Method, device and equipment for constructing planning mode set of power system

Abstract

The embodiment of the invention provides a construction method, device and equipment for a planning mode set of a power system, and the method comprises the steps: firstly constructing a safety risk and weak mode identification index system, and carrying out the risk and weak link identification of a large number of operation modes generated through time sequence production simulation; then, risk modes are classified according to categories of safety, economy, green and the like, and typical risk modes, extreme risk modes and deduction modes are generated through feature aggregation; constructing a multi-type and multi-index high-dimensional coordinate system, and extracting an extreme operation mode based on algorithms such as a fuzzy entropy weight to meet a security check requirement; and finally, by analyzing the influence of an extreme mode on planning, combining feature sensitivity setting and a compatible boundary, defining a minimum planning mode set, and forming a novel power system planning reliable mode cluster suitable for a target region. The technical problems that in the prior art, when an electric power system is planned, evaluation indexes are single, expert experience or a simple statistical method is mostly depended on, and the regional adaptability is low are solved.

Description

Technical Field

[0001] This application provides embodiments in the field of power system technology, and particularly relates to a method, apparatus, and equipment for constructing a set of planning methods for a power system. Background Technology

[0002] With the advancement of the "dual carbon" goals, the penetration rate of new energy power generation (wind power, photovoltaics, etc.) continues to climb, and the power system is accelerating its transformation into a new "dual-high" power system with a high proportion of renewable energy and a high proportion of power electronic equipment. This transformation has brought about new problems such as a surge in uncertainty on both the source and load sides, a decrease in system inertia, and a reduction in voltage and frequency stability margins, which significantly increases the volatility and complexity of the power system's operating state. Traditional "deterministic" planning models are no longer able to cope with the security risks under massive random scenarios.

[0003] Traditional power system planning often relies on a few typical operating modes, such as maximum load, minimum load, and typical seasonal scenarios, neglecting the impact of edge scenarios such as extreme weather, sudden changes in renewable energy output, and sudden load fluctuations. For example, the combined effect of a surge in load caused by an extreme cold wave and a sharp drop in wind power output may trigger transmission channel overload or voltage collapse; while cloud cover during high-proportion photovoltaic grid connection may cause voltage fluctuations exceeding safe thresholds. These uncovered risk scenarios often become weak links in system operation, threatening the safety and stability of the power grid.

[0004] Existing research has significant limitations in risk identification and scenario extraction: First, risk assessment indicators often focus on a single dimension, such as safety or economy, lacking comprehensive quantification of multiple objectives such as safety, economy, and green (e.g., carbon emission intensity), leading to planning schemes that are incomplete in one aspect; Second, the selection of extreme operating modes often relies on expert experience or simple statistical methods, with subjective setting of indicator weights, making it difficult to accurately identify key risk scenarios from massive amounts of operational data; Third, there is insufficient regional adaptability. Eastern coastal areas such as Jiangsu have characteristics such as high load density, centralized grid connection of new energy sources, and high dependence on inter-regional power transmission, and general planning methods cannot meet their special needs, resulting in an imbalance between the economy and safety of planning schemes. Summary of the Invention

[0005] The embodiments of this application provide a method, apparatus, and equipment for constructing a set of planning methods for a power system, which solves the technical problems of existing power system planning, such as single evaluation indicators, reliance on expert experience or simple statistical methods, and low regional adaptability.

[0006] In a first aspect, embodiments of the present invention provide a method for constructing a set of planning methods for a power system, the method comprising:

[0007] Acquire multi-source time-series data of the target power system, wherein the multi-source time-series data is a set of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems;

[0008] The multi-source time-series data is standardized to obtain a standardized dataset;

[0009] Based on the standardized dataset, a new risk indicator system for the target power system is constructed using massive operational modes generated by time-series production simulation.

[0010] Based on the novel risk indicator system, the causes of risks affecting the consumption of new energy sources and the main influencing factors are identified.

[0011] A method for extracting typical high-risk moments based on the tSNE-FCM algorithm is used to analyze the extreme operation indicators of the minimum planning method set and to screen high-risk moments.

[0012] Based on the determined extreme operating mode indicators and the initial screening results of high-risk moments, a minimum planning method set for the safety risks and weaknesses of the target power system is constructed.

[0013] Secondly, embodiments of the present invention also provide an apparatus for constructing a set of planning methods for a power system, the apparatus comprising:

[0014] The data acquisition unit is used to acquire multi-source time-series data of the target power system, wherein the multi-source time-series data is a set of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems;

[0015] The data processing unit is used to standardize the multi-source time-series data to obtain a standardized dataset;

[0016] The risk indicator construction unit is used to construct a new risk indicator system for the target power system based on the standardized dataset and a large number of operating modes generated by time-series production simulation.

[0017] The influencing factor analysis unit is used to determine the risk causes and main influencing factors affecting the consumption of new energy based on the constructed new risk indicator system.

[0018] The data filtering unit is used for the extraction of typical high-risk moments based on the tSNE-FCM algorithm, and for the analysis of extreme operation indicators and initial screening of high-risk moments in the set of minimum planning methods.

[0019] The model building unit is used to construct a minimum planning set of security risks and weaknesses of the target power system based on the determined extreme operating mode indicators and the initial screening results of high-risk moments.

[0020] Thirdly, embodiments of the present invention also provide an apparatus for constructing a set of planning methods for a power system, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the method for constructing a set of planning methods for a power system as described in the first aspect.

[0021] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method for constructing a power system planning scheme set as described in the first aspect.

[0022] The technical solution provided in this application first constructs a safety risk and vulnerability identification index system to identify risks and vulnerabilities in the massive operational modes generated by time-series production simulation. Then, it categorizes risk modes according to safety, economy, and environmental considerations, generating typical risk modes, extreme risk modes, and deductive modes through feature aggregation. Next, it constructs a multi-type, multi-index high-dimensional coordinate system and extracts extreme operational modes based on algorithms such as fuzzy entropy weighting to meet safety verification requirements. Finally, by analyzing the impact of extreme modes on planning, and combining feature sensitivity settings and compatibility boundaries, it defines a minimum planning mode set, forming a cluster of reliable new power system planning modes suitable for the target region. This solves the technical problems of existing power system planning technologies, such as single evaluation indicators, reliance on expert experience or simple statistical methods, and low regional adaptability. It achieves efficient extraction of key risk scenarios, providing scientific, economical, and safe solution support for new power system planning. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for generating a power system operation mode scenario provided in an embodiment of this application;

[0024] Figure 2 A photovoltaic curtailment characteristic diagram considering grid transmission constraints is provided for an embodiment of the present invention;

[0025] Figure 3 This is a diagram illustrating the characteristics of photovoltaic power curtailment without considering grid transmission constraints, provided in an embodiment of the present invention.

