CN117277307A - Data-driven-based active frequency control method and system for power system - Google Patents
Data-driven-based active frequency control method and system for power system Download PDFInfo
- Publication number
- CN117277307A CN117277307A CN202311280913.9A CN202311280913A CN117277307A CN 117277307 A CN117277307 A CN 117277307A CN 202311280913 A CN202311280913 A CN 202311280913A CN 117277307 A CN117277307 A CN 117277307A
- Authority
- CN
- China
- Prior art keywords
- frequency
- data
- power system
- frequency modulation
- new energy
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 125
- 238000004458 analytical method Methods 0.000 claims abstract description 54
- 230000004044 response Effects 0.000 claims description 59
- 238000013016 damping Methods 0.000 claims description 47
- 230000008569 process Effects 0.000 claims description 44
- 238000011156 evaluation Methods 0.000 claims description 41
- 230000002776 aggregation Effects 0.000 claims description 37
- 238000004220 aggregation Methods 0.000 claims description 37
- 238000011217 control strategy Methods 0.000 claims description 36
- 238000010248 power generation Methods 0.000 claims description 33
- 238000004088 simulation Methods 0.000 claims description 31
- 230000008859 change Effects 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 27
- 238000005516 engineering process Methods 0.000 claims description 24
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000004146 energy storage Methods 0.000 claims description 22
- 238000005457 optimization Methods 0.000 claims description 22
- 238000007781 pre-processing Methods 0.000 claims description 18
- 230000033228 biological regulation Effects 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000012512 characterization method Methods 0.000 claims description 11
- 230000007246 mechanism Effects 0.000 claims description 11
- 230000002787 reinforcement Effects 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 8
- 238000013210 evaluation model Methods 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 7
- 238000000513 principal component analysis Methods 0.000 claims description 6
- 238000003066 decision tree Methods 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 5
- 238000011084 recovery Methods 0.000 claims description 5
- 238000007619 statistical method Methods 0.000 claims description 5
- 238000007635 classification algorithm Methods 0.000 claims description 4
- 238000007621 cluster analysis Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 230000035945 sensitivity Effects 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000012847 principal component analysis method Methods 0.000 claims description 3
- 230000009469 supplementation Effects 0.000 claims description 3
- 230000005619 thermoelectricity Effects 0.000 claims 1
- 238000009826 distribution Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 230000001276 controlling effect Effects 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 230000001105 regulatory effect Effects 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005265 energy consumption Methods 0.000 description 3
- 238000013178 mathematical model Methods 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 208000012639 Balance disease Diseases 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004138 cluster model Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003137 locomotive effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000003211 malignant effect Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 239000002344 surface layer Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000010792 warming Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/002—Flicker reduction, e.g. compensation of flicker introduced by non-linear load
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/24—Arrangements for preventing or reducing oscillations of power in networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/24—Arrangements for preventing or reducing oscillations of power in networks
- H02J3/241—The oscillation concerning frequency
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Physics & Mathematics (AREA)
- Nonlinear Science (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The scheme provides a method and a system for controlling active frequency of a power system based on data driving. The method comprises the steps of collecting multi-source data of a power system to form a full sample data set, carrying out frequency modulation characteristic analysis of the power system based on the full sample data set, identifying frequency modulation characteristic parameters of the power system, carrying out frequency modulation capability assessment and situation prediction on the power system based on the frequency modulation characteristic parameters of the power system, and finally generating a frequency modulation strategy according to assessment results and prediction results to realize active frequency control of the power system. According to the scheme, based on a data driving method, the system source, load, active-frequency equipment level and system level characteristics are fully identified, the real-time online situation sensing and predicting capacity of the power grid is improved, the frequency characteristic sensing capacity of the high-duty power grid of the new energy power source and the frequency characteristic sensing capacity of the AC/DC interconnection receiving end power grid can be effectively improved, the active control capacity and frequency safety of the power system are guaranteed, and the safe and stable operation of the power grid is guaranteed.
Description
Technical Field
The invention relates to the technical field of active frequency control of an electric power system, in particular to an active frequency control method and an active frequency control system of the electric power system based on data driving.
Background
A complicated large power grid is an alternating current system for connecting a plurality of power generation parties and power utilization parties, and is generally formed by a plurality of power transmission lines and substations in a geographically dispersed state. The complex large power grid is a power transmission and distribution system in a national or regional level generally, and has the characteristics of huge scale, wide area and field involved, profound influence and the like.
In recent years, the strong development of renewable green energy sources and the enhancement of wide-area power transmission capability are the main trend of the future development of power systems in China, and the characteristics of the power grid 'high-proportion renewable energy sources and high-proportion power electronic interfaces' in the future are more obvious.
Due to the improvement of the duty ratio of new energy sources such as wind power, photovoltaic and the like and the development of direct-current ultra-high voltage technology, the duty ratio of a conventional synchronous generator is gradually reduced, the inertia of the system is reduced along with the reduction, the damping characteristic is changed, and the active balance capacity of the system is weakened. On the other hand, the energy utilization rules of the load side are changeable and the interface characteristics are diversified due to the occurrence of energy utilization terminals such as electric automobiles and flexible loads, and the system balance difficulty is increased due to the combined action of the energy utilization terminals, the randomness, the intermittence and the fluctuation of new energy, so that real-time frequency control becomes one of the difficulties in future system operation. Multiple frequency events occur at home and abroad, and the difficulty and importance of the real-time frequency control technology of the system are highlighted. Under the background of increased uncertainty and diversity interleaving of the source network load, the risk of uncontrollable active frequency of a large power grid is more and more remarkable, and how to accurately identify and safely control the active and frequency characteristics of the system fully plays the regulating potential of the resources of each party, so that the method is the core work of planning operation of a future power system.
In practical application, the inventor of the application finds that the multi-frequency process of the power system is crossed and the multi-interference factors are interwoven, so that the frequency characteristic analysis and the parameter calculation are difficult. In the actual operation of the system, active fluctuation exists all the time, and the AGC frequency modulation instruction issued by scheduling also changes at all times, so that a plurality of processes of inertial response, primary frequency modulation, secondary frequency modulation and even tertiary frequency modulation of the hydroelectric generating set and the thermal generating set coexist under the same space-time measurement, the dynamic change characteristics of the system are influenced by a plurality of characteristics together, and the difficulty is high in how to separate the dynamic process corresponding to each part of frequency modulation characteristics from actual data; secondly, the overall inertia, damping and active-frequency change proportional relation of the system are affected by the controllable power supply, new energy and load, an electromechanical transient process model of the synchronous unit is mature at present, the new energy is difficult to form a mechanism model due to diversified control strategies, and the load model is built by means of a method with weaker characterization capability such as regression for a long time; and secondly, the new energy power supply and the load are greatly influenced by external factors such as meteorological conditions, energy consumption rules and the like, and the frequency characteristic of the new energy power supply and the load is high in nonlinearity degree and high in uncertainty. Therefore, the frequency dynamics of the system has the characteristics of multi-process interleaving, multi-factor action and high unknown degree of the underlying mechanism, and is difficult to analyze and predict by using a model or simple and direct parameter regression. How to implement frequency characteristic analysis at the system level and the power station level based on actual operation data is one of the difficulties.
Disclosure of Invention
The invention provides a data-driven-based power system active frequency control method and a data-driven-based power system active frequency control system, which solve the problems, fully identify system source, load, active-frequency equipment level and system level characteristics and improve the real-time online situation sensing and predicting capacity of a power grid.
In a first aspect, the present invention provides a method for controlling an active frequency of a power system based on data driving, including:
acquiring global historical data of a power system, and forming a full sample data set based on the global historical data, wherein the global historical data comprises PMU operation data, AGC instruction data, meteorological data and user energy habit data of the power system at each historical moment of each node of the power system;
performing frequency modulation characteristic analysis based on the full-sample data set to generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system;
performing frequency modulation capability assessment and situation prediction on the power system based on the generated model;
and determining a frequency modulation strategy of the power system according to the evaluation result and the prediction result, and realizing active frequency control of the power system based on the frequency modulation strategy.
In a second aspect, the present invention provides a data-driven-based active frequency control system for an electric power system, comprising:
the data set forming unit is used for acquiring global historical data of the power system and forming a full sample data set based on the global historical data, wherein the global historical data comprises PMU operation data, AGC instruction data, meteorological data and user energy habit data of the power system at each historical moment of each node of the power system;
the frequency modulation characteristic analysis unit is used for carrying out frequency modulation characteristic analysis based on the full-sample data set and generating an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system;
the frequency modulation capability assessment and situation prediction unit is used for carrying out frequency modulation capability assessment and situation prediction on the power system based on the generated model;
and the frequency modulation strategy determining unit is used for determining the frequency modulation strategy of the power system according to the evaluation result and the prediction result and realizing the active frequency control of the power system based on the frequency modulation strategy.
In a third aspect, the present invention provides an electrical power system comprising: the system comprises a controllable power supply, a new energy power supply and a converter load, wherein the controllable power supply comprises a hydroelectric generating set power supply and/or a thermal power generating set power supply;
the power system also comprises an AGC control system and an active power control system;
the active frequency control system executes the active frequency control method of the power system based on data driving according to any one of the above steps, and outputs an AGC control strategy;
the AGC control system generates an AGC control instruction based on the AGC control strategy to control the power system to perform active frequency adjustment.
From the above, the invention is applied to a power system, and provides a data-driven power system active frequency control method and system, which are based on the data-driven method, so that the characteristics of a system source, a load, an active-frequency equipment level and a system level are fully identified, the real-time on-line situation sensing and predicting capacity of a power grid is improved, the frequency characteristic sensing capacity of a high-duty power grid and an alternating current-direct current interconnection receiving end power grid of a new energy power source can be effectively improved, the active control capacity and the frequency safety of the power system are ensured, and the safe and stable operation of the power grid is ensured.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of an implementation of a method for controlling an active frequency of a data-driven-based power system according to an embodiment of the present invention;
FIG. 2 is a flow chart of one implementation of step 102 of the method of FIG. 1 provided by an embodiment of the present invention;
FIG. 3 is a flow chart of one implementation of step 103 of the method of FIG. 1 provided by an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an active frequency control system of a data-driven power system according to an embodiment of the present invention;
fig. 5 is a schematic diagram of an electric power system according to an embodiment of the present invention.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
It should be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It should also be understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
As used in the present description and the appended claims, the term "if" may be interpreted in context as "when …" or "upon" or "in response to a determination" or "in response to detection. Similarly, the phrase "if determined" or "if detected [ described condition or event ]) may be interpreted in the context to mean" upon determination "or" in response to determination "or" upon detection [ described condition or event ] "or" in response to detection [ described condition or event ].
