;

ƒ _i - objective function of each subproblem of optimization, including:

- loss of active power;

- deviations from the allowable voltage ranges;

- Violations of regime restrictions.

To find the optimal value of continuous control variables, such as active and reactive power injections in the nodes of the electrical network graph, it is proposed to use a hybrid annealing simulation method for continuous variables and the gradient descent method.

Static optimization

The problem of static mode optimization, which is solved for each mode in the process of dynamic optimization, can be written in general form as follows:

the objective function ƒ(⋅) is formulated as:

optimization control parameters u _g , k _t , y _c - voltages in reactive power balancing nodes, transformation ratios and conductivities of compensators, respectively;

B - many branches;

N - set of nodes.

The priority of accounting and scaling of the parameters of the objective function are given by constants: c _ΔP - constant scaling accounting for losses, c _U - constant scaling accounting for voltage deviations in the nodes.

Restrictions on control parameters are simple boundaries of acceptable values:

u _g , k _t , g _c - optimization control parameters;

G ∪ Cu - a set of generator nodes involved in optimization and compensators that regulate voltage;

T is the set of transformer branches involved in the optimization;

C - a set of nodes with compensators that regulate the voltage.

The dependent parameters Δp (losses in the branches) and Δu (voltage deviations in the nodes) are calculated from the electrical network equations:

s - vector of independent parameters: powers P, Q in load nodes; power P and voltage U in reactive power balancing nodes; voltage U and phase d in nodes balancing simultaneously in P and Q.

U and u - matrix and vector of components of stress complexes;

Y and y _b - matrix and vector of conductance complexes;

u _b - voltage of the basic node.

As a method for finding a local optimum, a method based on the L-BFGS-B algorithm is used. This method is a quasi-Newtonian method that takes into account restrictions on control parameters.

Restrictions on control parameters are taken into account in the form of boundary conditions:

x - values of the system state vector;

l and u are control parameter constraints.

Accounting for restrictions on the dependent mode parameters is performed in the form of barrier functions, the values of which will be included in the overall objective optimization function and are calculated in the process of calculating the flow distribution at each optimization iteration for fixed values of the control parameters. Logarithmic functions are used as barrier functions by analogy with the interior point method:

F(x) - barrier function;

g _i (x)≤ 0, i=1, …, m;

m - the number of restrictions on the dependent mode parameters.

This approach allows us to divide the optimization problem into the calculation of the flow distribution and the actual optimization problem, which, on the one hand, simplifies the optimization algorithm, on the other hand, complex models of network elements and models of automatic control systems can be taken into account within the procedure for calculating the steady state.

As a result, the algorithm for finding a local optimum using the L-BFGS-B method is as follows:

1. At each step of the iterative process, we have x _k - the control vector, ƒ(x _k ) - the value of the objective function, g _k - the gradient of the objective function at the point x _k .

2. To take into account restrictions of the form l≤x≤u, the gradient projection method is used.

3. The quadratic model of the function ƒ is written as follows:

B _k - approximation of the Hessian matrix, obtained by the quasi-Newtonian algorithm L-BFGS.

Next, a piecewise linear function is determined that coincides in direction with the antigradient and takes into account the constraints:

x - values of the system state vector;

l and u are control parameter constraints.

After that, the local minimum (Cauchy point) of the model function is calculated along the chosen piecewise linear direction:

m _k (x) - quadratic model of the function ƒ;

The iterative process continues until the change in the control vector along the optimization direction is greater than the allowable calculation error:

x - values of the system state vector;

ε - allowable calculation error.

When moving along the optimization trajectory, situations are possible when the mode does not converge, including due to limitations taken into account when calculating the flow distribution. In this case, it is necessary to find the limiting regime along the given trajectory, which ensures the convergence of the steady-state regime and fix new restrictions.

The movement along the trajectory is carried out by the method of dividing in half the increments of the vector of control actions. The quadratic model in this case is written as:

B _k - approximation of the Hessian matrix, obtained by the quasi-Newtonian algorithm L-BFGS;

α - кратность деления шага.

α - step division multiplicity.

To monitor optimization results 207 (FIG. 1), check for violation of alarm limits 316 (FIG. 2), check for violation of warning limits 317 (FIG. 2), alarm messages 318 (FIG. 2), warning messages 319 (FIG. 2) The operation of the optimization algorithm provides for the output of the following information:

• Summary of final optimization mode:

- The value of the objective function before and after optimization;

- The amount of losses before and after optimization;

- Violations of stress limits before and after optimization.

• Summary of the result of static optimization for each predictive mode:

- The value of the objective function before and after optimization;

- The amount of losses before and after optimization;

- Violations of stress limits before and after optimization.

• List of control actions as a result of static optimization for each mode.

• List of control actions as a result of dynamic optimization with the time of their application.

The results of the system for issuing control actions 208 (Fig. 1) and 320 (Fig. 2)

The system of optimal voltage and reactive power control issues SSPD tasks containing:

• Voltage limits (maximum and minimum) for each local regulator:

- Each limitation can contain several voltage measurement points, corresponding to network points where the voltage is controlled by the local regulator.

- For each restriction, the time of its validity is set.

- Constraints are hierarchical. If the limitation expires, then the higher-level constraint takes effect. At the same time, it should be controlled that the restrictions are not contradictory, both in terms of values and in time. At the topmost level of the boundary hierarchy (the root element of the hierarchy) are the base boundaries with infinite duration.

• Control task for each local controller with the required time of its completion. Tasks can be of the following types:

- The value of the required voltage at the network points where the voltage is controlled by the local regulator.

- No. of tap tap and corresponding transformation ratio.

- The required value of CSR inductance.

- To control the operation of the system for optimal control of voltage and reactive power on the part of the operator, the following information is displayed on the workstation in the online mode.

• Forecast results for a certain depth:

- Estimated forecast error;

- The greatest differences between the predicted mode and the current one

mode;

- Predictive mode (on the scheme and in tables).

• Optimization results:

- Optimal mode (in the diagram and in tables);

- Differences in the objective function in the initial and initial modes;

- Control actions corresponding to this optimal mode.

• Statistics on the remaining resource of control devices.

Claims (5)

1. A method for intelligent control of voltage and reactive power of a power system, including the installation of a centralized coordinating system (CCS) for collecting and processing data from power system nodes, as well as generating and transmitting control actions to power system nodes, containing a coordinating control system connected by communication channels with a database and an optimal control subsystem (POA) for the voltage and reactive power of the power system, and at least one decentralized nodal voltage and reactive power control system (USUNRM) connected to the MSC by communication channels for data transmission, while the coordinating control system is configured to receive information about the state power system, forming a slice of data on it, transmitting a slice of data to the POU, obtaining a list of control actions and settings of local regulators of regime automation from the POU and transferring control actions and settings to the power system in accordance with the required time of their application, the POU is performed with the ability to evaluate and process the slice data generated by the coordinating control system, selection based on the data cut of the composition of control actions and settings, formation in the database for each control action of the output tag corresponding to the number of the power system node to which this control action is directed, and each decentralized USUNRM is performed with the ability to calculate control influences and settings of local regulators of regime automation for optimal control of voltage and reactive power of power system nodes in real time by transmitting requests and control actions to power system nodes, as well as autonomous operation in the event of the disappearance of all communication channels with the CCS. 2. The method according to claim 1, characterized in that the communication channels for data transmission are duplicated. 3. The method according to claim 1, characterized in that the centralized coordinating system includes a simulation subsystem, including server equipment and configured to simulate a real dynamic power system to form a preliminary database and work out the general interaction of the nodes of the intelligent control system for voltage and reactive power of the power system. 4. The method according to claim 1, characterized in that the centralized coordinating system includes a workstation, configured to visualize and control the workflow by the user. 5. The method according to claim 1, characterized in that the nodal voltage and reactive power control system includes an automated workstation, made with the ability to visualize and control the workflow by the user.

Images