[0026] Figure 4 This is a diagram illustrating the characteristics of wind and solar power curtailment considering grid transmission constraints, provided in an embodiment of the present invention.

[0027] Figure 5 This is a diagram illustrating the characteristics of wind and solar power curtailment without considering grid transmission constraints, provided in an embodiment of the present invention.

[0028] Figure 6 This is a diagram for identifying abnormal moments of photovoltaic curtailment under conditions where grid transmission constraints are not considered, provided in an embodiment of the present invention.

[0029] Figure 7 This is an example of identifying abnormal moments of wind and solar power curtailment without considering grid transmission constraints, provided by an embodiment of the present invention.

[0030] Figure 8 A stratified screening result diagram of the normalized wind power curtailment index provided in an embodiment of the present invention;

[0031] Figure 9 A stratified screening result diagram of the normalized photovoltaic curtailment index provided in an embodiment of the present invention;

[0032] Figure 10 A box plot showing the spatial distribution of centralized transformer power in the minimum operating mode provided in this embodiment of the invention;

[0033] Figure 11 A box plot showing the spatial distribution of concentrated line power in the minimum operating mode provided in this embodiment of the invention;

[0034] Figure 12 A box plot showing the spatial distribution of centralized photovoltaic output in the minimum operating mode provided in this embodiment of the invention;

[0035] Figure 13 A box plot showing the spatial distribution of concentrated wind power output in the minimum operating mode provided in this embodiment of the invention;

[0036] Figure 14 A box plot showing the spatial distribution of conventional unit output in the minimum operating mode provided by an embodiment of the present invention;

[0037] Figure 15 A box plot showing the spatial distribution of the centralized load level in the minimum operating mode provided in this embodiment of the invention;

[0038] Figure 16 A structural diagram of a device for constructing a set of planning methods for a power system provided in an embodiment of the present invention;

[0039] Figure 17 This is a structural diagram of a device for constructing a set of planning methods for a power system, provided in an embodiment of the present invention. Detailed Implementation

[0040] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] Figure 1 This is a flowchart of a method for constructing a planning mode set for a power system according to an embodiment of this application. The method can be executed by a device for constructing a planning mode set for a power system. The device can be implemented by software and / or hardware and can be configured in an electronic device such as a computer.

[0042] like Figure 1As shown, the technical solution provided in this application includes the following steps:

[0043] S101, acquire multi-source time-series data of the target power system. The multi-source time-series data is a collection of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems.

[0044] Optionally, S101 specifically includes:

[0045] Acquire historical active power data of wind farms and photovoltaic power plants; acquire the load curve of the entire power grid and typical regional load data of the target power system; acquire the maximum / minimum output, start-stop records, and ramp rate of thermal power units; acquire the charging power, discharging power, state of charge, and capacity parameters of energy storage systems.

[0046] Specifically, the historical active power data of wind farms and photovoltaic power stations belong to the category of new energy output data. The historical output records of wind farms and photovoltaic power stations are obtained through the dispatch and control system. The sampling time granularity is 1 hour, totaling 365 days, and the data format is an active power sequence with timestamps. The raw data is stored according to the power station category. After acquisition, it is uniformly converted into a standard structured format and initially classified and organized according to weekly, monthly, and other cycles.

[0047] Load data is extracted from the provincial dispatching system or the power grid-side load monitoring platform, including the entire network load curve and representative load data from typical regions, with a sampling granularity of 1 hour. The acquired data is stored in a database by region, and an index structure is established to support fast retrieval.

[0048] The thermal power unit operation data is collected through the energy management system to collect the regulation capacity parameters of the thermal power units. The data is archived by unit and associated with the scheduling period through a mapping table to construct a set of flexibility constraint parameters.

[0049] The energy storage system's operational data is extracted from the energy storage device's operational log data in the dispatch and monitoring system, with a sampling granularity of 1 hour. The data is archived using the energy storage unit's identification number. The collected data is managed according to the energy storage unit's identification number and will be subsequently used to construct the energy storage operational status matrix and set boundary conditions.

[0050] S102, standardize the multi-source time series data to obtain a standardized dataset.

[0051] Optionally, S102 specifically includes: aligning multi-source time-series data with different sampling frequencies to an hourly time scale to obtain a standardized dataset, wherein high-frequency data is downsampled using a segment averaging method and low-frequency data is completed using linear interpolation.

[0052] Specifically, data from different sources and sampling frequencies are time-aligned and uniformly converted to an hourly time scale to ensure consistency and comparability of various time-series data on the time axis. Missing values ​​in the data are repaired; short-term missing values ​​are imputed using the moving average method, while long-term missing segments are marked as invalid intervals and removed. Isolated outliers are replaced with the mean of nearest neighbors, and the Z-score method is used to identify outliers. Continuous outlier segments are directly removed and not included in subsequent analysis. After completing time synchronization, missing value repair, and post-outlier processing, all processed data are integrated into a unified spatial-temporal grid to obtain a standardized dataset.

[0053] Optionally, the multi-source time-series data with different sampling frequencies are uniformly aligned to an hourly time scale to obtain the standardized dataset, which includes:

[0054] For the multi-source time-series data with a sampling frequency higher than the target frequency, based on the formula... Aggregation is performed using the arithmetic mean resampling method, where x t This represents downsampled observations over a specified time period. The original sampling point observations are defined within a specified time period, where n is the number of sampling points within the specified time period, t is the sampling time, and i = 1, 2, 3, ...;

[0055] For the multi-source time-series data with a sampling frequency lower than the target frequency, based on the formula... Upsampling is performed using linear interpolation to fill in missing observations at specific time points, where y t The observed values ​​obtained through interpolation. and These are the observation values ​​at adjacent known time points;

[0056] For missing data segments, based on the formula Linear interpolation using the moving average method is employed to restore data continuity. These are estimates for the missing time points. Let n be the number of valid observation points located within the time window before and after the missing point, i = 1, 2, 3, ..., n. This represents the value at the corresponding time point;

[0057] For outliers that deviate from historical statistical patterns, based on the formula The Z-score standard deviation limit method was used for identification, where, Let μ be the observed value at time point t, μ be the mean of the corresponding variable over the same time period, and σ be the standard deviation.

[0058] It should be noted that when |Z(t)| > M, the value at the corresponding time point is considered an outlier. If at a certain time t, |Z(t)| > M, and at its two adjacent time points, |Z(t ± Δt)| ≤ M, then this point is identified as an isolated outlier, and the average of its immediate neighbors is used to repair it. The repair expression is as follows:

[0059] ;

[0060] If x consecutive time points t i The observations (i=1, 2, 3, ...) all satisfy |Z(t) i If |> M, then the segment is considered a continuous abnormal segment, which can be directly removed without interpolation repair and excluded in subsequent modeling.