Reference in the specification to "one embodiment" or "some embodiments" or the like means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the invention. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," and the like in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments" unless expressly specified otherwise. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.
It should be understood that, the sequence number of each step in this embodiment does not mean the execution sequence, and the execution sequence of each process should be determined by its function and internal logic, and should not limit the implementation process of the embodiment of the present invention in any way.
The data driving method can obtain the real-time frequency modulation parameters of the controllable power supply from the global angle of the system, thereby realizing the optimal control of the power system. The invention is applied to a power system, provides a data-driven power system active frequency control method and a data-driven power system active frequency control system, fully identifies the characteristics of a system source, a load, an active-frequency equipment level and a system level, improves the real-time on-line situation sensing and predicting capacity of a power grid, can effectively improve the frequency characteristic sensing capacity of a high-duty power grid of a new energy power supply and an AC/DC interconnected receiving end power grid, ensures the active control capacity and the frequency safety of the power system, and ensures the safe and stable operation of the power grid.
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the following description will be made by way of specific embodiments with reference to the accompanying drawings.
Referring to fig. 1, a flowchart of an implementation of a method for controlling an active frequency of a data-driven power system according to an embodiment of the present invention is shown, and the details are as follows:
step 101, global historical data of the power system is obtained, and a full sample data set is formed based on the global historical data.
In the embodiment of the invention, the global historical data refers to multi-source data in a historical period, and can comprise PMU operation data, AGC instruction data, meteorological data and energy consumption habit data of a user of the power system at each historical moment of each node of the power system.
On the one hand, through the sensor and the supervisory equipment of installing in the electric power system, can gather the operation data of electric power system in real time, including parameters such as voltage, electric current, power, frequency. On the other hand, some indirectly measured data, such as user energy usage habit data, can be obtained through statistical analysis of the directly collected data. In this embodiment, the PMU is referred to as Phasor Measurement Unit, meaning a synchrophasor measurement device. The PMU can collect information such as current, voltage and the like at a rate of thousands of hertz, obtain information such as power, phase, power angle and the like of a measuring point through calculation, and send the information to the WAMS master station at a frequency of tens of frames per second. The PMU can ensure the synchronism of the whole network data through GPS time synchronization, and the data and the time mark information are stored locally and transmitted to the master station. After the PMU is installed in the transformer substation and the power plant in the power grid, the dispatcher can monitor the dynamic process of the whole network in real time. In this embodiment, the AGC instruction data may include load frequency control data, economic dispatch control, spare capacity monitoring, AGC performance monitoring, link deviation control, and the like. In this embodiment, the weather data refers to weather information at a corresponding historical moment, for example, the illumination intensity directly affects the photovoltaic power generation capacity, and the user energy consumption habit data of the power system directly reflects the load characteristics of the power system.
In one class, the power system node class power system includes multiple types of nodes, which may be distinguished by their known and to-be-solved variables, such as the PQ node, which is commonly used to represent a substation bus, and certain generators of given output and reactive power, the generator at the PQ node being referred to as a PQ machine; PV nodes, commonly used to represent generators equipped with self-regulating excitation means and having an active power determined according to the optimal distribution principle, and nodes with constant active power and voltage levels; the balance node, which is typically used to represent the only balance point in the system, is known for both active and reactive power and has a voltage magnitude and phase angle of zero. In another classification, the power source side of the power system may be classified into a hydroelectric generating set, a thermal generating set, and a new energy source.
In the embodiment of the invention, the global historical data can be preprocessed first, then the whole sample data set is formed, and the preprocessing mode can comprise operations such as filtering, abnormal value removal, normalization and the like, so that the accuracy and the reliability of the data are improved.
And 102, performing frequency modulation characteristic analysis based on the whole sample data set to generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system.
In the embodiment, the purpose of frequency modulation characteristic analysis is to identify system inertia and damping, identify and cluster model typical conventional controllable power supply (hydroelectric and thermal power) primary frequency modulation main parameters; identifying a frequency response principle and a typical strategy for providing frequency support by converter type grid-connected equipment such as wind power, photovoltaic and the like, and establishing an aggregation model of large-scale wind power and photovoltaic power sources in frequency adjustment, namely a new energy frequency response strategy aggregation model; and identifying active-frequency characteristic parameters of the converter load in the extra-high voltage direct current near region, and generating a load frequency characteristic identification model.
Aiming at the problems that in the actual running process of a power system, PMU and other measurement data have more noise, a plurality of frequency response processes are mutually interweaved and influenced, and the like, the invention establishes a unified framework for frequency modulation characteristic analysis, adopts a mode of realizing accurate identification of sample set parameters by a white box model, and then realizing correlation from mass factors to frequency characteristics by a black box model, and carries out identification analysis under the same analysis paradigm.
In one embodiment, the process of performing the frequency modulation characteristic analysis in the step 102 to generate the inertial and damping parameter identification model of the power system is shown in fig. 2, and may specifically include:
Step 1021, obtaining a typical moment point when the power system generates active disturbance;
step 1022, sample data of the typical time point is extracted from the whole sample data set to form a typical sample set;
step 1023, carrying out noise attenuation and data preprocessing on the typical sample set;
step 1024, calculating system inertia and damping parameters of a typical moment point by using a parameter estimation method based on a frequency dynamic process mechanism model of the power system;
step 1025, expanding the calculated system inertia and damping parameters into a full-sample data set, and carrying out data enhancement by utilizing a time domain simulation or data generation technology;
in this embodiment, the generalization degree of the post-learning model can be improved by performing data enhancement through a time domain simulation or a data generation technology.
1026, performing feature learning based on the data-enhanced full-sample data set to obtain an inertial damping influence factor set of the power system;
step 1027, training the LSTM neural network based on the inertial damping influence factor set to obtain an inertial and damping parameter identification model of the power system.
In the embodiment, the system inertia and damping are identified by adopting a mode of establishing samples, parameter identification and feature learning, global historical data comprise external environment information such as weather temperature and the like, data of special moments such as active disturbance and the like of the system can be selected from the global historical data to be extracted to form a typical sample set, and noise attenuation and data preprocessing are carried out by adopting methods such as Kalman filtering and the like; based on a system frequency dynamic process mechanism model, solving system inertia and damping coefficients of typical time points by adopting a least square and other parameter estimation method, and completing parameter identification of a white box model; expanding the obtained system inertia and damping parameters to a full sample set, and realizing data enhancement by utilizing a time domain simulation or data generation technology, thereby improving the generalization degree of a later learning model; further, a main influence factor set of system inertia and damping is obtained based on a feature learning theory, and an LSTM neural network is utilized to solve the influence factor set to obtain an inertia and damping parameter identification model of the association relation of the feedback system inertia and damping.
In the present embodiment, the mechanism model analysis refers to a method of analyzing a certain mechanism or phenomenon by establishing a physical model. In mechanism model analysis, it is often necessary to abstract and simplify the real problem, find out major factors, ignore minor factors, and build a mathematical model describing the problem. This mathematical model may be a differential equation, a graph theory equation, etc., the specific form depending on the nature of the problem and the mathematical tool used.
In one embodiment, the process of performing the fm characteristic analysis in the step 102 to generate the primary fm parameter identification model of the power system may specifically include:
extracting PMU data of a controllable power supply high-voltage bus from the full-sample data set;
frequency analysis is carried out on the extracted PMU data to obtain a frequency dynamic curve of a primary frequency modulation process;
the active-frequency sensitivity relation of each node of the power system and each time unit is obtained in a time difference mode, and a unit frequency modulation parameter set with space-time dependency characteristics is formed;
based on the unit frequency modulation parameter set and the full sample data set, performing association relation learning between the conventional unit frequency modulation characteristic and the physical operation characteristic and the external environment characteristic of the system by using a neural network method to obtain a primary frequency modulation parameter influence factor set;
And based on the primary frequency modulation parameter influence factor set, carrying out full-network on-line unit aggregation of the power system by utilizing a classification algorithm to obtain a primary frequency modulation parameter identification model.
In the embodiment, PMU data of a high-voltage bus of a controllable power supply such as hydropower and thermal power is separated from a system full sample set by adopting a parameter identification, feature selection and aggregation analysis mode, a frequency dynamic curve of a primary frequency modulation process is obtained by adopting a data preprocessing and frequency domain analysis method, and active and frequency sensitivity relations of each node and each time unit are obtained by utilizing a time difference mode, so that a unit frequency modulation parameter set with space-time dependency characteristics is formed; based on the obtained frequency modulation parameter set and the system full sample data set, the association relation between the frequency modulation characteristics of the conventional unit and the physical operation characteristics and the external environment characteristics of the system is learned by using a neural network and other methods, main influence factors are obtained, the full network online unit is aggregated by using a classification algorithm based on the obtained main influence factors of the frequency modulation characteristics of the unit, and the characteristic of the typical unit is effectively used for representing the multi-unit cluster through actual simulation verification.
In one embodiment, the process of generating the new energy frequency response policy aggregation model in step 102 may specifically include:
Extracting PMU data of new energy station grid-connected points from the full-sample data set, and obtaining active-frequency time sequence of station level or distributed new energy after data preprocessing;
decomposing the active-frequency time sequence to obtain a primary frequency modulation response process sequence, identifying frequency modulation parameters of the primary frequency modulation response process sequence, and establishing a station-level power electronic interface external characteristic model based on the identified frequency modulation parameters;
extracting a wide-area feature set from the full-sample data set, training a neural network model based on the wide-area feature set, and determining the association relation between the wide-area feature set, a new energy station level ontology control strategy and new energy frequency response characteristics;
and determining a new energy station frequency modulation characteristic influence factor set based on the association relation, and carrying out classification aggregation of the new energy station based on the new energy station frequency modulation characteristic influence factor set to obtain a new energy frequency response strategy aggregation model.
In the embodiment, a mode of mechanism analysis modeling, parameter identification and characteristic aggregation is adopted, PMU data of new energy station grid-connected points are firstly separated from a full sample set, and active and frequency time sequences of station levels or distributed new energy sources are obtained through data preprocessing; decomposing a primary frequency modulation response process sequence by adopting methods such as component analysis and the like, and identifying frequency modulation parameters by adopting a time difference method; analyzing the frequency response principle of each type of new energy active control strategy, establishing an external characteristic model according to the composition condition of a station-level power electronic interface, and verifying the effectiveness of the theoretical model on the external characteristic representation of the station; based on a system full sample data set, acquiring wide area characteristic sets such as meteorological resources, loads, hydrologic resources, external environment attribute data and the like, and learning association relations between a station body control strategy, the wide area characteristic sets and new energy frequency response characteristics by using methods such as a neural network and the like; the new energy station classification aggregation is carried out by taking various features as input in a decision tree mode and the like, and the multi-station and distributed new energy is effectively characterized by typical station characteristics through actual simulation verification, and specifically, the wide-area feature set further comprises: and collecting and counting data which directly influence the active frequency of the system, such as active disturbance data of the system, frequency change data of the system, active response data of a new energy station and the like.
In one embodiment, the process of generating the load frequency characteristic recognition model in step 102 may specifically include:
extracting PMU data of a converter load bus from the full-sample data set and carrying out data preprocessing;
the active-frequency micro-variable proportion relation of each space point and each moment point is obtained by using a long-time differential sequence, so that the identification of load characteristic parameters is realized;
performing feature learning and principal component analysis on the system electrical quantity features and the external environment features in the full-sample data set, removing redundant features, and obtaining a load frequency characteristic influence factor set;
based on the load frequency characteristic influence factor set, the association rule of the load frequency characteristic and the influence factor is characterized and learned by utilizing a neural network algorithm with time memory characteristics, and a load frequency characteristic identification model is obtained.
In the embodiment, a mode of sample generation, data regression and characteristic mining is adopted, and PMU data of converter load buses such as an electric vehicle charging station, a railway locomotive line parallel point and the like are firstly extracted and are subjected to data preprocessing. Considering that the power load has complex constitution and lacks a unified active-frequency rational mathematical model, the active-frequency micro-variable proportional relation of each space point and each moment point can be directly obtained by using long-time differential sequences and adopting the technologies such as regression, data estimation and the like, so that the load characteristic parameter identification is realized; removing redundant features from the electrical quantity features and the external environment features of the system in the system full-sample set by utilizing feature learning and principal component analysis to obtain a key influence factor set of the load frequency characteristics; and finally, performing characterization learning on the association rule of the load frequency characteristic and the influence factor by using a neural network algorithm with time memory characteristics to obtain a load frequency characteristic identification model.
And 103, carrying out frequency modulation capability assessment and situation prediction on the power system based on the generated model.
In the embodiment, the frequency modulation capacity of the power system is estimated based on the generated model, so that the AGC reserve capacity considering the rotational inertia and the primary frequency modulation and the secondary frequency modulation of the system can be estimated in real time; carrying out wide-area large-scale new energy frequency modulation available capacity modeling based on wind/light resource information; and the active frequency security situation prediction of the large power grid considering the real-time adjustment capability of the new energy power generation and the system is carried out.
In one embodiment, the process of evaluating the rotational inertia of the system in real time in step 103 may specifically include:
based on the inertia damping influence factor set, extracting a core influence factor influencing the inertia of the system by using a principal component analysis method or a decision tree method;
based on an integrated learning theory, a plurality of learners are selected to relearn the association rule characteristics from the core influence factors to the inertia of the system;
real-time data of the power system are obtained, sequence restoration and supplementation are carried out on the real-time data based on a data fitting technology, low-time precision data are extracted, and the inertia level of the power system is analyzed in real time based on the low-time precision data and association rule characteristics.
The method comprises the steps of factor extraction, rule learning and online evaluation, wherein firstly, an inertial influence factor set is obtained by utilizing the inertial characteristic learning model of the system, and core influence factors are extracted by utilizing methods such as principal component analysis, decision trees and the like; then, based on an integrated learning theory, a plurality of learners are selected to relearn association rules from core influence factors to system inertia together, so that model generalization capability under the conditions of feature reduction and relatively insufficient samples is improved; further, in consideration of the requirement of rapidly obtaining key influence factors in real-time operation, a sequence reduction supplementing technology of low-time precision data is provided based on a data fitting technology.
In one embodiment, as shown in fig. 4, the process of evaluating the AGC reserve capacity of the primary frequency modulation and the secondary frequency modulation of the system in real time in step 103 may specifically include:
step 1031, determining the capacity and proportion of each type of online power supply of the power system;
step 1032, based on the primary frequency modulation parameter identification model, the new energy frequency response strategy aggregation model and the determined capacity and proportion of each type of online power supply of the power system, carrying out statistical analysis on the hydropower active frequency characteristic, the thermal power active frequency characteristic and the new energy cluster active frequency characteristic of the power system to obtain an online power supply frequency modulation characteristic description model of the whole power system;
Step 1033, extracting physical characteristics and external characteristics of the power grid based on the full sample data set to form a generalized characteristic set, and identifying the mapping relation between the generalized characteristic set and an online power supply frequency modulation characteristic description model of the whole power system by using an association rule learning model;
and 1034, linearly combining the mapping relations to form a frequency modulation capacity comprehensive evaluation model of the power system, and evaluating the primary frequency modulation and secondary frequency modulation reserve capacity level of the controllable power supply of the power system based on the frequency modulation capacity comprehensive evaluation model.
The comprehensive analysis learning and the induction learning model are carried out together, so that the power supply frequency modulation capacity evaluation is realized. The analysis learning model is based on a physical rule and follows a thought from bottom to top, and the embodiment establishes a frequency modulation characteristic description model of the whole power supply by using a statistical analysis method on the basis of obtaining the capacity and proportion of various types of online power supplies of the system and combining the active frequency characteristics of the hydroelectric power, the thermal power and the new energy clusters obtained in the embodiment;
in the embodiment, the relation between complex variables is regarded as a 'black box' model through induction learning, association rules are directly obtained by samples, and the method is carried out according to the thought from top to bottom; and finally, establishing an analysis-induction comprehensive learning model by using linear combination and other modes, thereby realizing real-time evaluation of AGC reserve capacity of primary frequency modulation and secondary frequency modulation of the system.
In one embodiment, the wide-area large-scale new energy frequency modulation available capacity modeling process based on the wind/light resource information in step 103 may specifically include:
acquiring historical operation data of a new energy station in the power system and ultra-short-term prediction data of new energy of a dispatching center;
based on historical operation data of new energy stations and ultra-short-term prediction data of new energy of a dispatching center, restoring the maximum power generation capacity of the new energy at each moment, and identifying a power generation capacity boundary;
extracting the power generation characteristics of the power supply equipment as a characteristic parameter set, using the characteristic parameter set and the historical NWP data as inputs of a learner, and constructing association rules between the maximum power generation capacity of new energy and meteorological resources and equipment characteristics by using a perceptron;
the method comprises the steps of extracting uncertain events of a new energy station to form an uncertain factor probability set, taking the uncertain factor probability set and the maximum power generation capacity of new energy at each moment together as input of a learning machine, utilizing a Bayesian learner to discover new energy frequency modulation available capacity probability description in an uncertain environment, and carrying out real-time evaluation on the new energy frequency modulation available capacity based on the new energy frequency modulation available capacity probability description in the uncertain environment.
The frequency modulation reserve capacity of the new energy power supply is jointly influenced by meteorological resources and various uncertain factors, and the boundary identification is firstly carried out on the upper limit of the new energy power generation capacity determined by the meteorological resources by adopting the modes of characteristic mining, probability modeling and rule learning. Firstly, restoring the maximum power generation capacity of new energy at each moment based on historical operation data of new energy stations and ultra-short-term prediction data of new energy of a dispatching center; and extracting the power generation characteristic of the power supply equipment as a characteristic parameter set, and inputting the characteristic parameter set and the historical NWP data as a learner. And constructing association rules between the maximum power generation capacity of the new energy and the meteorological resources and equipment characteristics by adopting a perceptron. And then, establishing a probability model of uncertain events such as station equipment faults, outgoing line faults, section out-of-limit and the like, taking the probability model and the power generation boundary capacity as learning machine input, and utilizing a Bayesian learner to discover new energy frequency modulation available capacity probability description in an uncertain environment. In the online evaluation of the power grid, the power generation capacity boundary can be directly replaced by ultra-short-term prediction, and the uncertain event probability model is replaced by using long-term statistics mastered by scheduling, so that the probability evaluation of the frequency modulation available capacity of the new energy is realized.
In one embodiment, the process of performing the active frequency security situation prediction of the large power grid for taking into account the new energy power generation and the real-time adjustment capability of the system in step 103 may specifically include:
selecting a typical event with higher perception of active change of a new energy station in an electric power system as a sample;
calculating a system frequency security situation representation index of the power system based on a frequency dynamic curve of the primary frequency modulation process;
taking the full sample data set as input, adopting an LSTM neural network as a learner, and establishing a mapping model from complex multi-element features to system security situation characterization indexes;
based on an analysis result of the inertia level of the power system, a primary frequency modulation and secondary frequency modulation reserve capacity level evaluation result of the controllable power supply and a real-time evaluation result of available frequency modulation capacity of new energy, predicting the frequency security situation of the power system by using a time domain simulation method;
the characterization index comprises a frequency lowest point, a frequency change rate, a post-hoc frequency steady state value and a recovery time corresponding to the typical event.
In the embodiment, a sample generation, feature selection and mapping modeling mode is adopted, an actual power grid is firstly selected as a research object, typical events with higher active change perceptibility such as new energy station off-grid or output dip, tie line power mutation and the like are selected as samples, a system frequency dynamic curve is obtained by utilizing the data preprocessing technology provided by the embodiment, and corresponding frequency minimum points, frequency change rates, post-frequency steady-state values, recovery time and the like are calculated to form a system frequency security situation representation index; the method comprises the steps of using a system full-sample feature set established by the embodiment as input, using an LSTM neural network as a learner, and establishing a mapping model from complex multi-element features to system security situation indexes; finally, based on the inertia and frequency modulation capacity evaluation method provided by the embodiment, the effect comparison is carried out on the safety situation prediction by combining a time domain simulation method, so that the active frequency safety situation prediction of the large power grid considering the power generation of new energy and the real-time adjustment capability of the system is realized.