[0061] S103, a new risk indicator system for the target power system is constructed based on a standardized dataset and a large number of operation modes generated by time-series production simulation.

[0062] Specifically, after acquiring and preprocessing multi-source time-series data, a system of time-based indicators for calculating the security risks of the target power system is constructed based on the massive operational modes generated by time-series production simulation. These indicators include the following:

[0063] (1) Power indicators of new energy sources

[0064] Based on output data processed with a unified time scale, this paper points out that current risk assessment indicators for wind and solar power generation systems largely follow the traditional reliability assessment indicator system. These indicators describe the system risk level before and after renewable energy integration through load shedding rate, expected power shortage, and expected power shortage amount. However, these indicators cannot reflect the individual contribution of renewable energy power plant output to system risk. Unlike traditional approaches, this invention links risk to wind and solar grid-connected electricity prices, project costs, and wind and solar power utilization hours. It measures risk from the perspective of future investment policies and renewable energy development, rather than conducting risk assessments based on operational modes.

[0065] The risk function used in conventional power system risk assessment is the expected value of power system losses, i.e., the unconditional expectation, which is defined as follows:

[0066] ;

[0067] ;

[0068] Where: X is a random variable representing a certain power system risk index; x is a value of the random variable X, i.e., a specific numerical value; p(t) is the probability density function of the risk index; Q(x) is the cumulative distribution function of the risk index; and V(X) is the risk value of the risk index.

[0069] Based on conventional risk indicators such as system-level power shortage risk and component-level overload risk and load shedding risk, a risk indicator for renewable energy power fluctuation is proposed to quantify the contribution of renewable energy access to system risk, thus expanding the risk indicator system.

[0070] Wind Power Variability Risk (WPVR) is the risk of system load loss and wind / solar curtailment caused by insufficient ramp-up capability of conventional generating units due to deviations in renewable power output from planned values, under system conditions that consider random outages of generating units and transmission lines. The risk indicators are defined as follows:

[0071] ;

[0072] ;

[0073] ;

[0074] Where n is the total number of system states; o i Let be the probability of the i-th state of the system; o Inet The system net load is the system load power. I With new energy power w The difference; o ac R represents the actual output of the conventional unit after being limited by the ramp-up capability and output boundary of the renewable energy source; R is the maximum ramp-up rate of the conventional unit; Δt is the ramp-up time; o g max and o g min These represent the maximum and minimum output of a conventional unit.

[0075] If WPVR is positive, it indicates that the system is at risk of insufficient power supply (i.e., there is a demand for load shedding); if WPVR is negative, it indicates that the system is at risk of curtailment of wind and solar power.

[0076] The calculation steps for WPVR are as follows:

[0077] The first step is to calculate the net load based on the current power output plan curve Pg and the predicted value of renewable energy power;

[0078] The second step is to determine the time and predicted wind speed corresponding to each state, obtain the conditional fluctuation value of the new energy power under each state by joint conditional sampling of k, and then update the net load curve according to the magnitude and positive or negative of the fluctuation value.

[0079] The third step is to consider the random outage of system components and refresh the power generation base value of conventional units at the initial time of each period according to the outage status of the components.

[0080] The fourth step is to calculate the unit output boundary value during the study period, based on the formula, since the ramp rate of the conventional unit is within the maximum ramp rate range and is adjustable, and the unit output can be rescheduled and adjusted at any time.

[0081] The fifth step is to calculate the WPVR value according to the formula and determine the system risk type based on its positive or negative sign.

[0082] The renewable energy power fluctuation risk index is used to assess the coordination between renewable energy power generation and the ramp-up rate of conventional generating units, and to study the system risks caused by the untimely or insufficient ramp-up capacity of conventional generating units in a short period of time. To address the risks of load shedding or wind / solar curtailment, a new power generation plan needs to be formulated, and wind curtailment operations can be implemented by activating standby generating units or coordinating with appropriate market mechanisms. This index can reflect the risk level of renewable energy power fluctuations at the system level, providing a reference for dispatchers to balance grid risks and benefits and to formulate and modify wind and solar grid connection plans.

[0083] (2) System reserve gap ratio

[0084] Spinning reserve capacity serves as the first line of defense for ensuring power system frequency stability, and its supply-demand balance directly impacts the system's ability to respond to sudden disturbances such as generator trips and load spikes. This indicator, by precisely quantifying the supply-demand gap of spinning reserve resources, can not only assess the short-term dynamic stability of the system but also reflect the rationality of grid frequency regulation resource allocation. Its definition is as follows:

[0085] ;

[0086] In the formula, O DEFICIENY P represents the proportion of the spinning reserve demand gap. DEMAND To meet the system's spin-off backup requirements, P ACTUAL This is a backup for rotation that the system actually needs to meet.

[0087] (3) System standby gap ratio

[0088] As a crucial support for the long-term regulation capacity of a power system, the supply-demand balance of outage reserves directly affects the system's regulatory margin in responding to issues such as sustained load fluctuations, planned unit maintenance, and fuel supply interruptions. This indicator, by quantifying the actual supply-demand gap in outage reserves, can effectively assess the system's power supply reliability over medium- to long-term timescales. Its definition is as follows:

[0089] ;

[0090] In the above formula, O SHUTDOWN_DEFICIENY The proportion of the demand gap for standby power outages, P SHUTDOWN_DEMAND For system downtime backup requirements, P SHUTDOWN_ACTUAL This is for the actual shutdown backup required by the system.

[0091] (4) System main transformer over-limit rate

[0092] As the core hub for energy conversion between different voltage levels in a power system, the main transformer's overload operation can lead to a significant increase in equipment losses and may also cause faults such as insulation aging and localized overheating, seriously threatening power supply reliability. As a critical link connecting the main grid and the load side, the main transformer's over-limit status directly reflects the capacity bottleneck of the transmission and transformation system. Its over-limit rate is defined as follows: ;

[0093] In the above formula, O OUTOFLIMIT Main variable over-limit rate, P OUTOFLIMIT Main transformer over-limit power, P CAPACITY Main variable capacity.

[0094] (5) Failure probability

[0095] Failure probability refers to the probability that a power system will fail or exceed its safe operating range when faced with various uncertainties. For example, the uncertainty of new energy power generation may lead to grid instability, while improper load and hydropower dispatch may cause power outages.

[0096] ;

[0097] Where: O(Wind) is the probability of system failure caused by wind energy fluctuations; O(Hydro) is the probability of failure caused by improper hydropower dispatch; and O(Grid Failure) is the probability of failure caused by grid equipment failure.

[0098] Failure probability helps power systems understand the likelihood of system failure under various uncertainties. For example, when grid equipment is old or poorly maintained, the probability of equipment failure is higher; in regions heavily reliant on wind and solar power, if the output of new energy sources fluctuates significantly, the probability of system failure will also increase accordingly. By calculating failure probabilities, power companies can strengthen the monitoring and optimization of potential failure points, thereby reducing the incidence of failure events.

[0099] S104. Based on the newly constructed risk indicator system, the causes of risks affecting the consumption of new energy sources and the main influencing factors are identified.

[0100] Optionally, S104 specifically includes:

[0101] An autoencoder model is constructed; based on the autoencoder model, the autoencoder reconstruction error analysis method is used to calculate the errors at different levels, providing fine-grained data for subsequent analysis; the simulation results based on time series are output to determine the time series characteristics of curtailment of wind power and photovoltaic power generation under different constraints.

[0102] Specifically, after constructing and simulating the indicator system for calculating the timing of power system security risks, it is necessary to analyze the risk causes and main influencing factors affecting the absorption of new energy. The capacity for new energy absorption is influenced by a combination of factors, with the most significant constraints stemming from insufficient peak-shaving capacity on the power supply side and bottlenecks in grid transmission capacity. The autoencoder reconstruction error analysis method is a highly practical approach. It does not rely on a large amount of labeled data, can automatically learn the characteristics of normal power generation modes, and identify anomalies (wind / solar curtailment is essentially an abnormal mode where "actual output is far lower than theoretical output") through reconstruction errors. Furthermore, it is highly adaptable to high-dimensional, nonlinear power data and is an effective sampling technique capable of generating the necessary samples from complex probability distributions of new energy power data.

[0103] 1. Construct an autoencoder model.

[0104] Clarify the mapping relationship between the encoder and decoder to provide a model basis for error calculation.

[0105] Encoder: Maps input data to a low-dimensional latent space, denoted as f. θ :R d →R k ( As a potential dimension, (For encoder parameters), output latent vector .

[0106] Decoder: Reconstructs the latent vectors into input space data, denoted as g. Φ :R k →R d ( (For decoder parameters), output reconstructed samples .

[0107] 2. Reconstruct the error function.

[0108] Reconstruction error is a measure of the difference between the input sample and the reconstructed sample. An appropriate loss function must be selected based on the data type. Commonly used error functions include the following:

[0109] (1) The error of a single sample is the average of the squared differences along the feature dimension: ;

[0110] in, It is a sample The j-th feature, These are the corresponding features of the reconstructed sample.

[0111] (2) Cross-entropy (CE, applicable to discrete / binary data, such as binary images and text):

[0112] For binary data Or [0,1], the error of a single sample is:

[0113] (Note: This should be avoided) In actual calculations, a small amount is often added. ).

[0114] (3) Total training loss.

[0115] The average error across the entire dataset, used for parameter optimization during model training: Where L is the single sample error function such as MSE or CE mentioned above.

[0116] 3. Multi-granularity calculation of reconstruction error.

[0117] Calculate errors at different levels (sample level, feature level) to provide fine-grained data for subsequent analysis.

[0118] (1) Sample-level error

[0119] The reconstruction error of a single sample (defined in step 2) is used to evaluate the reconstruction quality of a single sample:

[0120] ;

[0121] (2) Characteristic level error

[0122] The reconstruction error of a single feature dimension is used to analyze the model's ability to capture different features:

[0123] For MSE loss, the error of the j-th feature (average across all samples): ;

[0124] For the CE loss, the error of the j-th feature (average across all samples): ;

[0125] (3) Global error

[0126] The average reconstruction error of the entire dataset is used to evaluate the overall performance of the model. .

[0127] 4. Analysis of new power-related influencing factors.

[0128] In this embodiment of the invention, based on the time series simulation results, the curtailment time series characteristics of wind power and photovoltaic power generation under different constraints are systematically compared and analyzed, aiming to reveal the main constraint mechanisms for the consumption of new energy in the current regional power grid. Figures 2-5 The time-by-time curtailment characteristics of wind and solar power generation are demonstrated under two operating conditions: considering grid transmission constraints and not considering grid transmission constraints. Figure 2 and Figure 4 To account for the curtailment time series curves of photovoltaic and wind power under grid transmission constraints, Figure 3 and Figure 5 This corresponds to the simulation results without considering grid transmission constraints. As can be seen from the figure, both photovoltaic (PV) and wind power exhibit significant curtailment when grid transmission constraints are considered. PV curtailment is mainly characterized by high peak values ​​and frequent curtailment periods, with the maximum curtailment in some areas approaching 2500MW, and varying degrees of curtailment occurring for most hours throughout the year. Wind power curtailment is mainly concentrated in small amounts over many hours, with the maximum curtailment within the hundreds of megawatts, but occurring much more frequently. This indicates that under limited grid transmission capacity, the absorption capacity of new energy sources is significantly suppressed, resulting in widespread and long-term curtailment. In contrast, without considering grid transmission constraints, the curtailment of PV and wind power is significantly reduced. Figures 2-5 It is evident that, regardless of whether it is photovoltaic or wind power, without considering grid transmission constraints, the amount of wasted power is close to zero for the vast majority of periods, with only a small amount of localized wasted power occurring in very few periods. This result indicates that grid transmission capacity has a crucial impact on the level of renewable energy absorption, and grid bottlenecks have become one of the main physical constraints currently limiting the full utilization of wind and solar power.

[0129] Further analysis, such as Figures 6-7 As shown, this paper presents the identification results of abnormal wind and solar power curtailment at specific times, without considering grid transmission constraints. To more intuitively reflect extreme curtailment behavior within the system, this application employs the quantile threshold method to detect anomalies in the wind and solar power curtailment time-series data. Figure 6 It can be seen that abnormal photovoltaic curtailment is mainly concentrated in specific high-incidence periods, with extremely high peak curtailment levels, exceeding 12,000 MW for the entire province at some moments. The temporal distribution of these anomalies shows a clear early-stage concentration, mainly occurring in the first 3,000 hours of the year, while their frequency decreases significantly in subsequent periods. This characteristic indicates that even under conditions where grid transmission constraints are completely removed, photovoltaic systems are still affected by source-load mismatch and system peak-shaving capacity constraints during certain seasonally high-output phases, leading to severe curtailment. The concentration and seasonal variation of photovoltaic output further exacerbate the periodic concentration of these anomalies.

[0130] Figure 7This reflects the abnormal curtailment characteristics of wind power. The abnormal curtailment times of wind power are more dispersed than those of solar power, with anomalies covering different time periods throughout the year. While the peak anomalies are not as extreme as those of solar power, there are still curtailment values ​​exceeding 2000MW. The volatility of wind power output determines that its curtailment anomalies have greater uncertainty and seasonal dispersion. The above analysis shows that although grid transmission capacity constraints have been completely removed, wind and solar power systems still experience significant abnormal curtailment times, fully demonstrating that the bottleneck in peak-shaving capacity within the system remains significant. Especially during periods of highly concentrated renewable energy output, system regulation resources cannot effectively follow renewable energy fluctuations, resulting in a significant deficiency in renewable energy absorption capacity during certain periods.