In the embodiment of the invention, the common influence of uncertain factors such as power supply, load composition and meteorological resources on the system inertia and secondary frequency modulation characteristics is considered, so that the generalized characteristics are taken as situation prediction models to be input, and the system frequency characteristics are directly obtained by the multi-element mass information. According to the embodiment of the invention, on the basis of extracting the characteristics of the system operation electric physical quantity and the external environment as the input of the learning model, the core factors are further extracted by utilizing the inertial and frequency modulation capacity main influence factor association relation model obtained by the embodiment and utilizing the methods of principal component analysis and the like, so that the online evaluation speed is increased.
And 104, determining a frequency modulation strategy of the power system according to the estimated result and the predicted result, and realizing active frequency control of the power system based on the frequency modulation strategy.
In this embodiment, according to the evaluation result and the prediction result of step 103, in combination with the frequency modulation characteristic analysis result of step 102, an intelligent learning model can be established, and an AGC master station control strategy in the power system is fed back and parameter optimization is performed; the new energy AGC substation control performance can be comprehensively evaluated, and the new energy AGC control strategy and the optimizing mode can be fed back based on the comprehensive evaluation index.
In one embodiment, the process of feeding back the AGC master control strategy and performing parameter optimization in the power system in step 104 may specifically include:
establishing a model training data set of AGC frequency characteristic coefficients, conventional energy and new energy coordination control parameters and unit/station regulation performance based on an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the electric power system, as well as an analysis result of the inertia level of the electric power system, a primary frequency modulation and secondary frequency modulation reserve capacity level evaluation result of a controllable power supply and a real-time evaluation result of available frequency modulation capacity of the new energy;
extracting first, second and third sample data based on the full sample data set;
training an AGC frequency characteristic coefficient by using the first sample data;
training the coordination control parameters of the conventional energy and the new energy based on the second sample data to obtain the coordination control parameters of the dynamic conventional energy and the new energy applicable to different power grid running states;
training an adjustment performance model based on the third sample data to obtain an adjustment rule of a unit/station, and determining an AGC control instruction allocation strategy based on the adjustment rule, the trained AGC frequency characteristic coefficient and dynamic conventional energy and new energy coordination control parameters applicable to different power grid running states;
The first sample data comprises ACE, frequency, exchange power, CPS/A standard check and load data; the second sample data comprises ACE, frequency, exchange power, standby under conventional energy AGC, load variation and link variation data; the third sample data includes AGC command time, AGC command value, AGC command issue time output, installed capacity, response time, adjustment accuracy, and adjustment rate data.
Aiming at different typical application scenes of a complicated large power grid, the embodiment firstly acquires on-line operation data of the power grid, counts historical operation data records of all new energy stations, considers prediction data of all current new energy stations, and determines the power regulation direction and regulation capacity which can be carried out by all new energy stations; based on the power response characteristic difference of a conventional unit or a new energy station in the system, the regional unbalanced power is distributed in a differentiated mode, the coordination control of the conventional unit and the new energy station is realized, a large number of simulation calculation analyses are carried out aiming at different typical working conditions of a complex large power grid, and an AGC master station control strategy and optimization parameters adapting to the working conditions of various complex large power grids are formed through intelligent learning.
Based on the frequency adjustment characteristics and the frequency adjustment capability evaluation results in the above embodiments, a model training data set of AGC frequency characteristic coefficients, conventional energy and new energy coordination control parameters, and unit/station adjustment performance is established. Training AGC frequency characteristic coefficients based on a large amount of historical data, including sample data consisting of ACE, frequency, exchange power, CPS/A standard assessment, load and the like by utilizing an intelligent learning algorithm; based on a large amount of sample data consisting of data such as ACE, frequency, exchange power, standby under conventional energy AGC, load variation, tie line variation and the like, conventional energy and new energy coordination control parameter training is carried out, and dynamic conventional energy and new energy coordination control parameters suitable for different power grid running states are obtained; based on the AGC command time, the AGC command value, the command issuing time output, the installed capacity, the response time, the adjustment precision, the adjustment speed and other large amount of sample data of the unit/station, the adjustment performance model is trained to obtain the adjustment rule of the unit/station, the AGC control command distribution strategy is guided, and the AGC control effect is improved.
In one embodiment, the process of comprehensively evaluating the control performance of the new energy AGC substation in the step 104 and feeding back the new energy AGC control strategy and the optimization mode based on the comprehensive evaluation index specifically may include:
based on the designated safety and stability guidelines of the power system and new energy standards, quantifying new energy AGC control performance parameters;
and determining an AGC substation intelligent control strategy based on the quantized new energy AGC control performance parameters, the new energy frequency response strategy aggregation model and the real-time evaluation result of the available frequency modulation capacity of the new energy.
According to the embodiment, the AGC control performance parameters of the new energy are quantized by analyzing the national standard power system safety and stability guidelines and the related enterprise standards of the new energy; and optimizing an AGC substation intelligent control strategy by using a new energy source based on data driving to participate in a frequency adjustment model of AGC and a large power grid online active frequency adjustment model based on situation prediction, and improving AGC control performance parameters by optimizing the substation control strategy.
The embodiment of the invention also provides a method for participating in the active frequency adjustment control strategy based on the multi-type energy storage and adjustable load driven by data, on one hand, the method is based on the massive operation data of the power system regulation center, and the change rule of the adjustable load and the change of the active frequency of the power grid are predicted through intelligent learning; on the other hand, the active frequency regulation strategy of the power grid is participated in coordination and optimization aiming at multi-type energy storage and adjustable load.
In one embodiment, the method for predicting the adjustable load change rule and the power grid active frequency change by intelligent learning according to the mass operation data of the power system regulation center specifically includes:
acquiring actual operation data of a power system regulation and control center, preprocessing the actual operation data, extracting time sequence data, acquiring tide and frequency data of high-voltage buses of industrial load and resident load, and forming a frequency prediction data set;
based on the frequency prediction data set, carrying out component decomposition and feature extraction on the load active sequences of all nodes of the power system by utilizing a feature indicator to obtain a load reference energy curve, and establishing a frequency prediction model of the power system under a steady state based on the load reference energy curve;
acquiring frequency regulation resource participation condition data and power grid power generation condition data of a power system, constructing a self-coding network to reconstruct input information, and realizing feature extraction of a frequency prediction data set;
based on the feature extraction result of the frequency prediction data set, the feature representation of the load active change rule of the power system, the inertia and damping parameter identification model, the primary frequency modulation parameter identification model, the new energy frequency response strategy aggregation model and the load frequency characteristic identification model of the power system, deep learning is carried out by utilizing a neural network, and a disturbed frequency prediction model of the power system is established;
And determining an active frequency control strategy of the multi-type energy storage and the adjustable load participating in adjustment based on the frequency prediction model under the steady state of the power system and the frequency prediction model after disturbance.
In the embodiment, the actual operation data of the power grid regulation and control center, such as the actual operation data of D5000, is collected by means of sample generation, parameter identification and rule learning, bad data rejection and curve principal component extraction are performed by means of filtering, outlier detection and other technologies, tide and frequency data of high-voltage buses of industrial loads and resident loads are obtained, and a training sample set is formed.
And secondly, carrying out component decomposition and feature extraction on the load active sequences of all the nodes by using a feature indicator such as sparse representation and the like to obtain a representation model of a load reference energy curve, wherein the reference energy curve is used for representing the original energy characteristic of the load, and providing a data base for frequency prediction under the steady state of the system.
And thirdly, aiming at the problem of active frequency change prediction of the power grid, massive data of frequency adjustment resource participation conditions (including participation types of frequency adjustment resources, adjustable frequency capacity, frequency adjustment characteristics, related uncertainty distribution and the like) and power generation conditions of the power grid (including load size and uncertainty distribution thereof, active power size and uncertainty distribution thereof) are sorted, and self-coding network structure reconstruction input information is constructed to obtain more effective feature description.
Finally, features generated by a self-coding network are used for replacing original data, and a mapping model between each influencing factor and the power grid frequency change is established by utilizing a deep neural network by combining the frequency modulation characteristic parameters learned by the embodiment, so that the frequency recovery process after disturbance and the deviation distribution condition of the frequency under normal operation are predicted.
In another embodiment, the process of optimizing the participating grid active frequency adjustment strategy for multi-type energy storage and adjustable load coordination may specifically include:
probability sampling is carried out on the running trend of the power grid for energy storage and adjustable load active control requirements aiming at the power grid frequency change, and a certain number of initial scenes are generated;
scene reduction is carried out on a certain number of initial scenes through a cluster analysis algorithm, and a typical scene is constructed;
based on actual operation data, simulating a typical scene under the control of different coordination strategies on a simulation platform of the power system to obtain simulation data of the typical scene;
based on simulation data of a typical scene, adjusting the correction direction of a target and the characteristic guiding rewarding function of various frequency modulation resources with different frequencies, and establishing a coordinated optimization knowledge model between energy storage and adjustable load based on reinforcement learning, wherein the adjustment target is a comprehensive index reflecting the adjustment characteristic and economic characteristic of a power grid;
Acquiring control requirements of a target scene, and classifying the target scene into a target typical scene based on the control requirements; generating a frequency modulation strategy of a target typical scene based on a coordinated optimization knowledge model between energy storage and adjustable load; and carrying out self-adaptive optimization on the frequency modulation strategy of the target typical scene based on the difference between the target scene and the target typical scene to obtain the frequency modulation strategy of the target scene.
Aiming at the problem of designing a coordination optimization control strategy of multi-type energy storage and adjustable load, the embodiment adopts a knowledge model building method based on reinforcement learning, firstly, probability sampling is carried out on the running trend of the power grid on the active control requirements of the energy storage and the adjustable load aiming at the power grid frequency change, a large number of initial scenes are generated, scene reduction is carried out by adopting a cluster analysis algorithm such as K-means, and a typical scene is built from a large number of running scenes. And then, based on actual operation data of a D5000 platform and related data under the control of different coordination strategies of a typical scene generated by simulation of a corresponding simulation platform, guiding the correction direction of a reward function by utilizing different frequency adjustment targets and characteristics (available frequency adjustment capacity, response speed of frequency, cost generated by frequency adjustment and the like) of various frequency adjustment resources, and completing establishment of a coordination optimization knowledge model between energy storage and adjustable load based on reinforcement learning, wherein the adjustment targets can select comprehensive indexes comprehensively reflecting the adjustment characteristics and economic characteristics of a power grid. And finally, classifying any given scene based on the control requirement of the scene, converging the scene to the typical scene, and carrying out self-adaptive adjustment on the frequency adjustment strategy according to the difference between the target scene and the typical scene.
In the embodiment of the invention, aiming at the practical problems that the multi-frequency processes of the power system are crossed and the multi-interference factors are interwoven, and the frequency characteristic analysis and the parameter calculation are difficult, the invention adds a new technical idea of data driving on the basis of the mechanism model analysis, and realizes the mixed analysis of the white box and black box models. The embodiment of the invention is based on actual data, analyzes the time sequence characteristics of each frequency process by combining with the dynamic theory of the power system, realizes component identification and separation of multiple processes by utilizing frequency domain decomposition technologies such as wavelet analysis and the like on the actual data, and selects a scene with obvious inertial response characteristics to analyze the inertial response parameters of the system by combining with historical operation conditions; considering that the obtained typical sample data are less, a wide data range full sample set is obtained by utilizing a data enhancement technology, and the generalization capability of a later learning model is improved. Meanwhile, in order to solve the problem that the system frequency is influenced by multiple factors together, inertia, damping and power supply primary frequency modulation parameters are accurately calculated by utilizing parameter identification based on a white box model in a sample set, and a mapping model of system electric quantity characteristics, external environment characteristics and frequency characteristic parameters is learned by utilizing a neural network and other models based on a black box model in a characteristic analysis part, so that association relation development is realized.
The inertia and damping characteristics of the system are mainly reflected in the process of large-range dynamic change of frequency, in the classical power system stability theory, the values of transition time, frequency nadir and the like between different steady-state frequencies depend on the inertia and damping of the system and the frequency modulation characteristics of a power supply and a load. In normal steady-state operation of the power grid, the active power is in a small disturbance or even no disturbance state, the system inertia is difficult to identify, and the existing system inertia calculation method is required to depend on a specially designed active power large disturbance experiment. With the continuous increase of the uncertainty of the system, the diversity of the operation modes is greatly improved, the inertia change condition of the system is difficult to predict, disturbance tests cannot be carried out under various working conditions under the actual conditions to analyze the inertia constant, and a system inertia and frequency modulation capacity evaluation mode with strong generalization capability under the condition of small samples is provided in a system with multiple operation modes and increased uncertainty factors. The invention has the difficulty that under the condition of non-deep frequency fluctuation, the traditional method is difficult to identify the system inertia and the frequency modulation reserve.
Aiming at the difficulty, the embodiment of the invention firstly selects typical active frequency events, solves the system inertia and the power supply side frequency modulation reserve capacity under a typical scene by utilizing a parameter identification method, then expands the sample size by utilizing a sample enhancement technology, simultaneously solves the association rule between various characteristics and the system frequency characteristics by utilizing an integrated learning model, directly utilizes primary information which can be directly mastered by a power grid dispatching center in real-time operation as surface layer characteristics to carry out evaluation and analysis, and improves the model generalization capability in a multi-measure way so as to realize the inertia evaluation in a small-sample and complex unknown system.
For deep learning of large power grid operation data, a single algorithm is difficult to effectively extract, the composition of data clusters is not fully considered, and the accuracy of data cluster extraction is not enough, so that large errors exist, and further deep analysis and optimization of large power grid AGC control parameters are difficult. The rapid development of artificial intelligence, applications in the energy internet include intelligent sensing, intelligent analysis, intelligent decision making and intelligent control, wherein sensing, analysis and decision making are the preconditions and basis for control. However, regardless of how much and how powerful the artificial intelligence technology is developed, the "big data" is easily changed into "dead data" if not combined with the characteristics and scenarios of the power system, and big data driven technologies are difficult to combine with the grid physical characteristics. Physical priori knowledge is effectively combined with artificial intelligence, so that the advantages and strong items of the artificial intelligence in the ubiquitous electric power Internet of things can be exerted to the greatest extent. Therefore, how to deeply mine the large power grid operation data, establish the relationship set between the AGC control strategy and the parameters and the power grid operation data, explore the intelligent learning model training technology for researching the AGC control parameters of the large power grid, and is a difficulty for realizing the research of optimizing the AGC control strategy of the large power grid by utilizing intelligent learning. Based on the above, the embodiment of the invention firstly analyzes and establishes a theoretical data model among AGC control strategy, parameters and power grid operation data based on the operation rule of the large power grid and the AGC control target and strategy; and secondly, based on a large amount of power grid operation data in the D5000 dispatching system, performing accuracy analysis, validity screening, filtering and checking on the data by utilizing a big data analysis tool, and further establishing a model training data set of the AGC control parameters of the big power grid based on the valid data.
In addition, the difference between different types of energy storage and adjustable load control characteristics is large, the dispersion degree of the adjustable load control response process is high, the control uncertainty is large, the traditional deterministic modeling and control strategy is difficult to meet the requirements of coordinated control on economy and safety, and the coordination optimization strategy is found out by adopting a reinforcement learning mode, so that the method is an effective technical means. However, the reinforcement learning needs a large amount of operation basic data and a proper evaluation mode to train to obtain a proper knowledge model, so how to obtain a large amount of training sample sets from limited practical operation samples by means of simulation or data reasoning, and further propose an energy storage and adjustable load optimization control strategy based on reinforcement learning is still another important point and difficulty to be solved by the invention. Based on the actual response and adjustment data of all energy storage and adjustable loads in a power grid are collected, and a reinforcement learning sample set based on actual basic data is constructed. And verifying simulation results of a simulation system model by using actual operation coefficients based on a simulation model of an actual power grid, and obtaining a large amount of power grid operation basic learning data by using corrected power grid simulation systems and simulating simulation control results and data reasoning of different scenes and different control strategies, constructing a typical operation scene training data set by adopting modes such as clustering and the like, finally constructing a reinforcement learning evaluation function by combining actual power grid scheduling requirements and typical scenes and designing a proper evaluation mode, and providing an energy storage and adjustable load optimization control strategy based on reinforcement learning.
The invention has the following specific application: the method comprises the steps of researching the existing automatic power generation control function of XX company, including a conventional power supply automatic power generation control system and a new energy automatic power generation control system based on a D5000 platform, analyzing the control strategy of the existing conventional power supply AGC system and the new energy AGC system and the interaction relation between the two systems, and designing the whole architecture and the function division scheme of the AGC system based on intelligent learning.
The intelligent learning relies on a large amount of high-quality historical operation data of the power grid, massive control data related to AGC frequency control are filtered and extracted, and the data architecture and the data interface scheme of the AGC system based on the intelligent learning are designed by combining the results of frequency adjustment characteristics, frequency adjustment capability assessment and the like. And a unified intelligent learning algorithm interface is developed, so that the method is suitable for training various parameters. The interfaces of all parts are uniform and relatively independent, so that the system is easy to adapt to the continuously-changing service demands, and is more convenient, safer and more reliable to maintain.
The characterization model of the active, frequency external characteristics and the regulating capability of the multi-type power supply and the load, which is provided by the invention, fully considers the characteristics of system components such as hydroelectric power, thermal power, new energy power, converter load and the like in the power system from two aspects of external characteristics and cluster expression, can be applied to evaluation work such as power grid frequency modulation capability equipment level modeling and system frequency response characteristic description with 'double high' characteristics in the future, is beneficial to fully excavating the frequency regulating capability of the power grid at the power supply side and the load side, and provides a basis for power system operation decision under uncertain environments.
The invention can bring direct benefit, and can develop demonstration verification on a power grid by establishing a frequency modulation capacity identification model, a frequency modulation capacity online evaluation model and situation prediction technology, an AGC strategy optimization model and an energy storage-load combined frequency modulation strategy model of each link and the whole system so as to practice a result verification theoretical method. The method provided by the invention can guide the frequency control of the region and the power grid, improve the system safety level under the conditions of high-capacity new energy access and direct current feed-in, and ensure the low-carbon, economic and safe operation of the power system. The method can realize that the forecasting precision error of key indexes such as the lowest frequency point is less than 0.01 Hz, assist regulatory personnel to realize the forecasting and pre-controlling of the safety risk, avoid the occurrence of malignant frequency events similar to the occurrence of the great power failure of British '8.9', and effectively prevent accidents such as large-scale new energy off-grid, load loss and the like. The invention has obvious risk management value, taking a great Britain blackout accident as an example, the total period of 1000MW load is forced to be off-line for 15-45 minutes, over 110 ten thousand people are affected, the direct loss can reach 700 ten thousand pounds according to the loss load value calculation of 9200 pounds per megawatt hour (provided by London Econamics), and the indirect loss of hospital blackout, industrial production stopping and the like is more difficult to estimate. Therefore, after the achievement of the invention is applied, various frequency risks can be effectively prevented, various events can be avoided, and the economic benefit is huge.
The invention can bring indirect benefit, and the successful research and development application of the scheme can enhance the technical strength in the technical field of system stability control based on artificial intelligence technology and promote technical progress. The achievement of the invention can provide standard establishment basis and engineering practice foundation for the aspects of strategy establishment, data technology and the like in the aspects of large-scale system active power, frequency control, system stability control and the like for the power industry, the energy supervision department, the academic community and the like; the invention can improve the utilization efficiency of new energy, promote the safe balance of energy crossing areas, is beneficial to accelerating the green low-carbon transformation process of an energy system, is beneficial to building a resource-saving and environment-friendly society, and relieves the energy crisis and global climate warming caused by shortage of fossil energy.
The following are device embodiments of the invention, for details not described in detail therein, reference may be made to the corresponding method embodiments described above.
Fig. 4 is a schematic structural diagram of a data-driven-based power system active frequency control system according to an embodiment of the present invention, and for convenience of explanation, only a portion relevant to the embodiment of the present invention is shown, as shown in fig. 4, the data-driven-based power system active frequency control system 4 may include: a data set forming unit 41, a tuning characteristic analyzing unit 42, a tuning ability evaluating and situation predicting unit 43 and a tuning strategy determining unit 44.