[0131] S105, a typical high-risk moment extraction method based on the tSNE-FCM algorithm, analyzes the extreme operation mode indicators and high-risk moment screening of the minimum planning mode set.

[0132] Optionally, S105 specifically includes: normalizing the multidimensional risk indicators of the target power system; using a typical high-risk moment extraction method based on the tSNE-FCM algorithm to analyze the extreme operation mode indicators and high-risk moments of the minimum impact planning mode set.

[0133] Specifically, Figures 6-7 The analysis revealed typical periods of abnormal power curtailment and their main influencing factors, while also exposing potential weaknesses in the system under operating conditions. To further extract representative high-risk scenarios from the year-round operating data, this invention, based on a multi-dimensional risk indicator system constructed using S104, conducts hourly evaluations of the time-series production simulation results at 8760 time points. By setting risk thresholds for key indicators such as wind and solar power curtailment, equipment overload levels, and system reserve gaps, high-risk moments with significantly exceeded limits are identified. These high-risk moments will serve as the basis for subsequent uncertainty disturbance modeling and operational scenario generation, ensuring that the constructed scenarios cover various boundary operating states the system may face. The specific steps are as follows:

[0134] (1) Normalization of multidimensional risk indicators of power system.

[0135] To accurately identify high-risk states during power system operation, based on the established operational risk indicator system and relying on time-series production simulation results, the values ​​of various indicators are calculated hourly, covering core indicators such as renewable energy absorption capacity, system regulation margin, equipment physical load, and network transmission bottlenecks, thereby constructing a time-level operational state vector. This serves as the input basis for subsequent risk scoring and cluster analysis.

[0136] Because the dimensions and orders of magnitude of the various indicators differ significantly, direct aggregation or scoring would lose uniformity and comparability. Therefore, this study employs interval normalization to treat all indicators as dimensionless. The normalization method is shown below:

[0137] ;

[0138] in, This represents the original value of the p-th index at time t. and These are the minimum and maximum values ​​of the indicator across 8760 points throughout the year, respectively. The values ​​obtained after normalization are... This indicates the relative risk intensity of this dimension indicator at that moment.

[0139] For inverse indicators such as reserve capacity, where "lower values ​​indicate higher risk," a reverse normalization method is used, namely:

[0140] ;

[0141] This ensures that all indicators are within the normalized space, with larger values ​​indicating higher risk levels, thus facilitating unified measurement. To avoid distortion of the overall normalized distribution due to extreme values, the upper bound of some indicators (such as abandoned electricity) adopts a statistical quantile correction strategy, using the 95th quantile as the upper bound for normalization, thereby enhancing the model's robustness and generalization ability. The normalized vector is denoted as: .

[0142] (2) A method for extracting typical high-risk moments based on hierarchical criteria and cluster compression.

[0143] To identify typical high-risk moments with both strong risk characteristics and representativeness from year-round time-series production simulation data, this invention proposes a two-stage extraction method based on hierarchical criteria and cluster compression. The aim is to obtain a set of typical high-risk moments that are physically representative and structurally distinguishable in the indicator space. First, for the normalized multidimensional risk indicator sequence, a quantile-based hierarchical screening strategy is constructed. Based on the degree of anomaly of each indicator, the moments throughout the year are divided into three categories: mild (90–95%), moderate (95–99%), and extreme anomaly (>99%), initially eliminating normal periods with insignificant risk characteristics. This mechanism preserves the potential boundary states of the system while ensuring reasonable coverage of various typical risk patterns in the sample space.

[0144] Subsequently, to further reduce information redundancy in the candidate high-risk moment set and improve the representativeness of the sample structure, a two-stage identification method integrating tSNE nonlinear dimensionality reduction and fuzzy C-means clustering (FCM) was introduced. While preserving the risk distribution structure of multiple indicators, the tSNE algorithm maps the normalized indicator matrix to a low-dimensional separable space, significantly improving the pattern separability under high-dimensional indicator combinations. Then, FCM clustering is introduced in the projection space to obtain soft membership relationships between samples, accurately characterizing the risk transition state driven by multiple mechanisms. Finally, representative samples with the highest membership degrees are selected from each cluster to construct a typical high-risk moment set with balanced structural distribution and diverse risk characteristics, providing crucial foundational support for the subsequent generation of uncertain risk scenarios.

[0145] A. A hierarchical identification method for high-risk candidate moments.

[0146] To systematically identify operational risk states of varying degrees, this invention sets three-tiered quantile thresholds based on the statistical distribution characteristics of each indicator in the annual data. , , It is used to classify the anomaly level corresponding to each moment, aiming to cover multiple operating conditions from mild boundary disturbances to extreme risk shocks, and to retain early warning signals of system deviations from normal.

[0147] The set of high-risk candidate moments is defined as follows: ;

[0148] Among them, if Determined as mildly abnormal. The diagnosis was moderately abnormal. These are identified as extreme anomalies. These three types of anomalies will serve as diverse candidate risk samples, undertaking tasks such as normal boundary detection, medium-intensity disturbance analysis, and extreme condition simulation in subsequent analyses, thereby enhancing the breadth and depth of risk identification.

[0149] To more intuitively demonstrate the application effect of the stratified criteria in time series data of indicators, Figure 8 and Figure 9 The paper presents the annual time-series distribution of wind and solar curtailment indicators after normalization, along with the identification results of high-value segments. The two types of indicators exhibit different operational characteristics: wind curtailment fluctuates frequently with relatively flat amplitudes, mainly concentrated during typical periods of insufficient system regulation margin; while solar curtailment shows concentrated peaks with significant amplitudes, reflecting its high coupling with daytime operation scenarios characterized by low load and concentrated irradiance. The figures show only typical examples of candidate indicators. In the actual screening process, hierarchical criteria will be applied simultaneously across multiple normalized indicator dimensions to construct a diverse set of high-risk candidate moments with varying risk sources, laying the foundation for subsequent clustering compression and scenario generation.

[0150] B. A method for extracting typical high-risk moments based on the tSNE-FCM algorithm.