A data set forming unit 41, configured to obtain global historical data of a power system, and form a full sample data set based on the global historical data, where the global historical data includes PMU operation data, AGC instruction data, weather data, and user energy habit data of the power system at each historical time of each node of the power system;
the frequency modulation characteristic analysis unit 42 is configured to perform frequency modulation characteristic analysis based on the full-sample data set, and generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model, and a load frequency characteristic identification model of the electric power system;
a frequency modulation capability assessment and situation prediction unit 43, configured to perform frequency modulation capability assessment and situation prediction on the power system based on the generated model;
and the frequency modulation strategy determining unit 44 is configured to determine a frequency modulation strategy of the power system according to the evaluation result and the prediction result, and implement active frequency control of the power system based on the frequency modulation strategy.
In one embodiment, the fm characteristics analysis unit 42 may be specifically configured to obtain a typical point in time when an active disturbance occurs in the power system; extracting sample data of typical time points from the whole sample data set to form a typical sample set; noise attenuation and data preprocessing are carried out on a typical sample set; calculating system inertia and damping parameters of a typical moment point by using a parameter estimation method based on a frequency dynamic process mechanism model of the power system; expanding the calculated system inertia and damping parameters into a full-sample data set, and carrying out data enhancement by utilizing a time domain simulation or data generation technology; performing feature learning based on the full sample data set after data enhancement to obtain an inertial damping influence factor set of the power system; and training the LSTM neural network based on the inertial damping influence factor set to obtain an inertial and damping parameter identification model of the power system.
In one embodiment, the frequency modulation characteristic analysis unit 42 may be specifically configured to extract PMU data of the controllable-power high-voltage bus from the full-sample dataset; frequency analysis is carried out on the extracted PMU data to obtain a frequency dynamic curve of a primary frequency modulation process; the active-frequency sensitivity relation of each node of the power system and each time unit is obtained in a time difference mode, and a unit frequency modulation parameter set with space-time dependency characteristics is formed; based on the unit frequency modulation parameter set and the full sample data set, performing association relation learning between the conventional unit frequency modulation characteristic and the physical operation characteristic and the external environment characteristic of the system by using a neural network method to obtain a primary frequency modulation parameter influence factor set; and based on the primary frequency modulation parameter influence factor set, carrying out full-network on-line unit aggregation of the power system by utilizing a classification algorithm to obtain a primary frequency modulation parameter identification model.
In one embodiment, the fm characteristics analysis unit 42 may be specifically configured to extract PMU data of a new energy station grid-connected point from the full sample data set, and obtain an active-frequency time sequence of a station level or a distributed new energy source after data preprocessing; decomposing the active-frequency time sequence to obtain a primary frequency modulation response process sequence, identifying frequency modulation parameters of the primary frequency modulation response process sequence, and establishing a station-level power electronic interface external characteristic model based on the identified frequency modulation parameters; extracting a wide-area feature set from the full-sample data set, training a neural network model based on the wide-area feature set, and determining the association relation between the wide-area feature set, a new energy station level ontology control strategy and new energy frequency response characteristics; and determining a new energy station frequency modulation characteristic influence factor set based on the association relation, and carrying out classification aggregation of the new energy station based on the new energy station frequency modulation characteristic influence factor set to obtain a new energy frequency response strategy aggregation model.
In one embodiment, the fm characteristics analysis unit 42 may be specifically configured to extract PMU data of the converter load bus from the full sample dataset and perform data preprocessing; the active-frequency micro-variable proportion relation of each space point and each moment point is obtained by using a long-time differential sequence, so that the identification of load characteristic parameters is realized; performing feature learning and principal component analysis on the system electrical quantity features and the external environment features in the full-sample data set, removing redundant features, and obtaining a load frequency characteristic influence factor set; based on the load frequency characteristic influence factor set, the association rule of the load frequency characteristic and the influence factor is characterized and learned by utilizing a neural network algorithm with time memory characteristics, and a load frequency characteristic identification model is obtained.
In one embodiment, the frequency modulation capability assessment and situation prediction unit 43 may be specifically configured to extract, based on the set of inertial damping influence factors, a core influence factor that influences inertia of the system by using a principal component analysis method or a decision tree method; based on an integrated learning theory, a plurality of learners are selected to relearn the association rule characteristics from the core influence factors to the inertia of the system; real-time data of the power system are obtained, sequence restoration and supplementation are carried out on the real-time data based on a data fitting technology, low-time precision data are extracted, and the inertia level of the power system is analyzed in real time based on the low-time precision data and association rule characteristics.
In one embodiment, the frequency modulation capability assessment and situation prediction unit 43 may be specifically configured to determine the capacity and proportion of each type of online power source of the power system; based on the primary frequency modulation parameter identification model, the new energy frequency response strategy aggregation model and the determined online power capacity and proportion of each type of the power system, carrying out statistical analysis on the hydropower active frequency characteristic, the thermal power active frequency characteristic and the new energy cluster active frequency characteristic of the power system to obtain an online power frequency modulation characteristic description model of the whole power system; extracting physical characteristics and external characteristics of a power grid based on a full-sample data set to form a generalized characteristic set, and identifying a mapping relation between the generalized characteristic set and an online power supply frequency modulation characteristic description model of the whole power system by using an association rule learning model; and linearly combining the mapping relations to form a frequency modulation capacity comprehensive evaluation model of the power system, and evaluating the primary frequency modulation and secondary frequency modulation reserve capacity level of the controllable power supply of the power system based on the frequency modulation capacity comprehensive evaluation model.
In one embodiment, the frequency modulation capability assessment and situation prediction unit 43 may be specifically configured to obtain historical operation data of a new energy station in the power system and ultra-short term prediction data of new energy of the dispatching center; based on historical operation data of new energy stations and ultra-short-term prediction data of new energy of a dispatching center, restoring the maximum power generation capacity of the new energy at each moment, and identifying a power generation capacity boundary; extracting the power generation characteristics of the power supply equipment as a characteristic parameter set, using the characteristic parameter set and the historical NWP data as inputs of a learner, and constructing association rules between the maximum power generation capacity of new energy and meteorological resources and equipment characteristics by using a perceptron; the method comprises the steps of extracting uncertain events of a new energy station to form an uncertain factor probability set, taking the uncertain factor probability set and the maximum power generation capacity of new energy at each moment together as input of a learning machine, utilizing a Bayesian learner to discover new energy frequency modulation available capacity probability description in an uncertain environment, and carrying out real-time evaluation on the new energy frequency modulation available capacity based on the new energy frequency modulation available capacity probability description in the uncertain environment.
In one embodiment, the frequency modulation capability assessment and situation prediction unit 43 may be specifically configured to select a typical event with a higher active change perceptibility of a new energy station in the power system as a sample; calculating a system frequency security situation representation index of the power system based on a frequency dynamic curve of the primary frequency modulation process; taking the full sample data set as input, adopting an LSTM neural network as a learner, and establishing a mapping model from complex multi-element features to system security situation characterization indexes; based on an analysis result of the inertia level of the power system, a primary frequency modulation and secondary frequency modulation reserve capacity level evaluation result of the controllable power supply and a real-time evaluation result of available frequency modulation capacity of new energy, predicting the frequency security situation of the power system by using a time domain simulation method; the characterization index comprises a frequency lowest point, a frequency change rate, a post-hoc frequency steady state value and a recovery time corresponding to the typical event.
In one embodiment, the frequency modulation strategy determining unit 44 may be specifically configured to establish a model training data set of AGC frequency characteristic coefficients, conventional energy and new energy coordination control parameters, and unit/station adjustment performance based on an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model, a load frequency characteristic identification model of the power system, an analysis result of an inertia level of the power system, a primary frequency modulation and secondary frequency modulation reserve capacity level evaluation result of the controllable power supply, and a real-time evaluation result of available frequency modulation capacity of the new energy; extracting first, second and third sample data based on the full sample data set; training an AGC frequency characteristic coefficient by using the first sample data; training the coordination control parameters of the conventional energy and the new energy based on the second sample data to obtain the coordination control parameters of the dynamic conventional energy and the new energy applicable to different power grid running states; training an adjustment performance model based on the third sample data to obtain an adjustment rule of a unit/station, and determining an AGC control instruction allocation strategy based on the adjustment rule, the trained AGC frequency characteristic coefficient and dynamic conventional energy and new energy coordination control parameters applicable to different power grid running states; the first sample data comprises ACE, frequency, exchange power, CPS/A standard check and load data; the second sample data comprises ACE, frequency, exchange power, standby under conventional energy AGC, load variation and link variation data; the third sample data includes AGC command time, AGC command value, AGC command issue time output, installed capacity, response time, adjustment accuracy, and adjustment rate data.
In one embodiment, the frequency modulation strategy determination unit 44 may be specifically configured to quantify new energy AGC control performance parameters based on specified power system safety and stability guidelines and new energy standards; and determining an AGC substation intelligent control strategy based on the quantized new energy AGC control performance parameters, the new energy frequency response strategy aggregation model and the real-time evaluation result of the available frequency modulation capacity of the new energy.
In one embodiment, the active frequency control system 4 may further include:
the frequency prediction data set forming unit is used for acquiring actual operation data of the power system regulation and control center, preprocessing the actual operation data, extracting time sequence data, acquiring tide and frequency data of the industrial load and resident load high-voltage buses, and forming a frequency prediction data set;
the first frequency prediction model building unit is used for carrying out component decomposition and feature extraction on the load active sequences of all nodes of the power system by utilizing the feature indicator based on the frequency prediction data set to obtain a load reference energy utilization curve, and building a frequency prediction model of the power system under a steady state based on the load reference energy utilization curve;
the self-coding network construction unit is used for acquiring frequency regulation resource participation condition data of the power system and power grid power generation condition data, constructing self-coding network reconstruction input information and realizing feature extraction of a frequency prediction data set;
The second frequency prediction model building unit is used for building a disturbed frequency prediction model of the power system by utilizing a neural network based on a feature extraction result of the frequency prediction data set, a feature representation of a load active change rule of the power system, an inertia and damping parameter identification model of the power system, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model;
the active frequency control strategy determining unit is used for determining an active frequency control strategy of the multi-type energy storage and the adjustable load participating in adjustment based on the frequency prediction model in the steady state of the power system and the frequency prediction model after disturbance.