[0151] Dimensionality reduction algorithms for high-dimensional data mainly include linear and nonlinear methods. Traditional linear methods offer faster dimensionality reduction, but significant information loss occurs when generating low-dimensional data. tSNE (t-Distributed Stochastic Neighbor Embedding, a nonlinear technique for high-dimensional data dimensionality reduction and visualization) is a nonlinear dimensionality reduction method. Compared to traditional dimensionality reduction algorithms such as principal component analysis, tSNE has a stronger ability to capture nonlinear relationships in the data, more realistically reflecting the distribution of complex high-dimensional data, and more completely preserving the local structure of the original data. That is, data points that are close to each other in the high-dimensional space remain close after being mapped to the low-dimensional space. The main idea of ​​tSNE is to characterize the similarity between sample points through the joint probability between two sample points, and finally obtain the optimal low-dimensional sample points by minimizing the KL divergence.

[0152] Fuzzy C-Means (FCM) is an unsupervised learning algorithm developed from the traditional K-means clustering algorithm. As an extension of K-means, it differs from the hard clustering method of K-means by introducing the concept of fuzzy membership to achieve flexible fuzzy partitioning. Therefore, it has greater flexibility and robustness, and is widely used among many clustering algorithms with excellent clustering results. Specifically, the core idea of ​​FCM is to divide all samples into several fuzzy subsets. Each sample has a membership degree with each fuzzy subset, which is used to characterize the similarity between the sample and the subset. By calculating the membership degree values ​​of a sample to each subset, the category to which the sample belongs is determined, thereby achieving data clustering. Its advantage lies in its ability to handle the fuzzy classification problem of sample points and more accurately characterize the fuzzy classification relationship between samples.

[0153] S106, based on the determined extreme operating mode indicators and the initial screening results of high-risk moments, constructs the minimum planning method set of the security risks and weak modes of the target power system.

[0154] Optionally, S106 specifically includes: combining the parallel computing capabilities of the Spark distributed data processing framework, using a self-paced deep clustering (SPDC) algorithm to efficiently cluster and automatically classify high-risk operating moment data; performing box plot analysis on the power spatial distribution of various devices in a set number of minimum operating modes extracted by clustering, and constructing a minimum planning mode set for the safety risks and weaknesses of the target power system.

[0155] The preferred setting is 10. Under the various boundary operating states that the scenario coverage system constructed in steps S101-S105 may face, a minimum planning method set for new power system security risks and weaknesses is defined. For example, the reliable method set for the target power system planning of a certain province is taken as an example.

[0156] 1. A parallel deep clustering method based on Spark.

[0157] Currently, many methods in power system state analysis and typical operating condition identification still rely on supervised learning frameworks, which often require a large number of manually labeled samples. However, in actual operating data, the operating moments often lack clear category information, making supervised models difficult to apply. To address this issue, this invention combines the parallel computing capabilities of the Spark distributed data processing framework and proposes embedding a self-paced deep clustering (SPDC) algorithm based on self-paced learning into the Spark platform. This enables efficient clustering and automated classification of large-scale, high-risk operating moment data, thereby constructing a representative minimal set of operating modes without the need for prior label information. The specific steps are as follows:

[0158] (1) Data preprocessing and partitioning. Preprocessing operations include removing outlier data, filling in missing data, and normalization. Data partitioning adopts a stratified sampling strategy, which extracts preprocessed load training samples proportionally, divides the dataset into multiple data blocks, and ensures that the load categories and quantities of each sample block are similar. These data blocks are then distributed to multiple nodes of the Spark cluster for parallel processing.

[0159] (2) Construction of deep clustering model. In the TensorFlow on Spark environment, the computation graph was constructed using the TensorFlow API. The model was trained in two stages: first, the synthetic encoder was pre-trained to obtain low-dimensional features. This process used the Adam optimizer (learning rate 0.001, batch size 128) for 500 iterations, and the autoencoder parameters were adjusted using the proposed loss function; then, clustering optimization was performed using the Adam optimizer with a learning rate of 0.001 and a maximum number of iterations of 2000.

[0160] (3) Spark Cluster Construction. Spark clusters are constructed by defining the TFConfig, TFNode, and TFCluster functions: The TFConfig function is used to define the configuration information of the TensorFlow cluster, including the number of nodes, node types, and the address and port of the parameter server nodes; the TFNode function is used to start each node in the TensorFlow cluster, and each TFNode contains information such as the node's role, number, address, and port; the TFCluster function is used to embed the TensorFlow graph into the Spark computation graph, which will create the TensorFlow cluster, add TensorFlow nodes, and pass the defined parameters.

[0161] (4) Distributed Network Training. Each Worker node starts one or more Executors, each containing an SPDC model. The Driver node acts as a parameter server, responsible for transmitting SPDC model parameters: initializing SPDC model parameters and distributing them to the SPDC models of the Executors in each Worker node; each Worker node independently trains on the allocated data to update its parameters, and obtains new parameters for its respective model after training. When the training reaches a certain number of iterations, each Worker node feeds back the latest parameters of its SPDC model to the Driver node. The Driver parameter server aggregates all parameters and calculates the average parameter value, then distributes it to the SPDC models of each Worker node to update the parameters. After the training stopping condition is met, the training result of each base clusterer is represented as (where is the data and is the label category). The results of each base clusterer are merged and output to the HDFS file system using the ReduceByKey function.

[0162] For example, in the previous step, this invention utilizes deep clustering to extract a minimal set of typical operating modes covering different boundary conditions, thereby reducing scenario redundancy and highlighting key risk characteristics. To more intuitively depict the differences between these typical operating states, Figures 10-15 Box plot analysis was performed on the power spatial distribution of various equipment in the 10 minimum operating modes extracted by clustering, providing a basis for identifying the output range, spatial dispersion and potential bottleneck characteristics of various equipment at typical high-risk moments.

[0163] Depend on Figures 10-15It is evident that the transformer power distribution is relatively compact, with median power levels close at different typical times and moderate dispersion, indicating that the output levels of equipment in each substation are relatively balanced and no obvious overload phenomenon has occurred. This suggests that the load of each substation in the network is relatively reasonable under the extracted minimum operating mode. In contrast, the power distribution of line flows exhibits greater dispersion, especially with some lines approaching full load at certain typical times, indicating that some lines may form local transmission bottlenecks. This is consistent with the key boundary condition characteristics obtained from the aforementioned clustering.

[0164] For renewable energy output, photovoltaic (PV) output exhibits significant spatial dispersion, with some sites operating at near full capacity while others experience low output. This characteristic corresponds to meteorological variations and grid constraints, highlighting the uneven distribution of PV power. In contrast, wind power output shows relatively smaller spatial fluctuations, with median output values ​​across sites being similar. This reflects a more consistent spatial characteristic of wind power output at typical times, implying a more balanced impact of wind power-side constraints in the extracted boundary conditions.

[0165] Meanwhile, the output levels of conventional generating units are distributed over a narrow range, with similar median outputs across units. This indicates that each unit undertakes a relatively similar output task under boundary operating conditions, and there are no units with long-term overload. This suggests that the distribution of power among conventional generating units in these boundary conditions is relatively reasonable. However, the load side exhibits significant spatial dispersion, with some nodes showing higher power levels. This is directly related to the differences in electricity demand across regions, highlighting the impact of user-side distribution characteristics on the boundary operating condition features.

[0166] In summary, the minimum set of operating modes generated by clustering can fully preserve the spatial distribution characteristics of key boundary conditions in the original large-scale scenario, providing strong support for identifying vulnerable links in the network, assessing spatial differences in equipment output, and formulating corresponding operation and investment strategies.

[0167] In summary, this invention effectively solves multiple challenges in the planning of new power systems by constructing a scientific risk identification and planning method set generation system, and has the following beneficial effects:

[0168] (1) Breaking through the limitations of traditional single-dimensional assessment: Constructing a multi-dimensional indicator system covering safety, economy, and greenness. This system can not only quantify safety risks such as line overload and voltage over-limit, but also incorporate economic parameters such as cost per kilowatt-hour and return on investment, while taking into account green indicators such as carbon emission intensity, so as to achieve multi-objective synergistic consideration and avoid planning schemes from emphasizing one dimension and neglecting other key elements, thus fundamentally solving the problem of the one-sidedness of traditional assessment.

[0169] (2) Improve planning efficiency: The defined minimum planning method set significantly reduces the scale of scenarios while fully covering risks through feature sensitivity settings and compatibility boundary constraints. Compared with the traditional planning method set, the number of scenarios is reduced by 50%-70%, which not only reduces the computational complexity of planning, but also saves a lot of manpower and time costs, while avoiding resource waste caused by scenario redundancy.

[0170] (3) Regional adaptability: The reliable power supply cluster is designed specifically for Jiangsu Province, precisely matching its characteristics of high load density, centralized grid connection of new energy sources, and high dependence on inter-regional power transmission. Through customized scenario aggregation and risk simulation, the power supply layout and grid structure can be optimized while ensuring the safety and stability of the power grid, thereby improving the economic efficiency of the planning scheme by more than 15%, and providing a replicable and efficient paradigm for the planning of new power systems in similar load center areas.

[0171] (4) Scene extraction: Significantly reduces interference from subjective human factors. By constructing a high-dimensional coordinate system and allocating objective weights, extreme scenes that have a significant impact on planning can be accurately identified from massive operational data, ensuring that safety verification covers the most critical risk points. Compared with the traditional experience-based screening method, the scene capture accuracy is improved by more than 30%, providing solid support for the reliability of the planning scheme.

[0172] Figure 16 This is a structural diagram of a device for constructing a set of planning methods for a power system, provided in an embodiment of the present invention.

[0173] like Figure 16 As shown, the device includes:

[0174] The data acquisition unit 161 is used to acquire multi-source time-series data of the target power system. The multi-source time-series data is a set of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems.

[0175] Data processing unit 162 is used to standardize multi-source time series data to obtain a standardized dataset;

[0176] Risk indicator construction unit 163 is used to construct a new risk indicator system for the target power system based on a standardized dataset and a large number of operating modes generated by time-series production simulation.

[0177] The influencing factor analysis unit 164 is used to determine the risk causes and main influencing factors affecting the consumption of new energy based on the constructed new risk index system.

[0178] Data filtering unit 165 is used for the extraction method of typical high-risk moments based on the tSNE-FCM algorithm, and to analyze the extreme operation mode indicators and high-risk moments of the minimum planning mode set.

[0179] Model building unit 166 is used to construct a minimum planning set of security risks and weaknesses of the target power system based on the determined extreme operating mode indicators and the initial screening results of high-risk moments.

[0180] Optionally, the data acquisition unit 161 is specifically used for:

[0181] Obtain historical active power data for wind farms and photovoltaic power plants;

[0182] Obtain the load curve of the entire power grid and load data of typical areas of the target power system;

[0183] Obtain the maximum / minimum output, start-stop records, and ramp rate of the thermal power unit;

[0184] Obtain the charging power, discharging power, state of charge, and capacity parameters of the energy storage system.

[0185] Optionally, the data processing unit 162 is specifically used for:

[0186] The multi-source time-series data with different sampling frequencies are uniformly aligned to an hourly time scale to obtain the standardized dataset. For high-frequency data, segment averaging is used for downsampling, and for low-frequency data, linear interpolation is used for completion.

[0187] Optionally, the data processing unit 162 is further specifically used for:

[0188] For the multi-source time-series data with a sampling frequency higher than the target frequency, based on the formula... Aggregation is performed using the arithmetic mean resampling method, where x t This represents downsampled observations over a specified time period. The original sampling point observations are defined within a specified time period, where n is the number of sampling points within the specified time period, t is the sampling time, and i = 1, 2, 3, ...;

[0189] For the multi-source time-series data with a sampling frequency lower than the target frequency, based on the formula... Upsampling is performed using linear interpolation to fill in missing observations at specific time points, where y t The observed values ​​obtained through interpolation. and These are the observation values ​​at adjacent known time points;

[0190] For missing data segments, based on the formula Linear interpolation using the moving average method is employed to restore data continuity. These are estimates for the missing time points. Let n be the number of valid observation points located within the time window before and after the missing point, i = 1, 2, 3, ..., n. This represents the value at the corresponding time point;

[0191] For outliers that deviate from historical statistical patterns, based on the formula The Z-score standard deviation limit method was used for identification, where, Let μ be the observed value at time point t, μ be the mean of the corresponding variable over the same time period, and σ be the standard deviation.

[0192] Optionally, the influencing factor analysis unit 164 is specifically used for:

[0193] Construct an autoencoder model;

[0194] Based on the autoencoder model, the autoencoder reconstruction error analysis method is used to calculate the error at different levels, providing fine-grained data for subsequent analysis;

[0195] Output time-series-based simulation results to determine the time-series characteristics of curtailment of wind and solar power under different constrained conditions.

[0196] Optionally, the data filtering unit 165 is specifically used for:

[0197] The multidimensional risk indicators of the target power system are normalized.

[0198] A method for extracting typical high-risk moments based on the tSNE-FCM algorithm is used to analyze the extreme operation indicators of the minimum planning method set and to screen high-risk moments.

[0199] Optionally, the model building unit 166 is specifically used for:

[0200] By combining the parallel computing capabilities of the Spark distributed data processing framework, a self-learning deep clustering algorithm is used to efficiently cluster and automatically classify data from high-risk runtime events.

[0201] Box plot analysis was performed on the power spatial distribution of various equipment in a set number of minimum operating modes extracted by clustering, and a minimum planning mode set of security risks and weak points of the target power system was constructed.

[0202] The apparatus for constructing a power system planning method set provided in this embodiment of the invention has the same technical features as the method for constructing a power system planning method set provided in the above embodiments, so it can also solve the same technical problems and achieve the same technical effects.