In one embodiment, the active frequency control system 4 may further include:
the initial scene generation unit is used for probability sampling of the running trend of the power grid on the active control requirements of the energy storage and the adjustable load aiming at the power grid frequency change and generating a certain number of initial scenes;
the typical scene construction unit is used for carrying out scene reduction on a certain number of initial scenes through a cluster analysis algorithm to construct typical scenes;
the data simulation unit is used for simulating the typical scene under the control of different coordination strategies on the basis of the actual operation data on a simulation platform of the power system to obtain simulation data of the typical scene;
The coordination optimization model building unit is used for guiding the correction direction of the rewarding function according to the simulation data of the typical scene, adjusting the characteristics of the targets and various frequency modulation resources according to different frequencies, and building a coordination optimization knowledge model between the energy storage and the adjustable load based on reinforcement learning, wherein the adjusting targets are comprehensive indexes reflecting the adjusting characteristics and the economic characteristics of the power grid;
the scene frequency modulation strategy determining unit is used for obtaining the control requirement of the target scene and classifying the target scene into a target typical scene based on the control requirement; generating a frequency modulation strategy of a target typical scene based on a coordinated optimization knowledge model between energy storage and adjustable load; and carrying out self-adaptive optimization on the frequency modulation strategy of the target typical scene based on the difference between the target scene and the target typical scene to obtain the frequency modulation strategy of the target scene.
As shown in fig. 5, the embodiment of the present invention further provides an electric power system 5, including: the controllable power supply, the new energy power supply and the converter load can be a hydropower station or a thermal power station, the new energy power supply can be a photovoltaic power station or a wind power station, and the power system 5 also comprises an AGC control system and an active power control system; the active frequency control system can execute the active frequency control method of the power system based on data driving, which is proposed by any embodiment, and output an AGC control strategy; the AGC control system generates AGC control commands based on an AGC control strategy to control the power system 5 for active frequency adjustment.
According to the power system provided by the invention, based on a data driving method, the system source, load, active-frequency equipment level and system level characteristics are fully identified, the real-time online situation sensing and predicting capacity of the power grid is improved, the frequency characteristic sensing capacity of the high-duty power grid of the new energy power supply and the frequency characteristic sensing capacity of the AC/DC interconnected receiving end power grid can be effectively improved, the active control capacity and frequency safety of the power system are ensured, and the safe and stable operation of the power grid is ensured.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.
Claims (15)
1. The utility model provides a power system active frequency control method based on data drive which characterized in that includes:
acquiring global historical data of a power system, and forming a full sample data set based on the global historical data, wherein the global historical data comprises PMU operation data, AGC instruction data, meteorological data and user energy habit data of the power system at each historical moment of each node of the power system;
Performing frequency modulation characteristic analysis based on the full-sample data set to generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system;
performing frequency modulation capability assessment and situation prediction on the power system based on the generated model;
and determining a frequency modulation strategy of the power system according to the evaluation result and the prediction result, and realizing active frequency control of the power system based on the frequency modulation strategy.
2. The method of claim 1, wherein the generating the inertia and damping parameter identification model, the primary frequency modulation parameter identification model, the new energy frequency response strategy aggregation model and the load frequency characteristic identification model of the power system based on the full sample data set comprises:
acquiring a typical moment point when the power system generates active disturbance;
extracting sample data of the typical time point from the whole sample data set to form a typical sample set;
performing noise attenuation and data preprocessing on the typical sample set;
Calculating system inertia and damping parameters of the typical moment by using a parameter estimation method based on a frequency dynamic process mechanism model of the power system;
expanding the calculated system inertia and damping parameters into the full-sample data set, and carrying out data enhancement by utilizing a time domain simulation or data generation technology;
performing feature learning based on the full-sample data set after data enhancement to obtain an inertial damping influence factor set of the power system;
and training an LSTM neural network based on the inertial damping influence factor set to obtain an inertial and damping parameter identification model of the power system.
3. The method of claim 2, wherein the performing frequency modulation characteristic analysis based on the full sample data set to generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system comprises:
extracting PMU data of a controllable power supply high-voltage bus from the full-sample data set;
performing frequency analysis on the extracted PMU data to obtain a frequency dynamic curve of a primary frequency modulation process;
Obtaining the active-frequency sensitivity relation of each node and each time unit of the power system in a time difference mode to form a unit frequency modulation parameter set with space-time dependency characteristics;
based on the unit frequency modulation parameter set and the full sample data set, performing association relation learning between the conventional unit frequency modulation characteristic and the physical operation characteristic and the external environment characteristic of the system by using a neural network method to obtain a primary frequency modulation parameter influence factor set;
and based on the primary frequency modulation parameter influence factor set, carrying out whole-network online unit aggregation of the electric power system by using a classification algorithm to obtain the primary frequency modulation parameter identification model.
4. The method of claim 3, wherein the performing frequency modulation characteristic analysis based on the full sample data set to generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system comprises:
extracting PMU data of new energy station grid-connected points from the full-sample data set, and obtaining active-frequency time sequences of station levels or distributed new energy after data preprocessing;
Decomposing the active-frequency time sequence to obtain a primary frequency modulation response process sequence, identifying frequency modulation parameters of the primary frequency modulation response process sequence, and establishing a station-level power electronic interface external characteristic model based on the identified frequency modulation parameters;
extracting a wide-area feature set from the full-sample data set, training a neural network model based on the wide-area feature set, and determining association relations among the wide-area feature set, a new energy station level ontology control strategy and new energy frequency response characteristics, wherein the wide-area feature set comprises: meteorological data, load data, hydrological resource data and external environment attribute data; the wide-area feature set further includes: active disturbance data of a system, frequency change data of the system and active response data of a new energy station;
and determining a new energy station frequency modulation characteristic influence factor set based on the association relation, and carrying out classification aggregation of the new energy station based on the new energy station frequency modulation characteristic influence factor set to obtain the new energy frequency response strategy aggregation model.
5. The method of claim 4, wherein the performing frequency modulation characteristic analysis based on the full sample data set to generate an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system comprises:
Extracting PMU data of a converter load bus from the full-sample data set and carrying out data preprocessing;
the active-frequency micro-variable proportion relation of each space point and each moment point is obtained by using a long-time differential sequence, so that the identification of load characteristic parameters is realized;
performing feature learning and principal component analysis on the system electrical quantity features and the external environment features in the full-sample data set, removing redundant features, and obtaining a load frequency characteristic influence factor set;
and based on the load frequency characteristic influence factor set, performing characterization learning on association rules of the load frequency characteristic and the influence factor by using a neural network algorithm with time memory characteristics to obtain the load frequency characteristic identification model.
6. The data-driven power system active frequency control method of claim 5, wherein the performing frequency modulation capability assessment and situation prediction on the power system based on the generated model comprises:
based on the inertia damping influence factor set, extracting a core influence factor influencing the inertia of the system by using a principal component analysis method or a decision tree method;
based on an integrated learning theory, a plurality of learners are selected to relearn the association rule characteristics from the core influence factors to the inertia of the system;
And acquiring real-time data of the power system, performing sequence restoration and supplementation on the real-time data based on a data fitting technology, extracting low-time precision data, and analyzing the inertia level of the power system in real time based on the low-time precision data and the association rule characteristics.
7. The data-driven power system active frequency control method of claim 6, wherein the performing frequency modulation capability assessment and situation prediction on the power system based on the generated model comprises:
determining the capacity and proportion of various online power supplies of the power system;
based on the primary frequency modulation parameter identification model, the new energy frequency response strategy aggregation model and the determined online power capacity and proportion of each type of the power system, carrying out statistical analysis on the hydropower active frequency characteristic, the thermoelectricity active frequency characteristic and the new energy cluster active frequency characteristic of the power system to obtain an online power frequency modulation characteristic description model of the whole power system;
extracting physical characteristics and external characteristics of a power grid based on the full-sample data set to form a generalized characteristic set, and identifying a mapping relation between the generalized characteristic set and an online power supply frequency modulation characteristic description model of the whole power system by using an association rule learning model;
And linearly combining the mapping relation to form a frequency modulation capacity comprehensive evaluation model of the power system, and evaluating the primary frequency modulation and secondary frequency modulation reserve capacity level of the controllable power supply of the power system based on the frequency modulation capacity comprehensive evaluation model.
8. The data-driven power system active frequency control method of claim 7, wherein the performing frequency modulation capability assessment and situation prediction on the power system based on the generated model comprises:
acquiring historical operation data of a new energy station in the power system and ultra-short-term prediction data of new energy of a dispatching center;
based on the historical operation data of the new energy station and the ultra-short-term prediction data of the new energy of the dispatching center, restoring the maximum power generation capacity of the new energy at each moment, and identifying a power generation capacity boundary;
extracting the power generation characteristics of the power supply equipment as a characteristic parameter set, using the characteristic parameter set and the historical NWP data as inputs of a learner, and constructing association rules between the maximum power generation capacity of new energy and meteorological resources and equipment characteristics by using a perceptron;
and extracting uncertain events of the new energy station to form an uncertain factor probability set, taking the uncertain factor probability set and the maximum power generation capacity of the new energy at each moment together as input of a learning machine, utilizing a Bayesian learner to discover new energy frequency modulation available capacity probability description in an uncertain environment, and carrying out real-time evaluation on the new energy frequency modulation available capacity based on the new energy frequency modulation available capacity probability description in the uncertain environment.
9. The data-driven power system active frequency control method of claim 8, wherein the performing frequency modulation capability assessment and situation prediction on the power system based on the generated model comprises:
selecting a typical event with higher active change perceptibility of a new energy station in the power system as a sample;
calculating a system frequency security situation characterization index of the power system based on the frequency dynamic curve of the primary frequency modulation process, wherein the characterization index comprises a frequency lowest point, a frequency change rate, a post-event frequency steady-state value and a recovery time corresponding to the typical event;
taking the full sample data set as input, adopting an LSTM neural network as a learner, and establishing a mapping model from complex multi-element features to system security situation characterization indexes;
and predicting the frequency security situation of the power system by using a time domain simulation method based on the analysis result of the inertia level of the power system, the primary frequency modulation and secondary frequency modulation reserve capacity level evaluation result of the controllable power supply and the real-time evaluation result of the available frequency modulation capacity of the new energy.