[0203] like Figure 17As shown in the illustration, this application embodiment also provides a device for constructing a power system planning method set, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0204] Memory 113 is used to store computer programs;

[0205] In one embodiment of this application, when the processor 111 executes a program stored in the memory 113, it implements the method provided in any of the foregoing method embodiments, including:

[0206] Acquire multi-source time-series data of the target power system. The multi-source time-series data is a collection of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems.

[0207] Multi-source time-series data are standardized to obtain a standardized dataset.

[0208] A novel risk indicator system for the target power system is constructed based on standardized datasets and massive operational data generated by time-series production simulations.

[0209] Based on a new risk indicator system, the causes of risks affecting the consumption of new energy sources and the main influencing factors are identified.

[0210] A method for extracting typical high-risk moments based on the tSNE-FCM algorithm is used to analyze the extreme operation indicators of the minimum planning method set and to screen high-risk moments.

[0211] Based on the determined extreme operating mode indicators and the initial screening results of high-risk moments, a minimum planning method set for the safety risks and weaknesses of the target power system is constructed.

[0212] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method provided in any of the foregoing method embodiments.

[0213] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0214] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0215] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A method for constructing a set of planning methods for a power system, characterized in that, The method includes: Acquire multi-source time-series data of the target power system, wherein the multi-source time-series data is a set of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems; The multi-source time-series data is standardized to obtain a standardized dataset; Based on the standardized dataset, a new risk indicator system for the target power system is constructed using massive operational modes generated by time-series production simulation. Based on the novel risk indicator system, the causes of risks affecting the consumption of new energy sources and the main influencing factors are identified. A method for extracting typical high-risk moments based on the tSNE-FCM algorithm is used to analyze the extreme operation indicators of the minimum planning method set and to screen high-risk moments. Based on the determined extreme operating mode indicators and the initial screening results of high-risk moments, a minimum planning method set for the safety risks and weaknesses of the target power system is constructed.

2. The method for constructing a power system planning method set according to claim 1, characterized in that, Acquiring multi-source time-series data of the target power system includes: Obtain historical active power data for wind farms and photovoltaic power plants; Obtain the load curve of the entire power grid and load data of typical areas of the target power system; Obtain the maximum / minimum output, start-stop records, and ramp rate of the thermal power unit; Obtain the charging power, discharging power, state of charge, and capacity parameters of the energy storage system.

3. The method for constructing a power system planning method set according to claim 1, characterized in that, The multi-source time-series data is standardized to obtain a standardized dataset, which includes: The multi-source time-series data with different sampling frequencies are uniformly aligned to an hourly time scale to obtain the standardized dataset. For high-frequency data, segment averaging is used for downsampling, and for low-frequency data, linear interpolation is used for completion.

4. The method for constructing a power system planning method set according to claim 3, characterized in that, The multi-source time-series data with different sampling frequencies are uniformly aligned to an hourly time scale to obtain the standardized dataset, which includes: For the multi-source time-series data with a sampling frequency higher than the target frequency, based on the formula... Aggregation is performed using the arithmetic mean resampling method, where x t This represents downsampled observations over a specified time period. The original sampling point observations are defined within a specified time period, where n is the number of sampling points within the specified time period, t is the sampling time, and i = 1, 2, 3, ...; For the multi-source time-series data with a sampling frequency lower than the target frequency, based on the formula... Upsampling is performed using linear interpolation to fill in missing observations at specific time points, where y t The observed values ​​obtained through interpolation. and These are the observation values ​​at adjacent known time points; For missing data segments, based on the formula Linear interpolation using the moving average method is employed to restore data continuity. These are estimates for the missing time points. Let n be the number of valid observation points located within the time window before and after the missing point, i = 1, 2, 3, ..., n. This represents the value at the corresponding time point; For outliers that deviate from historical statistical patterns, based on the formula The Z-score standard deviation limit method was used for identification, where, Let t be the observed value at time point t, μ be the mean of the corresponding variable over the same time period, and σ be the standard deviation.

5. The method for constructing a power system planning method set according to claim 1, characterized in that, Based on the newly constructed risk indicator system, the causes and main influencing factors affecting the absorption of new energy sources are identified as follows: Construct an autoencoder model; Based on the autoencoder model, the autoencoder reconstruction error analysis method is used to calculate the errors at different levels, providing fine-grained data for subsequent analysis; Output time-series-based simulation results to determine the time-series characteristics of curtailment of wind and solar power under different constrained conditions.

6. The method for constructing a power system planning method set according to claim 1, characterized in that, A method for extracting typical high-risk moments based on the tSNE-FCM algorithm is used to analyze the extreme operating mode indicators and initial screening of high-risk moments for the set of minimum planning methods: The multidimensional risk indicators of the target power system are normalized. A method for extracting typical high-risk moments based on the tSNE-FCM algorithm is used to analyze the extreme operation indicators of the minimum planning method set and to screen high-risk moments.

7. The method for constructing a power system planning method set according to claim 1, characterized in that, Based on the determined extreme operating mode indicators and the initial screening results of high-risk moments, the minimum planning method set for the security risks and weaknesses of the target power system is constructed, including: By combining the parallel computing capabilities of the Spark distributed data processing framework, a self-learning deep clustering algorithm is used to efficiently cluster and automatically classify data from high-risk runtime events. Box plot analysis was performed on the power spatial distribution of various equipment in a set number of minimum operating modes extracted by clustering, and a minimum planning mode set of the security risks and weaknesses of the target power system was constructed.

8. A device for constructing a set of planning methods for a power system, characterized in that, The device includes: The data acquisition unit is used to acquire multi-source time-series data of the target power system, wherein the multi-source time-series data is a set of data collected from wind farms, photovoltaic power plants, power system loads, thermal power units, and energy storage systems; The data processing unit is used to standardize the multi-source time-series data to obtain a standardized dataset; The risk indicator construction unit is used to construct a new risk indicator system for the target power system based on the standardized dataset and a large number of operating modes generated by time-series production simulation. The influencing factor analysis unit is used to determine the risk causes and main influencing factors affecting the consumption of new energy based on the constructed new risk indicator system. The data filtering unit is used for the extraction of typical high-risk moments based on the tSNE-FCM algorithm, and for the analysis of extreme operation indicators and initial screening of high-risk moments in the set of minimum planning methods. The model building unit is used to construct a minimum planning set of security risks and weaknesses of the target power system based on the determined extreme operating mode indicators and the initial screening results of high-risk moments.

9. A device for constructing a set of planning methods for a power system, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method for constructing a planning mode set for a power system as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed in a computer, causes the computer to perform the method for constructing a set of planning methods for a power system as described in any one of claims 1-7.

Images