10. The method according to claim 9, wherein determining a tuning strategy of the power system according to the result of the evaluation and the result of the prediction, and implementing the active frequency control of the power system based on the tuning strategy comprises:
Establishing a model training data set of AGC frequency characteristic coefficients, conventional energy and new energy coordination control parameters and unit/station regulation performance based on an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the electric power system, as well as an analysis result of the inertia level of the electric power system, a primary frequency modulation and secondary frequency modulation reserve capacity level assessment result of the controllable power supply and a real-time assessment result of the available frequency modulation capacity of the new energy;
extracting first, second and third sample data based on the full sample data set;
training an AGC frequency characteristic coefficient by using the first sample data, wherein the first sample data comprises ACE, frequency, switching power, CPS/A standard assessment and load data;
performing conventional energy and new energy coordination control parameter training based on the second sample data to obtain dynamic conventional energy and new energy coordination control parameters applicable to different power grid running states; wherein the second sample data comprises ACE, frequency, exchange power, standby under conventional energy AGC, load variation and link variation data;
And training an adjustment performance model based on the third sample data to obtain an adjustment rule of a unit/station, and determining an AGC control instruction allocation strategy based on the adjustment rule, the trained AGC frequency characteristic coefficient and the dynamic conventional energy and new energy coordination control parameters applicable to different power grid running states, wherein the third sample data comprises AGC instruction time, AGC instruction value, AGC instruction issuing time output, installed capacity, response time, adjustment precision and adjustment rate data.
11. The method according to claim 10, wherein determining a tuning strategy of the power system according to the result of the evaluation and the result of the prediction, and implementing the active frequency control of the power system based on the tuning strategy comprises:
based on the designated safety and stability guidelines of the power system and new energy standards, quantifying new energy AGC control performance parameters;
and determining an AGC substation intelligent control strategy based on the quantized new energy AGC control performance parameters, the new energy frequency response strategy aggregation model and the real-time evaluation result of the new energy available frequency modulation capacity.
12. The data-driven based power system active frequency control method according to any one of claims 1 to 11, characterized in that the method further comprises:
acquiring actual operation data of the power system regulation center, preprocessing the actual operation data, extracting time sequence data, acquiring tide and frequency data of an industrial load and resident load high-voltage bus, and forming a frequency prediction data set;
based on the frequency prediction data set, carrying out component decomposition and feature extraction on a load active sequence of each node of the power system by utilizing a feature indicator to obtain a load reference energy curve, and establishing a frequency prediction model of the power system under a steady state based on the load reference energy curve;
acquiring frequency regulation resource participation condition data and power grid power generation condition data of the power system, constructing a self-coding network to reconstruct input information, and realizing feature extraction of the frequency prediction data set;
based on the feature extraction result of the frequency prediction data set, the feature representation of the load active change rule of the power system, the inertia and damping parameter identification model, the primary frequency modulation parameter identification model, the new energy frequency response strategy aggregation model and the load frequency characteristic identification model of the power system, deep learning is carried out by utilizing a neural network, and a frequency prediction model after disturbance of the power system is established;
And determining an active frequency control strategy of the multi-type energy storage and the adjustable load participating in adjustment based on the frequency prediction model of the power system under the steady state and the frequency prediction model after disturbance.
13. The data-driven based power system active frequency control method according to any one of claims 1 to 11, characterized in that the method further comprises:
probability sampling is carried out on the running trend of the power grid for energy storage and adjustable load active control requirements aiming at the power grid frequency change, and a certain number of initial scenes are generated;
performing scene reduction on the initial scenes with a certain number through a cluster analysis algorithm to construct a typical scene;
based on the actual operation data, performing simulation of the typical scene under the control of different coordination strategies on a simulation platform of the power system to obtain simulation data of the typical scene;
based on simulation data of the typical scene, adjusting the characteristic of a target and various frequency modulation resources to guide the correction direction of a reward function with different frequencies, and establishing a coordinated optimization knowledge model between energy storage and adjustable load based on reinforcement learning, wherein the adjustment target is a comprehensive index reflecting the adjustment characteristic and economic characteristic of a power grid;
Acquiring control requirements of a target scene, and classifying the target scene into a target typical scene based on the control requirements;
generating a frequency modulation strategy of the target typical scene based on the coordinated optimization knowledge model between the energy storage and the adjustable load;
and carrying out self-adaptive optimization on the frequency modulation strategy of the target typical scene based on the difference between the target scene and the target typical scene to obtain the frequency modulation strategy of the target scene.
14. A data-driven power system active frequency control system, comprising:
the data set forming unit is used for acquiring global historical data of the power system and forming a full sample data set based on the global historical data, wherein the global historical data comprises PMU operation data, AGC instruction data, meteorological data and user energy habit data of the power system at each historical moment of each node of the power system;
the frequency modulation characteristic analysis unit is used for carrying out frequency modulation characteristic analysis based on the full-sample data set and generating an inertia and damping parameter identification model, a primary frequency modulation parameter identification model, a new energy frequency response strategy aggregation model and a load frequency characteristic identification model of the power system;
The frequency modulation capability assessment and situation prediction unit is used for carrying out frequency modulation capability assessment and situation prediction on the power system based on the generated model;
and the frequency modulation strategy determining unit is used for determining the frequency modulation strategy of the power system according to the evaluation result and the prediction result and realizing the active frequency control of the power system based on the frequency modulation strategy.
15. An electrical power system, the electrical power system comprising: the system comprises a controllable power supply, a new energy power supply and a converter load, wherein the controllable power supply comprises a hydroelectric generating set power supply and/or a thermal power generating set power supply;
the power system also comprises an AGC control system and an active power control system;
the active frequency control system performs the data drive-based power system active frequency control method according to any one of claims 1 to 13, outputting an AGC control strategy;
the AGC control system generates an AGC control instruction based on the AGC control strategy to control the power system to perform active frequency adjustment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311280913.9A CN117277307A (en) | 2023-09-28 | 2023-09-28 | Data-driven-based active frequency control method and system for power system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311280913.9A CN117277307A (en) | 2023-09-28 | 2023-09-28 | Data-driven-based active frequency control method and system for power system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117277307A true CN117277307A (en) | 2023-12-22 |
Family
ID=89204237
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311280913.9A Pending CN117277307A (en) | 2023-09-28 | 2023-09-28 | Data-driven-based active frequency control method and system for power system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117277307A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117913996A (en) * | 2024-01-24 | 2024-04-19 | 江苏同合电气有限公司 | Intelligent monitoring management method and system for operation of power distribution cabinet based on data analysis |
| CN118332703A (en) * | 2024-06-13 | 2024-07-12 | 中国航空工业集团公司西安飞机设计研究所 | Automatic generation method for shear slice structural parameters |
| CN118586677A (en) * | 2024-08-02 | 2024-09-03 | 国网山西省电力公司电力科学研究院 | Power distribution prediction method and system based on association rule analysis |
-
2023
- 2023-09-28 CN CN202311280913.9A patent/CN117277307A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117913996A (en) * | 2024-01-24 | 2024-04-19 | 江苏同合电气有限公司 | Intelligent monitoring management method and system for operation of power distribution cabinet based on data analysis |
| CN117913996B (en) * | 2024-01-24 | 2024-06-07 | 江苏同合电气有限公司 | Intelligent monitoring management method and system for operation of power distribution cabinet based on data analysis |
| CN118332703A (en) * | 2024-06-13 | 2024-07-12 | 中国航空工业集团公司西安飞机设计研究所 | Automatic generation method for shear slice structural parameters |
| CN118586677A (en) * | 2024-08-02 | 2024-09-03 | 国网山西省电力公司电力科学研究院 | Power distribution prediction method and system based on association rule analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | The comprehensive evaluation of smart distribution grid based on cloud model | |
| CN117277307A (en) | Data-driven-based active frequency control method and system for power system | |
| CN107947165A (en) | A kind of power distribution network flexibility evaluation method towards regulatory demand | |
| CN111428355A (en) | Modeling method for power load digital statistics intelligent synthesis | |
| CN104915889A (en) | Method of acquiring comprehensive load model parameters in online mode based on daily load curve | |
| Trojani et al. | Stochastic security-constrained unit commitment considering electric vehicles, energy storage systems, and flexible loads with renewable energy resources | |
| Cai et al. | Electric power big data and its applications | |
| Zhang et al. | Distributed Architecture of Power Grid Asset Management and Future Research Directions | |
| Qin et al. | An integrated situational awareness tool for resilience-driven restoration with sustainable energy resources | |
| CN110929968A (en) | Comprehensive regulation and control method and system for multi-energy combined supply of smart city | |
| Guan et al. | Enhancing security and resilience of bulk power systems via multisource big data learning | |
| Habib et al. | Opportunities and challenges of AC/DC transmission network planning considering high proportion renewable energy | |
| Sun et al. | Assessment of AC/DC hybrid microgrids based on grey comprehensive evaluation theory | |
| Ou et al. | A Review of Highly Reliable Operation Technologies and Evaluation Techniques for Microgrids in Remote Rural Areas | |
| Zhao et al. | Optimal Operation Control Strategies for Active Distribution Networks under Multiple States: A Systematic Review | |
| Zhang et al. | Research on refined prediction of coastal wind power based on dynamic downscale and deep learning prediction | |
| Jie et al. | Investigation on Power System Dynamic Monitoring System Using Artificial Neural Networks | |
| Li et al. | Post evaluation index system of smart grid pilot city planning | |
| CN117374946B (en) | Power distribution network scheduling management method and system based on digestion capability evaluation | |
| Xiang et al. | Application of Artificial Intelligence Technology in Power System Stability Assessment | |
| Tang et al. | Application technology of big data in smart grid and its development prospect | |
| Li et al. | SELF-ADAPTIVE OPTIMAL ALLOCATION STRATEGY OF EMERGENCY RESOURCES FOR POWER DISTRIBUTION NETWORK FAILURES | |
| Yang et al. | Artificial Intelligence Based Fault Diagnosis and Analysis in the Distribution Network | |
| Yang et al. | Risk Assessment Strategy for Distribution Network Operation considering New Power Access | |
| CN113283702B (en) | Power flow state evaluation method and device integrating safety and stability characteristics of power system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |