A power grid intelligent dispatching method based on multi-source data fusion

Abstract

This invention discloses a smart grid dispatching method based on multi-source data fusion, belonging to the field of grid data acquisition and control. This invention senses and analyzes the type and coupling strength of broadband oscillation modes, decides to implement a broadband feature selective retention strategy and grid broadband data processing to retain key broadband dynamic features or filter out redundant oscillation signals. Then, based on the retained broadband dynamic features and power frequency steady-state data, deep multi-source data fusion is performed, and load and renewable energy output probability predictions are made based on the fusion results. Subsequently, a reliability assessment is performed based on the prediction results. If the assessment is satisfactory, the prediction results are input into a multi-objective optimization dispatching model to generate the corresponding dispatching strategy; if the assessment is unsatisfactory, a prediction reliability insufficient warning is sent. This solves the problems of data distortion, accumulated prediction deviations, and discrepancies between grid dispatching strategies and actual dynamic operation caused by the filtering of broadband oscillation signals in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of power grid data acquisition and control technology, and in particular to a power grid intelligent scheduling method based on multi-source data fusion. Background Technology

[0002] With the deepening of the global energy transition, a high proportion of new energy sources such as wind power and photovoltaics are being connected to the grid on a large scale, and new loads such as electric vehicles and virtual power plants are growing rapidly. The structure of various links in the power grid is becoming increasingly complex. The traditional dispatch mode, which relies on manual experience and is driven by plans, is no longer suitable for the needs of power grid operation under the new situation. The randomness and intermittency of new energy output have led to a surge in the difficulty of power grid power balance. The increased interactivity on the load side has placed higher demands on dispatch response speed. At the same time, the need for coordinated regulation of multiple objectives such as safe and stable operation of the power grid and consumption of new energy is becoming increasingly urgent. Therefore, carrying out research on intelligent dispatch of the power grid has become an inevitable choice to ensure the safe and efficient operation of the power grid and promote the clean and low-carbon development of energy.

[0003] The main implementation process of existing smart grid dispatch is as follows: First, collect multi-source heterogeneous data such as steady-state grid operation data (e.g., voltage, current, power), user load data (e.g., residential electricity load, industrial production load), new energy power station output data (e.g., wind turbine speed, energy storage power station charging and discharging power), and meteorological monitoring data (e.g., wind speed, solar intensity, temperature). Next, preprocess and fuse the collected multi-source heterogeneous data. Data purification and standardization are performed through operations such as outlier detection, missing value imputation, and spatiotemporal alignment. Based on algorithms such as statistical fusion and machine learning fusion, the scattered multi-source heterogeneous data are integrated into a fused dataset. Then, based on the fused dataset, intelligent analysis and modeling are carried out to construct predictive models for load and new energy output, predicting load demand and new energy power generation capacity in the future. Subsequently, with grid security and stability, maximizing new energy consumption, and environmental compliance as core objectives, multi-objective optimization scheduling is solved based on optimization algorithms such as genetic algorithms and reinforcement learning to generate optimal scheduling strategies for power flow control, load allocation, and energy storage charging and discharging.

[0004] In the field of smart grid dispatching, various related contents based on multi-source heterogeneous data fusion have been proposed. For example, the smart grid dispatching decision system and method based on multi-source heterogeneous data fusion disclosed in Chinese invention patent application CN120638517B includes: Step 1: Collect multi-source heterogeneous data, construct a meteorological feature vector set, predict grid load changes, and output load forecast values; Step 2: Fuse load forecast values ​​and equipment health coefficients to construct a dual-risk collaborative analysis matrix, generate risk entropy values ​​and map them to standard risk models; Step 3: According to the standard risk models, perform generator output pre-adjustment, energy storage charging and discharging control and interruptible load management in stages, and coordinate microgrid resources through blockchain.

[0005] The above-mentioned technology has at least the following technical problems: Existing smart grid dispatch mainly involves collecting multi-source data, combining it with equipment health coefficients for risk analysis, and finally executing hierarchical dispatch based on risk patterns. The core focus is on dispatch execution under load forecasting and risk management. However, with the large-scale grid connection of high-proportion renewable energy, the peak and valley load fluctuations of the power grid are becoming increasingly severe, and existing technologies and methods are difficult to adapt to the needs of multi-source data collection and fusion under high-frequency disturbance scenarios.

[0006] When collecting and fusing multi-source heterogeneous power grid data under conditions where peak renewable energy generation coincides with off-peak loads, leading to power flow reversal, situations arise such as a sudden surge in photovoltaic output at midday while industrial load is significantly reduced due to holidays; distributed photovoltaic clusters concentrating their output at the end of the distribution network causing reverse power flow and heavy overload on feeders; and energy storage power stations and renewable energy plants experiencing alternating power oscillations due to simultaneous charging and discharging of electricity price signals. Under these conditions, the power grid exhibits characteristics such as rapid fluctuations in renewable energy output, delayed load response resulting in a sharp drop in net load, and a surge in voltage and reactive power regulation demands, entering a state of concurrent cluster regulation. Photovoltaic inverters and flexible... The frequent operation of power electronic equipment such as DC converter stations to support voltage leads to the generation of broadband oscillation signals in the power grid, including subsynchronous oscillation modes, supersynchronous oscillation components, and broadband oscillations caused by the interaction of multiple converter control. Existing technologies typically use power grid signal processing methods such as digital filtering and Fourier transform to filter out or simply average the broadband oscillation signals as noise. This may result in the acquired multi-source heterogeneous power grid data failing to accurately reflect the broadband dynamic operating status of the power grid under the concurrent regulation of the power electronic equipment cluster, thus causing distortion of the multi-source heterogeneous power grid data.

[0007] When fusing multi-source heterogeneous data based on distorted multi-source data of the power grid, existing technologies typically employ shallow machine learning fusion algorithms. This may result in the fused dataset failing to uncover the correlation between renewable energy output fluctuations, load mutations, and broadband dynamic disturbances, leading to insufficient accuracy in the generated fused dataset. In the case of a fused dataset with insufficient accuracy due to the superposition of distorted multi-source heterogeneous data of the power grid, when using prediction models such as Long Short-Term Memory Networks (LSTM) and Temporal Convolutional Networks (TCN) to predict load and renewable energy output, significant deviations occur in the prediction results. When multi-objective optimization scheduling is performed based on these deviation prediction results, the final result is that the power grid scheduling does not match the actual dynamic operation requirements of the power grid.

[0008] In summary, existing multi-source heterogeneous data acquisition and fusion technologies for the power grid cannot meet the complex demands of scenarios with high proportions of renewable energy grid connection and drastic peak-valley load fluctuations, such as a sudden surge in photovoltaic output at midday while industrial load is significantly reduced due to holidays, a surge in wind power generation at night combined with a sharp drop in residential heating load, and concentrated photovoltaic output during holidays leading to reverse heavy overload of the distribution network. Existing technologies lack synergistic efficiency in multi-source heterogeneous data acquisition and fusion, load and renewable energy output prediction, and multi-objective optimized scheduling. The adaptability of data authenticity, fusion accuracy, and prediction accuracy under high-frequency disturbance scenarios has not been specifically explored. With the widespread integration of high proportions of renewable energy and the increasingly severe peak-valley load fluctuations of the power grid, problems such as data distortion, insufficient fusion accuracy, and prediction bias are gradually accumulating and amplifying. The adaptability and reliability of power grid scheduling continue to decline, making it difficult to meet the precise requirements of intelligent power grid scheduling under scenarios with high proportions of renewable energy grid connection and drastic peak-valley load fluctuations. The support capacity for safe and stable operation of the power grid and efficient consumption of renewable energy is limited. Summary of the Invention

[0009] To address the technical problems of continuously declining adaptability and reliability of existing power grid dispatching technologies, and the mismatch between power grid dispatching and actual dynamic operation requirements of the power grid, this invention provides a power grid intelligent dispatching method based on multi-source data fusion. This method includes: performing broadband oscillation mode analysis of the power grid signal to capture broadband oscillations, identify broadband oscillation mode types, and quantify broadband oscillation coupling strength; determining, based on the analysis results, whether a broadband feature selective retention strategy and power grid broadband data processing adapted to the conventional steady-state dispatching requirements of the power grid are needed; after the broadband oscillation mode analysis of the power grid signal is completed, performing deep fusion of multi-source power grid data, and predicting the probability of power grid load and renewable energy output based on the fusion results; after the prediction is completed, assessing the reliability of the predicted power grid load and renewable energy output based on the prediction results; if the reliability assessment is satisfactory, performing intelligent power grid dispatching based on the prediction results; if the reliability is unsatisfactory, sending a warning of insufficient prediction reliability.

[0010] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention provides a power grid intelligent dispatching method based on multi-source data fusion. By performing broadband oscillation mode analysis of the power grid signal, and based on the analysis results determining whether a broadband feature selective retention strategy and broadband data processing of the power grid are necessary, it helps to capture the dynamic characteristics of broadband oscillations of the power grid, identify subsynchronous and supersynchronous oscillation modes and power frequency steady-state components. This effectively solves the problems in existing technologies where it is impossible to accurately distinguish between broadband oscillation signals and power frequency steady-state signals, and where it is difficult to adapt to broadband coupling resonance scenarios caused by the parallel operation of multiple power electronic devices. By selectively choosing retention strategies or data processing methods, it reduces the redundancy or loss of key features in broadband oscillation characteristics, providing high-quality basic data for subsequent deep fusion of multi-source power grid data. After the broadband oscillation mode analysis of the power grid signal is completed, deep fusion of multi-source power grid data is performed, and based on... The fusion results are used to predict the probability of grid load and renewable energy output. After the prediction is completed, a reliability assessment of the grid load and renewable energy output prediction is conducted based on the prediction results. This helps to solve the problem that traditional data fusion methods are unable to effectively explore the deep correlation between broadband dynamic disturbances and load and renewable energy output, realize the quantification of the uncertainty of load and renewable energy output, reduce the scheduling mismatch problem caused by unreliable prediction results, and improve the accuracy of prediction results. If the reliability assessment is qualified, intelligent grid scheduling is carried out based on the prediction results. If the reliability is unqualified, a prediction reliability insufficient warning is sent. This helps to realize the closed-loop management of grid broadband dynamic security prevention and control and scheduling decision-making, solves the problems of existing scheduling methods ignoring broadband oscillation risks and poor adaptability of scheduling strategies to actual grid operating conditions, and ensures the safe and stable operation of the grid.

[0011] 2. The advantage of this invention in acquiring the original voltage and current signal sequences lies in the fact that the voltage signal directly reflects the voltage amplitude and phase angle changes at each node of the power grid, serving as a core indicator of voltage stability and reactive power balance. The current signal directly reflects the power flow and load distribution in each branch of the power grid, acting as a key indicator of active power balance and power flow distribution. Physically, the two are strictly coupled through Ohm's law and Kirchhoff's laws. Existing technologies typically rely solely on voltage signals for broadband analysis or process voltage and current independently, neglecting the propagation direction of oscillation energy and the location information of the oscillation source implied by the phase difference and amplitude ratio between voltage and current. This invention, by acquiring the original voltage and current signal sequences in parallel and performing synchronous analysis, can simultaneously extract voltage oscillation modes and current... The flow oscillation mode provides a key basis for subsequent oscillation source localization and damping control. Based on the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters, the complementarity test of the power grid broadband oscillation frequency is performed to obtain qualified broadband oscillation mode pairs. Based on the qualified broadband oscillation mode pairs, the characteristics of the power grid broadband oscillation mode are fused and quantified to determine whether the power grid broadband oscillation characteristic parameters meet the broadband feature retention trigger condition. If so, a selective retention strategy for the power grid broadband features is adopted; otherwise, power grid broadband data processing is adopted. This helps to solve the problem that existing technologies confuse harmonic interference with broadband oscillation signals and cannot accurately identify true broadband oscillations. It realizes the quantitative characterization of broadband dynamic stability state and helps to solve the fundamental problem that existing technologies directly filter broadband oscillation signals as noise, resulting in distortion of multi-source heterogeneous data of the power grid.

[0012] 3. When a large number of distributed photovoltaic, energy storage converters, and other power electronic devices are simultaneously put into parallel operation in the distribution network, strong coupling resonance occurs between broadband oscillation modes. This makes it impossible for a single prediction confidence interval and the subsynchronous oscillation mode damping ratio to fully characterize the degree of oscillation coupling. This may lead to misjudgments in the assessment methods, failing to identify coupled broadband risks, resulting in inflated prediction reliability and mismatched dispatch strategies. Therefore, an alternative scheme for assessing the reliability of grid load and renewable energy output predictions needs to be implemented. Based on the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters of each qualified broadband oscillation mode pair, a broadband mode coupling resonance index of the grid can be obtained. This helps to quantify the broadband oscillation modes caused by the parallel operation of multiple power electronic devices with high accuracy. This method assesses the intensity of coupled resonance, addressing the shortcomings of existing evaluation methods in fully characterizing the degree of coupled broadband oscillations. It reduces evaluation misjudgments caused by coupled resonance, ensuring that evaluation results accurately reflect prediction reliability and the broadband operating status of the power grid. If the broadband mode coupled resonance index is less than a preset coupling threshold, intelligent grid dispatch is initiated; if it is not less than the preset coupling threshold, a prediction reliability deficiency warning is sent. This helps solve the problems of poor adaptability and susceptibility to misjudgment in coupled resonance scenarios of existing evaluation methods. Under the premise of ensuring broadband grid security, it achieves synergistic optimization of new energy consumption and economic grid operation, while providing clear feedback to the intelligent grid dispatch center, ensuring the long-term safe and stable operation of the power grid. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart of a power grid intelligent dispatching method based on multi-source data fusion is provided for an embodiment of the present invention; Figure 2 A flowchart outlining a general overview of a smart power grid dispatching method based on multi-source data fusion, provided in this embodiment of the invention. Figure 3 A power grid broadband oscillation mode feature fusion quantization logic diagram for a power grid intelligent dispatching method based on multi-source data fusion provided in an embodiment of the present invention; Figure 4 The present invention provides a logic diagram of a broadband feature selective retention strategy for a power grid intelligent dispatching method based on multi-source data fusion. Detailed Implementation

[0015] The technical solution provided by the present invention will now be described with reference to the accompanying drawings.

[0016] To facilitate understanding of the embodiments of the present invention, the following points will be explained first: First, in this invention, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural.

[0017] Secondly, in this invention, the use of prefixes such as "first" and "second" is merely for the purpose of distinguishing and describing different things belonging to the same name category, and does not constrain the order, size, or quantity of things. For example, "first message" and "second message" are simply different messages, and there is no temporal sequence, size, or priority relationship between them.

[0018] Third, in this invention, a database storing various types of preset data is established before the design of a power grid intelligent dispatching method based on multi-source data fusion. The database includes, but is not limited to, preset tolerance range, preset power frequency threshold, preset frequency complementarity deviation threshold, preset time interval, preset power grid damping rate threshold, etc.

[0019] This database is constructed based on historical operation archives from the power grid dispatch center, grid connection technical parameters of new energy power plants, phasor measurement units, data acquisition and monitoring control systems, and other multi-source sensor monitoring data, including raw heterogeneous datasets of the power grid, industry monitoring standard values, historical broadband oscillation event records, and field measurement calibration data. Data sources cover basic power grid topology information, power electronic equipment control parameters, monitoring equipment calibration indicators, industry standard thresholds, historical oscillation event characteristics, and dispatch operation results. The data structure adopts a hierarchical and relational architecture, integrating power grid topology geographic data, monitoring standard threshold data, broadband oscillation mode parameter benchmark data, and more. Spatiotemporal registration benchmark data, broadband dynamic characteristic index data, predictive reliability classification data, and scheduling strategy constraint data are classified and stored according to logical hierarchy, and field relationships are established. The storage method adopts a hybrid storage mode of structured and unstructured data. Standardized parameters and threshold data are stored through relational data tables, and high-frequency sampling data of phasor measurement units, broadband oscillation mode decomposition results, fused feature vector time series data, and scheduling strategy execution records are stored through time series databases and file indexes. This enables unified management, rapid retrieval, and dynamic updating of the set data, providing stable data support and benchmark basis for the entire process of intelligent scheduling method of the power grid.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] Example 1, Reference Figure 1 The present invention provides a flowchart of a power grid intelligent dispatching method based on multi-source data fusion, comprising: power grid broadband oscillation monitoring; during the acquisition and fusion of multi-source heterogeneous power grid data, performing power grid signal broadband oscillation mode analysis to capture power grid broadband oscillations, identify broadband oscillation mode types, and quantify broadband oscillation coupling strength; and determining, based on the analysis results, whether it is necessary to adopt a broadband feature selective retention strategy to retain broadband dynamic correlation characteristics and power frequency steady-state related information, construct a complete power grid multi-source heterogeneous dataset, and a power grid broadband data processing strategy to filter out broadband oscillation redundant signals, retain power frequency steady-state characteristics, and adapt to the power grid's conventional steady-state dispatching requirements.

[0022] The power grid multi-source data fusion and prediction assessment monitoring involves the following steps: After the broadband oscillation mode analysis of the power grid signal is completed, deep fusion of multi-source power grid data is performed to integrate broadband dynamic correlation characteristics and power frequency steady-state correlation information, weaken redundant interference, and strengthen the correlation characteristics of power grid nodes. Based on the fusion results, probability prediction of power grid load and renewable energy output is performed to quantify broadband oscillation disturbances, load fluctuations, and uncertainties in renewable energy output. After the prediction is completed, reliability assessment of power grid load and renewable energy output prediction is performed based on the prediction results to verify the authenticity and effectiveness of the prediction results and identify prediction deviations caused by coupled broadband oscillations.

[0023] If the reliability assessment is satisfactory, the system will implement intelligent grid dispatching based on the prediction results to achieve broadband dynamic security control, optimize grid power allocation, and reduce operating losses. If the reliability is unsatisfactory, the system will send a warning indicating insufficient predicted reliability.

[0024] As described above, monitoring of power grid broadband oscillations, monitoring of power grid multi-source data fusion and prediction, and monitoring of power grid dispatch status help achieve high-precision control and closed-loop prevention and control of the entire process of intelligent power grid dispatch. This solves the problems of lack of full-process monitoring in power grid dispatch and the disconnect between various links, such as missed detection of broadband risks, prediction distortion, and dispatch mismatch, which are caused by the lack of monitoring in the existing technology. Specifically, power grid broadband oscillation monitoring, power grid multi-source data fusion and prediction, and monitoring of power grid dispatch status are interconnected, progressive, and synergistic, forming a complete monitoring closed loop. Power grid broadband oscillation monitoring provides high-quality and reliable basic data for the subsequent power grid multi-source data fusion and prediction monitoring, while power grid multi-source data fusion and prediction monitoring provides accurate decision-making basis for the subsequent power grid dispatch status monitoring. At the same time, the abnormal feedback detected by it can guide the upstream power grid broadband oscillation monitoring and power grid multi-source data fusion and prediction monitoring to optimize parameters and adjust processes. The three support and verify each other to ensure the continuity, accuracy, and reliability of the entire intelligent power grid dispatch process.

[0025] refer to Figure 2This is a flowchart outlining a general overview of a power grid intelligent dispatching method based on multi-source data fusion, provided by an embodiment of the present invention. The method includes the following steps: performing broadband oscillation mode analysis of the power grid signal and performing a broadband oscillation frequency complementarity test to obtain the power grid synchronization frequency complementarity deviation value; determining whether the power grid synchronization frequency complementarity deviation value is less than a preset frequency complementarity deviation threshold; if not, removing the corresponding subsynchronous oscillation mode and supersynchronous oscillation component of the power grid; otherwise, performing broadband oscillation mode feature fusion quantization; after the broadband oscillation mode feature fusion quantization is completed, obtaining a multi-source heterogeneous dataset of the power grid; and performing deep fusion of multi-source data of the power grid based on the multi-source heterogeneous dataset to obtain a deeply fused dataset of the power grid; and performing probability prediction of power grid load and new energy output based on the deeply fused dataset of the power grid. After the prediction is completed, the reliability assessment of the power grid load and renewable energy output prediction is performed based on the obtained prediction confidence interval and real-time broadband dynamic characteristic parameters. It is determined whether the prediction confidence interval width is less than the preset confidence interval width threshold and whether the damping ratio of the subsynchronous oscillation mode in the broadband oscillation mode is greater than the preset minimum damping ratio threshold. If not, a prediction reliability insufficient prompt is sent, and the corresponding prediction confidence interval parameters and broadband oscillation mode damping ratio parameters are fed back to the power grid intelligent dispatch center. Otherwise, power grid intelligent dispatch is performed, and the power grid operation status quantitative parameters are obtained. It is determined whether the power grid operation status quantitative parameters meet the power grid operation status qualification conditions. If yes, power grid dispatch is performed; otherwise, a dispatch strategy security verification failure prompt is sent, and the corresponding power grid intelligent dispatch strategy is sent to the power grid intelligent dispatch center.

[0026] Preferably, the specific process of broadband oscillation mode analysis of power grid signals is as follows: First, collect multi-source heterogeneous data from the power grid, including voltage synchronous phasor data such as voltage amplitude, voltage phase angle, current amplitude, and current phase angle acquired by deployed phasor measurement units, voltage RMS and current RMS values ​​acquired by the data acquisition and monitoring control system, and operating status data such as active power and reactive power from new energy power plants and energy storage power stations; resample the voltage synchronous phasor data and current synchronous phasor data to obtain the original voltage... The signal sequence and the original current signal sequence, for example, the original voltage signal sequence can be represented as [U1, U2, U3, ..., Un], where U represents the original voltage signal sequence, U1 represents the instantaneous voltage value at the first sampling time, U2 represents the instantaneous voltage value at the second sampling time, ..., Un represents the instantaneous voltage value at the nth sampling time, and n is the total number of samples; resampling processing means unifying the voltage synchronization phasor data and current synchronization phasor data with different sampling frequencies to the same time base based on an interpolation algorithm; the second step is based on variational mode decomposition. The method performs adaptive frequency band decomposition on the original voltage signal sequence and the original current signal sequence respectively, obtaining multiple intrinsic mode function (IMF) components. Each IMF component corresponds to a specific center frequency. The IMF components include voltage IMF components and current IMF components, representing independent signal components separated from the original signal through variational mode decomposition. These components have a single center frequency, limited bandwidth, and no mode aliasing, and can accurately reflect the oscillation dynamic characteristics of the corresponding frequency band. For example, a certain IMF component can be expressed as A1cos(2... πf1t+φ1), where t represents time, A1 represents the oscillation amplitude of the intrinsic mode function component, f1 represents the center frequency of the intrinsic mode function component, and φ1 represents the initial phase of the component; wherein, intrinsic mode function components whose center frequency is within the subsynchronous frequency band are labeled as subsynchronous oscillation mode components, intrinsic mode function components whose center frequency is within the supersynchronous frequency band are labeled as supersynchronous oscillation components, and intrinsic mode function components whose center frequency is within the preset tolerance range of the power frequency are labeled as power frequency steady-state components; the subsynchronous frequency band indicates a frequency range of 0.The 1Hz to 50Hz frequency band is primarily oscillated by the control interactions of power electronic equipment such as new energy power plants and flexible DC converter stations, commonly seen in large-scale wind and solar power generation scenarios. This frequency range closely matches the actual oscillation characteristics of the power grid. The supersynchronous frequency band represents a range from 50Hz to several hundred Hz. Oscillations in this band are mostly caused by the high-frequency switching actions and control strategies of photovoltaic inverters and energy storage converters, and are one of the core frequency bands for wideband power grid oscillations. The preset tolerance range represents a preset tolerance range centered on the 50Hz power frequency, considering the actual operating frequency fluctuations of the power grid. This range covers normal fluctuations in the power frequency signal, can separate the steady-state component and the oscillating component of the power frequency, and avoids misinterpreting power frequency fluctuations as oscillation signals. The preset tolerance range includes both endpoints and is preset by designated personnel.

[0027] The third step involves identifying the mode parameters of each voltage intrinsic mode function (EMF) component and each current intrinsic mode function (EMF) component based on Prony analysis. The process is as follows: The subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters corresponding to each voltage EMF component are obtained using Prony analysis; the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters corresponding to the current EMF component are also obtained using Prony analysis. The subsynchronous oscillation mode parameters include the subsynchronous oscillation mode frequency, which characterizes the oscillation speed and spectral position of the subsynchronous oscillation mode, and the parameters used to quantify the subsynchronous oscillation... The subsynchronous oscillation mode amplitude, which represents the intensity and energy of the subsynchronous oscillation mode, the subsynchronous oscillation mode phase, which determines the timing and phase relationship of the subsynchronous oscillation mode, and the subsynchronous oscillation mode damping ratio, which evaluates the attenuation characteristics and stability risk of the subsynchronous oscillation mode. The supersynchronous oscillation component parameters include the supersynchronous oscillation component frequency, which characterizes the speed and spectral position of the supersynchronous oscillation component, the supersynchronous oscillation component amplitude, which quantifies the intensity and energy of the supersynchronous oscillation component, and the supersynchronous oscillation component phase, which determines the timing and phase relationship of the supersynchronous oscillation component.

[0028] The fourth step involves performing a broadband oscillation frequency complementarity test on the power grid based on the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters corresponding to each voltage intrinsic mode function component and current intrinsic mode function component. It should be noted that the following description uses the voltage intrinsic mode function component as an example; the calculation process, parameter extraction, and test logic for the current intrinsic mode function component are completely consistent with those for the voltage intrinsic mode function component. The specific process of the broadband oscillation frequency complementarity test is as follows: The subsynchronous oscillation mode frequency and the supersynchronous oscillation component frequency corresponding to each voltage intrinsic mode function component are summed. The result of this summation is then compared with a preset power frequency threshold. This difference is used as the power grid synchronization frequency complementarity deviation value to quantify the frequency coupling degree between the subsynchronous oscillation mode and the supersynchronous oscillation component of the power grid. The system is structured as follows: A preset power frequency threshold is set in advance by designated personnel, such as twice the power frequency (100Hz). If the complementary deviation of the power grid synchronization frequency is less than the preset complementary deviation threshold, the corresponding subsynchronous oscillation mode and supersynchronous oscillation component are identified as a qualified broadband oscillation mode pair. Based on the qualified broadband oscillation mode pair, the characteristics of the power grid broadband oscillation mode are fused and quantized. The preset complementary deviation threshold is represented by the average value of the complementary deviation of the power grid synchronization frequency over a historical period. If the complementary deviation of the power grid synchronization frequency is not less than the preset complementary deviation threshold, the corresponding subsynchronous oscillation mode and supersynchronous oscillation component are identified as harmonic interference or measurement noise and discarded. A qualified broadband oscillation mode pair represents a pairing combination of a qualified subsynchronous oscillation mode and a qualified supersynchronous oscillation component whose frequencies satisfy a complementary relationship.

[0029] As described above, wideband oscillation mode analysis of power grid signals and frequency complementarity verification of wideband oscillations help to accurately capture the dynamic characteristics of wideband oscillations in the power grid, identify effective wideband oscillation modes, quantify oscillation coupling strength, and eliminate invalid interference components. This enables precise characterization and reasonable screening of wideband oscillation characteristics, providing high-quality and reliable basic data for subsequent deep fusion of multi-source data from the power grid. It effectively solves the problems of incomplete capture of wideband oscillation characteristics, invalid mode interference in subsequent links, and inability to accurately distinguish between effective and invalid oscillation signals in existing technologies, thereby improving the overall level of intelligent dispatching of the power grid.

[0030] refer to Figure 3 This application provides a power grid broadband oscillation mode feature fusion quantization logic diagram for a power grid intelligent dispatching method based on multi-source data fusion, which is derived from... Figure 3 It can be seen that: the power grid broadband oscillation mode feature fusion quantization is performed, and the power grid broadband oscillation feature parameters are obtained. It is then determined whether the power grid broadband oscillation feature parameters meet the broadband feature retention trigger condition. If not, power grid broadband data processing is adopted; otherwise, a power grid broadband feature selective retention strategy is adopted.

[0031] Preferably, the specific process of fusion and quantification of broadband oscillation mode characteristics of the power grid is as follows: Obtain the broadband oscillation characteristic parameters of the power grid, including the rate of change of the power grid damping ratio and the coupling strength coefficient of the power grid synchronous mode; take the minimum value of the damping ratio of the subsynchronous oscillation mode among all qualified subsynchronous oscillation modes as the minimum damping ratio of the power grid, used to quantify the divergence risk of the current broadband oscillation of the power grid; perform differential operation on the minimum damping ratio sequence of the power grid in continuous time sections, and then use the result of the differential operation and the corresponding time interval as the rate of change of the power grid damping ratio, used to predict the development trend of the oscillation; the continuous time section represents two adjacent sampling times divided according to a preset time interval, wherein the preset time interval is set in advance by preset personnel; the minimum damping ratio sequence of the power grid represents a time sequence formed by arranging the minimum damping ratios of the power grid calculated from each continuous time section in chronological order; the differential operation represents performing a difference operation on the minimum damping ratio of the power grid in the current time section and the minimum damping ratio of the power grid in the previous time section in the minimum damping ratio sequence; and then perform a differential operation on the minimum damping ratio of the power grid in the current time section and the minimum damping ratio of the power grid in the previous time section; and then perform a differential operation on the minimum damping ratio sequence ... The ratio of the synchronous oscillation mode amplitude to the supersynchronous oscillation component amplitude is used as the grid synchronous mode coupling strength coefficient to assess the intensity of multi-converter control interaction. The system determines whether the grid broadband oscillation characteristic parameters meet the broadband characteristic retention trigger condition. If so, it indicates a current risk of broadband oscillation divergence in the grid, and a selective retention strategy for grid broadband characteristics is adopted. Conversely, if not, it indicates that the current broadband dynamic operating state of the grid is within a safe range, and grid broadband data processing is implemented. The broadband characteristic retention trigger condition indicates that the grid damping ratio change rate is less than a preset grid damping rate threshold, and the grid synchronous mode coupling strength coefficient is greater than a preset strong coupling threshold. The preset grid damping rate threshold is represented by the average value of the grid damping ratio change rate over a historical time period, and the preset strong coupling threshold is represented by the average value of the grid synchronous mode coupling strength coefficient over a historical time period. Grid broadband data processing involves filtering out broadband oscillation signals using a digital filtering algorithm, retaining only the power frequency steady-state characteristics, and marking them as a multi-source heterogeneous dataset of the grid.

[0032] As described above, the fusion and quantization of broadband oscillation mode characteristics of the power grid helps to integrate the subsynchronous and supersynchronous oscillation characteristics of qualified broadband oscillation mode pairs with high precision, realize the centralized and standardized characterization of broadband dynamic characteristics, reduce the feature distortion problem caused by the dispersion of broadband oscillation mode parameters, improve the synergistic adaptability between broadband oscillation characteristics and power frequency steady-state characteristics, and lay the characteristic foundation for the reliability assessment of power grid load and new energy output prediction and subsequent intelligent dispatch.

[0033] refer to Figure 4 This is a logic diagram of a broadband feature selective retention strategy for a power grid intelligent dispatching method based on multi-source data fusion, provided in an embodiment of this application. Figure 4It can be seen that by implementing a broadband feature selective retention strategy, broadband oscillation feature parameters and power frequency steady-state features of the power grid are obtained. The broadband oscillation feature parameters and power frequency steady-state features of the power grid are then structurally concatenated to obtain a multi-source heterogeneous dataset of the power grid. Based on the multi-source heterogeneous dataset of the power grid, deep fusion of multi-source data of the power grid is carried out.

[0034] Preferably, the specific process of the broadband feature selective retention strategy is as follows: The voltage intrinsic mode function components and current intrinsic mode function components, labeled as power frequency steady-state components, are extracted by amplitude analysis to obtain the effective voltage value, effective current value, active power, reactive power, and other features of each node, which are then used as power frequency steady-state features. The power grid broadband oscillation feature parameters and the power frequency steady-state features are structurally concatenated to form a two-layer data representation structure containing both steady-state and broadband dynamic information, and this structure is labeled as a multi-source heterogeneous dataset of the power grid. The structured concatenation means aligning the power frequency steady-state features and broadband dynamic feature parameters according to the power grid node numbers, forming a two-dimensional data matrix with nodes as indexes and power frequency steady-state features and broadband dynamic feature parameters as fields. The power grid node number represents a unique identifier assigned to each physical electrical node in the intelligent dispatching of the power grid, used to distinguish measurement points and control points at different locations in the power grid.

[0035] Deep fusion of multi-source data from the power grid is performed based on a multi-source heterogeneous dataset. This deep fusion involves inputting the multi-source heterogeneous dataset into a pre-defined deep fusion model and outputting a deeply fused dataset. The deeply fused dataset contains fused feature vectors for each node after reconstruction by the pre-defined deep fusion model. These fused feature vectors simultaneously encode power frequency steady-state characteristics, broadband dynamic characteristic parameters, and topological association information between nodes. The topological association information between nodes is derived from the multi-source heterogeneous dataset and encodes the connection relationships, power transmission paths, and mutual influence between power grid nodes. For example, if the power grid has nodes M, N, and P, with node M directly connected to node N, node N directly connected to node P, and node M indirectly connected to node P, then the topological association information can be encoded as node M-node N (direct connection, power transmission coefficient a), node N-node P (direct connection, power transmission coefficient b), and node M-node P (indirect connection, power transmission coefficient c), intuitively reflecting the topological relationships and power transmission characteristics between nodes.

[0036] It should be noted that the preset deep fusion model used in this embodiment is specifically a hybrid deep fusion model combining graph attention network and adaptive weighted wavelet fusion. In addition, Transformer fusion models, convolutional neural networks, long short-term memory networks, etc., can also be used, all of which can achieve cross-dimensional and cross-node deep fusion of qualified power grid multi-source heterogeneous data. The specific training process of the preset deep fusion model is as follows: Obtain power grid operating conditions under different conditions in historical time periods, such as the superposition of high renewable energy generation and low load with power flow reversal, conventional steady-state conditions, and wideband oscillation with slight disturbances, etc. The multi-source heterogeneous power grid dataset is used as the original training data and randomly divided into training and validation sets according to a preset partitioning ratio; initialize the preset deep fusion model parameters, setting the number of attention heads and hidden layer dimensions of the graph attention network, the number of decomposition layers and wavelet basis functions of the adaptive weighted wavelet fusion, and setting the optimizer of the model to an adaptive momentum estimation optimizer. The loss function adopts a composite loss function that combines the mean square error and the physical constraint penalty term. The model employs a physical constraint penalty term to ensure that the fusion result satisfies Kirchhoff's voltage and power balance laws, guaranteeing that the fusion result conforms to the actual operating rules of the power grid. The training set data is input into the initialized deep fusion model. The model first mines the correlation weights of multi-source data between different power grid nodes using a graph attention network, then performs multi-scale decomposition and fusion of power frequency steady-state data and broadband oscillation feature parameters through adaptive weighted wavelet fusion, dynamically adjusting the fusion weights of the two types of data and outputting a preliminary fusion feature vector. The optimizer updates the parameters of each layer of the model through backpropagation, iteratively adjusting the attention weights, wavelet decomposition parameters, and network hidden layer parameters to minimize the loss value. After model training, verification metrics such as fusion accuracy and feature recognition on the test set are obtained. If all test metrics meet the preset standards, such as fusion accuracy exceeding the preset accuracy or feature recognition exceeding the preset recognition, the model can be put into practical use. If the standards are not met, the model parameters and loss function weights are readjusted, and the above training process is repeated until the model performance meets the standards.

[0037] As described above, by employing a broadband feature selective retention strategy and deep fusion of multi-source power grid data, it is helpful to selectively retain the core dynamic features of the power grid's broadband frequency and key information on power frequency steady state, eliminate redundant interference signals, achieve orderly integration and efficient utilization of multi-source data, improve the integrity, consistency and accuracy of multi-source heterogeneous power grid data, help solve the problems of prediction distortion and scheduling mismatch caused by insufficient data fusion and loss of key features in existing technologies, strengthen the ability to identify broadband oscillation risks and prediction deviations, and improve the adaptability and reliability of intelligent power grid scheduling methods.

[0038] Preferably, the probability prediction of power grid load and renewable energy output is based on a deep fusion dataset of the power grid. The probability prediction of power grid load and renewable energy output means that the deep fusion dataset is input into a preset power grid probability prediction model, and the outputs load prediction results and renewable energy output prediction results with prediction confidence intervals. The confidence interval means that at each prediction time, the output is an uncertainty interval composed of the prediction mean and the prediction standard deviation. The prediction mean reflects the central estimate of the load or renewable energy output at that time. The prediction standard deviation is used to quantify the influence of broadband dynamic disturbances on the prediction results. The larger the standard deviation, the higher the contribution of the current broadband oscillation to the uncertainty of the prediction results. The load prediction result means the load prediction mean and load prediction standard deviation of each node at each prediction time within the future prediction period. The renewable energy output prediction result means the output prediction mean and output prediction standard deviation of each renewable energy station at each prediction time within the future prediction period.

[0039] It should be noted that the preset power grid probability prediction model used in this embodiment is a long short-term memory network model. In addition, gated recurrent unit models, convolutional neural networks combined with recurrent neural networks, etc., can also be used. The specific training process of the preset power grid probability prediction model is as follows: First, select historical multi-source heterogeneous power grid data, historical broadband oscillation characteristic data, historical load data, and historical renewable energy output data. Preprocess the selected data to remove abnormal and missing data to obtain training and test datasets. Then, initialize the relevant parameters of the preset power grid probability prediction model, determine the training hyperparameters of the model, and input the training dataset into the initialized model. With the goal of minimizing the error between the prediction result and the actual data, iteratively train the model. During the training process, monitor the training error and validation error of the model in real time. When the error reaches the preset error threshold, stop the model training. The preset error threshold is represented by the average error of the model training over a historical time period. Finally, use the test dataset to verify the performance of the trained model, and check the prediction accuracy and stability of the model. If the model performance meets the preset requirements, the model is determined as the final preset power grid probability prediction model. If it does not meet the requirements, adjust the model parameters and retrain and validate until the model performance meets the standards.

[0040] The reliability assessment of power grid load and renewable energy output predictions is based on the predicted confidence interval and real-time broadband dynamic characteristic parameters. The specific process is as follows: It is determined whether the predicted confidence interval width is less than a preset confidence interval width threshold, and whether the damping ratio of the subsynchronous oscillation mode in the broadband oscillation mode is greater than a preset minimum damping ratio threshold. If so, it indicates that the reliability of the current prediction result is high and the broadband dynamic operation of the power grid is within a stable range. Then, intelligent power grid dispatch is performed based on the load prediction result and the renewable energy output prediction result. The preset confidence interval width threshold is represented by the average value of the predicted confidence interval width over a historical time period, and the preset minimum damping ratio threshold is represented by the average value of the subsynchronous oscillation mode damping ratio over a historical time period. Conversely, if the predicted confidence interval width and the damping ratio are not greater than the preset minimum damping ratio threshold, it indicates that the reliability of the current prediction result is insufficient or that the power grid has a risk of broadband oscillation divergence. In this case, a prediction reliability insufficient warning is sent, and the corresponding predicted confidence interval parameters and broadband oscillation mode damping ratio parameters are fed back to the power grid intelligent dispatch center. The power grid intelligent dispatch center uses the feedback content to incrementally update the preset power grid probability prediction model. After data fusion and prediction model correction are completed and the reassessment is deemed satisfactory, intelligent power grid dispatch based on the updated prediction results is restored.

[0041] As described above, predicting the probability of grid load and renewable energy output helps to accurately predict grid load and renewable energy output, providing reliable support for grid power allocation and renewable energy consumption optimization. This promotes the transformation of grid intelligent dispatch from experience-driven to data-driven, improving the intelligence level of dispatch decisions. Compared with the single prediction method in the existing technology, this solution can better cope with the complex operating conditions of the grid, effectively reduce grid overload and renewable energy curtailment caused by prediction deviations, and improve the stability of grid operation.

[0042] Preferably, the specific process of intelligent power grid dispatch is as follows: Step 1, input the load forecasting results, the renewable energy output forecasting results, and the deep fusion dataset into a preset multi-objective optimization dispatching model, and output the optimal intelligent power grid dispatching strategy, including the output allocation of each renewable energy power station, the charging and discharging timing of energy storage power stations, and the load allocation of each node. For example, the set of renewable energy power stations is represented as {G1,G2,...,Gm}, where m represents the total number of renewable energy power stations, and Gk represents the output allocation value of the k-th renewable energy power station; the set of energy storage power stations is represented as {S1,S2,...,Sn}, where n represents the total number of energy storage power stations, and Sj represents the output allocation value of the k-th renewable energy power station. The charging and discharging power sequence of the j-th energy storage power station in each time period, with positive values ​​representing discharging and negative values ​​representing charging; the load allocation of each node is represented as {L1, L2, ..., Lp}, where p represents the total number of grid nodes, and Li represents the active and reactive load allocation values ​​of the i-th node; Step 2, substitute the generated optimal grid intelligent dispatch strategy into the grid power flow simulation model to simulate the grid operation state during the future dispatch period, and determine whether the grid operation state quantification parameters meet the grid operation state qualification conditions. If so, grid dispatch is carried out based on the optimal grid intelligent dispatch strategy; otherwise, a dispatch strategy security check failure is sent. The system will display a failure message and send the corresponding smart grid dispatch strategy to the smart grid dispatch center. The power flow simulation model represents a steady-state simulation calculation model that can fully deduce the voltage distribution, power flow direction, and power loss changes of the entire network under different dispatch strategies, based on the grid topology, node electrical constraints, and the operating boundaries of various power sources and loads. This model can reproduce the actual operating conditions of the grid during future dispatch periods and can use models such as the Newton-Raphson power flow model and the fast decoupled power flow model. The quantitative parameters of the grid operating status include voltage, current, renewable energy absorption rate, and power flow imbalance index. The qualified conditions for grid operating status are indicated by voltage and current. The system meets the requirements of a safe operating range, a renewable energy absorption rate greater than a preset absorption threshold, and a balanced power flow distribution. The preset absorption threshold is represented by the average renewable energy absorption rate over a historical period. The safe operating range means that the voltage is within a preset acceptable voltage range and the current is within a preset acceptable current range. Both the preset acceptable voltage and current ranges include both endpoints and are set in advance by the designated personnel. The renewable energy absorption rate is calculated by comparing the sum of the actual power generation of renewable energy plants in the optimal grid intelligent dispatch strategy with the preset renewable energy generation threshold. The preset renewable energy generation threshold is also set in advance by the designated personnel.

[0043] The process for verifying the balance of power flow distribution is as follows: The load rate of each transmission line is obtained, and the ratio of the number of lines with load rates exceeding a preset overload threshold to the total number of lines is used as the power flow imbalance index. If the power flow imbalance index is less than the preset balance threshold, the power flow distribution is determined to be balanced; if the power flow imbalance index is not less than the preset balance threshold, the power flow distribution is determined to be unbalanced. The preset overload threshold is represented by the average load rate over a historical period, and the preset balance threshold is represented by the average power flow imbalance index over a historical period. The specific process of power grid dispatching is as follows: The optimal intelligent power grid dispatching strategy is sent to the intelligent power grid dispatching center; the intelligent power grid dispatching center obtains the power grid... Dispatch control commands are issued to execution units such as new energy power plants, energy storage power plants, distribution network node control terminals, and power electronic equipment controllers. Each execution unit executes synchronously according to the grid dispatch control commands to ensure that the dispatch strategy is implemented effectively. The grid dispatch control commands represent operation commands with clear action parameters generated for different execution units. For example, for new energy power plants, the grid dispatch control command is represented as CG={Gk,Pset,k,tstart,tend}, where Gk represents the identifier of the k-th new energy power plant, Pset,k represents the active power output setpoint of the power plant, tstart represents the start time of the command's effective time, and tend represents the end time of the command's effective time.

[0044] It should be noted that the preset multi-objective optimization scheduling model used in this embodiment is specifically a non-dominated sorting genetic algorithm model. In addition, particle swarm optimization algorithm models, mixed integer linear programming models, etc. can also be used. The specific training process of the preset multi-objective optimization scheduling model is as follows: First, select historical grid scheduling data, historical load data, historical renewable energy output data, and broadband oscillation characteristic related data. Preprocess the selected data to remove outliers and fill missing values, and divide it into training dataset and validation dataset. Input the training dataset into the initialized model. Then, with the core optimization objectives of minimizing grid operation losses, maximizing renewable energy absorption rate, and minimizing broadband oscillation risk, iteratively optimize and train the model. During the training process, monitor the deviation between the model's output scheduling strategy and the actual scheduling data in real time. Finally, use the validation dataset to verify the performance of the trained model, and check the feasibility, stability, and optimization effect of the model's output scheduling strategy. If the model performance meets the preset scheduling accuracy and optimization objective requirements, such as the renewable energy absorption rate being greater than the preset absorption threshold, then the model is determined to be the final preset multi-objective optimization scheduling model. If it does not meet the requirements, adjust the model parameters and hyperparameters, and retrain and validate until the model performance meets the standards.

[0045] Example 2, based on Example 1, addresses the issue of strong coupling resonance between broadband oscillation modes when a large number of distributed photovoltaic and energy storage converters and other power electronic devices in the distribution network are simultaneously put into parallel operation. This can lead to misjudgments in the evaluation method due to the inability to fully characterize the degree of oscillation coupling by a single prediction confidence interval and the subsynchronous oscillation mode damping ratio, resulting in an inflated prediction reliability and a mismatch in dispatching strategies. Therefore, an alternative scheme for the reliability assessment of grid load and renewable energy output prediction needs to be implemented. The specific process is as follows: First, extract the subsynchronous oscillation mode ratio of each qualified broadband oscillation mode pair. The synchronous oscillation mode parameters and supersynchronous oscillation component parameters are used to construct a broadband mode coupling feature matrix for the power grid. The specific construction process is as follows: The number of qualified broadband oscillation mode pairs is used as the number of matrix rows, and the total number of extracted subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters is used as the number of matrix columns. The corresponding parameters of each broadband oscillation mode pair are sequentially filled into the corresponding positions in the matrix to form an initial coupling feature matrix. The initial coupling feature matrix is ​​then normalized using the min-max normalization algorithm, mapping all parameters to the [0,1] interval to eliminate dimensional differences. This matrix can completely characterize each broadband oscillation mode. The system first obtains a broadband modal coupling characteristic matrix of the power grid, which shows the parameter characteristics and interrelationships of the pairs of devices. Second, it obtains a broadband modal coupling resonance index for quantifying the broadband coupling strength caused by the parallel connection of multiple devices. The broadband modal coupling resonance index is obtained as follows: The subsynchronous oscillation amplitude and the supersynchronous oscillation amplitude of each row in the broadband modal coupling characteristic matrix are multiplied to obtain the single-mode pair coupling strength. The sum of all single-mode pair coupling strengths is then compared with the total number of qualified broadband oscillation mode pairs to obtain the average single-mode pair coupling strength. The average single-mode pair coupling strength is then... The broadband mode coupling resonance index of the power grid is obtained by weighted summation of the frequency coupling coefficients of each qualified broadband oscillation mode pair. The larger the value of the broadband mode coupling resonance index, the stronger the coupling resonance between broadband oscillation modes, and the greater the interference on the prediction results. The weighted summation means multiplying the average single-mode pair coupling strength and the frequency coupling coefficient of each qualified broadband oscillation mode pair by their respective weighting coefficients and then summing them up. The weighting coefficients include the single-mode pair coupling strength influence factor and the frequency coupling influence factor. The frequency coupling coefficient is represented by the ratio of the subsynchronous frequency to the supersynchronous frequency.

[0046] The process involves comparing the broadband mode coupling resonance index of the power grid with a preset coupling threshold. Specifically, if the broadband mode coupling resonance index is less than the preset coupling threshold, the prediction result is deemed reliable, and intelligent power grid dispatch is performed based on the load forecast and renewable energy output forecast results. The preset coupling threshold is represented by the average value of the broadband mode coupling resonance index over a historical time period. If the broadband mode coupling resonance index is not less than the preset coupling threshold, the prediction result is deemed unreliable, a prediction reliability insufficient warning is sent, and the corresponding broadband mode coupling resonance index is fed back to the intelligent power grid dispatch center. The intelligent power grid dispatch center uses the feedback to incrementally update the preset power grid probability prediction model. After the data fusion center and prediction model center complete the correction and the prediction reliability is reassessed, intelligent power grid dispatch based on the updated prediction results is restored.

[0047] It should be noted that in this embodiment, the calculation of the broadband mode coupling resonance index of the power grid relies on a pre-constructed set of coupling weight parameters. This set is set by professional technicians and stored in a database, providing weights for the single-mode influence factor on coupling strength and the frequency coupling influence factor, ensuring that the quantitative evaluation of coupling resonance strength in multi-mode coexistence scenarios has a clear reference standard.

[0048] Specifically, for the single-mode pair coupling strength influence factor and frequency coupling influence factor required for the broadband mode coupling resonance index of the power grid, the mapping set needs to be constructed based on historical broadband oscillation coupling experimental data. This includes the correspondence between the number of mode pairs, amplitude ratio, frequency matching degree and oscillation divergence probability under different coupling strengths. Combined with the historical scheduling strategy execution effect verification results, each set of characteristic parameter combinations is assigned a weight quantization value, and the effective range of the weight coefficient is recorded.

[0049] Specifically, during the data processing stage, the aforementioned mapping sets all need to undergo correlation analysis, such as Pearson correlation analysis and Spearman correlation coefficient analysis, to eliminate abnormal correlated data caused by measurement noise and modal parameter identification errors. The correspondence between statistically significant feature parameters and weight parameters is retained. The final integrated mapping set uses a 0-1 value range to represent the influence ratio of each weight parameter, achieving a one-to-one correspondence or many-to-one adaptation between feature parameters and influence weights. When the system calculates the broadband modal coupling resonance index of the power grid, it can quickly retrieve the corresponding influence parameters to ensure that the quantitative results of the coupling resonance intensity are objective and reliable, providing an accurate basis for predictive reliability assessment.

[0050] As described above, reliability assessment of grid load and renewable energy output prediction helps to overcome the shortcomings of traditional assessment methods that rely on a single indicator to fully characterize the degree of oscillation coupling under complex operating conditions caused by the parallel operation of a large number of power electronic devices and strong coupling resonance of broadband oscillation modes. This enables refined quantitative identification of coupled broadband risks, reduces the problems of judgment bias, risk omission, and overestimation of prediction reliability that are prone to occur in the original assessment methods, and forms a closed-loop management process of risk identification, anomaly alerts, and model updates, ensuring the safety and stability of the intelligent dispatch operation of the power grid under complex coupling conditions.

[0051] In summary, the present invention provides a power grid intelligent dispatching method based on multi-source data fusion. By performing broadband oscillation mode analysis of the power grid signal, and determining whether a broadband feature selective retention strategy and broadband data processing are needed based on the analysis results, this method helps to capture the dynamic characteristics of broadband oscillations in the power grid, identify subsynchronous and supersynchronous oscillation modes and power frequency steady-state components. It effectively solves the problems in existing technologies, such as the inability to accurately distinguish between broadband oscillation signals and power frequency steady-state signals, and the difficulty in adapting to broadband coupling resonance scenarios caused by the parallel operation of multiple power electronic devices. By selectively choosing retention strategies or data processing methods, it reduces the redundancy or loss of key features in broadband oscillation characteristics, providing high-quality foundational data for subsequent deep fusion of multi-source power grid data. After the broadband oscillation mode analysis of the power grid signal is completed, deep fusion of multi-source power grid data is performed, and... Based on the fusion results, the probability of power grid load and renewable energy output is predicted. After the prediction is completed, the reliability of the prediction is assessed. This helps to solve the problem that traditional data fusion methods are unable to effectively explore the deep correlation between broadband dynamic disturbances and load and renewable energy output, quantify the uncertainty of load and renewable energy output, reduce scheduling mismatch caused by unreliable prediction results, and improve the accuracy of prediction results. If the reliability assessment is qualified, intelligent power grid scheduling is carried out based on the prediction results. If the reliability is unqualified, a warning of insufficient prediction reliability is sent. This helps to realize closed-loop management of broadband dynamic security prevention and control and scheduling decision-making of the power grid, and solves the problems of existing scheduling methods ignoring broadband oscillation risks and poor adaptability of scheduling strategies to actual power grid operating conditions, thus ensuring the safe and stable operation of the power grid.

[0052] The embodiments described herein have been described in sufficient detail to enable those skilled in the art to practice the disclosed teachings. Other embodiments may be used and derived therefrom, such that structural and logical substitutions and changes may be made without departing from the scope of this disclosure. Therefore, the detailed description should not be construed as limiting, and the scope of the various embodiments is defined only by the appended claims and the full scope of their equivalents.

Claims

1. A smart power grid dispatching method based on multi-source data fusion, characterized in that, Includes the following steps: Broadband oscillation mode analysis of power grid signals is performed to capture broadband oscillations in the power grid, identify broadband oscillation mode types, and quantify broadband oscillation coupling strength. Based on the analysis results, it is determined whether a broadband feature selective retention strategy and power grid broadband data processing are needed to adapt to the regular steady-state dispatch requirements of the power grid. After the broadband oscillation mode analysis of the power grid signal is completed, deep fusion of multi-source data of the power grid is carried out, and the probability prediction of power grid load and new energy output is made based on the fusion results. After the prediction is completed, the reliability assessment of the prediction of power grid load and new energy output is carried out based on the prediction results. If the reliability assessment is satisfactory, intelligent power grid dispatch will be carried out based on the prediction results; if the reliability is unsatisfactory, a warning message indicating insufficient predicted reliability will be sent.

2. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 1, characterized in that, The specific process of broadband oscillation mode analysis of the power grid signal is as follows: Collect multi-source heterogeneous data from the power grid, including voltage synchronous phasor data and current synchronous phasor data acquired based on deployed phasor measurement units, and operating status data acquired based on the data acquisition and monitoring control system; The voltage synchronization phasor data and the current synchronization phasor data are resampled to obtain the original voltage signal sequence and the original current signal sequence. The resampling process refers to unifying voltage synchronization phasor data and current synchronization phasor data with different sampling frequencies to the same time base based on an interpolation algorithm; Based on the variational mode decomposition method, frequency band adaptive decomposition is performed on the original voltage signal sequence and the original current signal sequence to obtain multiple intrinsic mode function components. The intrinsic mode function components include voltage intrinsic mode function components and current intrinsic mode function components; Based on the Prony analysis method, the modal parameters of each voltage intrinsic mode function component and each current intrinsic mode function component are identified, as follows: Based on the Prony analysis method, the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters corresponding to the intrinsic mode function components of each voltage are obtained; Based on the Prony analysis method, the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters corresponding to the current eigenmode function components are obtained; The subsynchronous oscillation mode parameters include the subsynchronous oscillation mode frequency, the subsynchronous oscillation mode amplitude, the subsynchronous oscillation mode phase, and the subsynchronous oscillation mode damping ratio; The parameters of the supersynchronous oscillation component include the supersynchronous oscillation component frequency, the supersynchronous oscillation component amplitude, and the supersynchronous oscillation component phase. Based on the subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters corresponding to the voltage intrinsic mode function components and current intrinsic mode function components, a broadband oscillation frequency complementarity test of the power grid is performed.

3. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 2, characterized in that, The specific process for verifying the complementarity of the broadband oscillation frequencies of the power grid is as follows: The subsynchronous oscillation mode frequency and the supersynchronous oscillation component frequency corresponding to each voltage intrinsic mode function component are summed, and the result of the summation is calculated by subtracting from the preset power frequency threshold. This result is used as the complementary deviation value of the power grid synchronization frequency. If the complementary frequency deviation of the power grid synchronization frequency is less than the preset complementary frequency deviation threshold, the corresponding power grid subsynchronous oscillation mode and the power grid supersynchronous oscillation component are determined as a qualified broadband oscillation mode pair, and the power grid broadband oscillation mode feature fusion quantization is performed based on the qualified broadband oscillation mode pair. If the complementary frequency deviation of the power grid is not less than the preset complementary frequency deviation threshold, then the corresponding subsynchronous oscillation mode and supersynchronous oscillation component of the power grid will be eliminated.

4. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 3, characterized in that, The specific process of fusion quantization of the broadband oscillation mode characteristics of the power grid is as follows: Obtain characteristic parameters of broadband oscillation of the power grid, including the rate of change of the grid damping ratio and the grid synchronous mode coupling strength coefficient; The minimum damping ratio of the subsynchronous oscillation mode among all qualified subsynchronous oscillation modes is taken as the minimum damping ratio of the power grid. The power grid minimum damping ratio sequence of continuous time section is subjected to differential operation, and the result of differential operation and corresponding time interval ratio operation is used as the power grid damping ratio change rate. The result of the ratio calculation between the subsynchronous oscillation mode amplitude and the supersynchronous oscillation component amplitude is used as the synchronous mode coupling strength coefficient of the power grid. Determine whether the broadband oscillation characteristic parameters of the power grid meet the broadband characteristic retention triggering condition. If so, adopt a broadband characteristic selective retention strategy for the power grid. Conversely, broadband data processing of the power grid is adopted; The broadband feature retention trigger condition indicates that the rate of change of the grid damping ratio is less than the preset grid damping rate threshold, and the grid synchronization mode coupling strength coefficient is greater than the preset strong coupling threshold. The aforementioned power grid broadband data processing refers to filtering out broadband oscillation signals based on digital filtering algorithms, retaining only the power frequency steady-state characteristics, and marking them as a multi-source heterogeneous dataset of the power grid.

5. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 4, characterized in that, The specific process of the broadband feature selective preservation strategy is as follows: The voltage intrinsic mode function components and current intrinsic mode function components, which are labeled as power frequency steady-state components, are used to extract the amplitude of voltage RMS value, current RMS value, active power and reactive power of each node, which are used as power frequency steady-state characteristics. The power grid broadband oscillation characteristic parameters and the power frequency steady-state characteristics are structurally spliced ​​together to form a two-layer data representation structure containing steady-state information and broadband dynamic information, and it is marked as a power grid multi-source heterogeneous dataset. Deep fusion of multi-source data from the power grid based on multi-source heterogeneous datasets.

6. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 5, characterized in that, The deep fusion of multi-source data in the power grid means inputting the heterogeneous multi-source dataset of the power grid into a preset deep fusion model and outputting a deep fusion dataset of the power grid. The power grid deep fusion dataset contains the fused feature vectors of each node after being reconstructed by a preset deep fusion model. Probabilistic prediction of power grid load and renewable energy output based on deep fusion dataset of power grid.

7. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 6, characterized in that, The power grid load and renewable energy output probability prediction means that the deeply fused dataset is input into the preset power grid probability prediction model, and the output is the load prediction result and renewable energy output prediction result with prediction confidence interval. Based on the predicted confidence interval and real-time broadband dynamic characteristic parameters, the reliability assessment of power grid load and new energy output prediction is performed. The specific process is as follows: Determine whether the predicted confidence interval width is less than the preset confidence interval width threshold, and whether the damping ratio of the subsynchronous oscillation mode in the wide-frequency oscillation mode is greater than the preset minimum damping ratio threshold. If so, perform intelligent grid dispatch based on the load forecast results and the new energy output forecast results. Conversely, if the prediction reliability is insufficient, a warning will be sent, and the corresponding prediction confidence interval parameters and wideband oscillation mode damping ratio parameters will be fed back to the power grid intelligent dispatch center.

8. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 7, characterized in that, The specific process of the intelligent power grid dispatching is as follows: The load forecasting results, the new energy output forecasting results, and the deep fusion dataset are input into the preset multi-objective optimization scheduling model, and the optimal power grid intelligent scheduling strategy is output, including the output allocation of each new energy power station, the charging and discharging sequence of energy storage power stations, and the load allocation of each node. The generated optimal power grid intelligent dispatch strategy is substituted into the power grid power flow simulation model to simulate the power grid operation status during the future dispatch period. It is then determined whether the power grid operation status quantification parameters meet the qualified conditions for power grid operation status. If so, power grid dispatch is carried out based on the optimal power grid intelligent dispatch strategy. Otherwise, a dispatch strategy security verification failure prompt is sent, and the corresponding power grid intelligent dispatch strategy is sent to the power grid intelligent dispatch center. The quantitative parameters of the power grid operation status include voltage, current, renewable energy absorption rate, and power flow imbalance index. The specific process of power grid dispatching is as follows: Send the optimal smart grid dispatch strategy to the smart grid dispatch center; The smart grid dispatch center will send the acquired grid dispatch control commands to the execution unit; Each execution unit executes synchronously according to the power grid dispatch control instructions.

9. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 7, characterized in that, The reliability assessment of power grid load and renewable energy output forecasting also includes: The subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters of each qualified broadband oscillation mode pair are extracted to construct the broadband mode coupling characteristic matrix of the power grid. The specific construction process is as follows: The number of qualified broadband oscillation mode pairs is used as the number of rows in the matrix, and the total number of categories of extracted subsynchronous oscillation mode parameters and supersynchronous oscillation component parameters is used as the number of columns in the matrix. The corresponding parameters of each broadband oscillation mode pair are filled into the corresponding positions in the matrix in order to form the initial coupling feature matrix. The initial coupling feature matrix is ​​normalized. Obtain the broadband modal coupling resonance index of the power grid; The broadband modal coupling resonance index of the power grid is obtained through the following methods: The single-mode pair coupling strength is obtained by multiplying the subsynchronous oscillation amplitude and the supersynchronous oscillation amplitude in each row of the broadband mode coupling characteristic matrix of the power grid. The average single-mode pair coupling strength is obtained by summing the coupling strengths of all single-mode pairs and then comparing the result with the total number of qualified broadband oscillation mode pairs. The average single-mode pair coupling strength is weighted and summed with the frequency coupling coefficient of each qualified broadband oscillation mode pair to obtain the broadband mode coupling resonance index of the power grid. The broadband mode coupling resonance index of the power grid is compared with the preset coupling threshold.

10. The intelligent power grid dispatching method based on multi-source data fusion as described in claim 9, characterized in that, The specific process of comparing the broadband mode coupling resonance index of the power grid with the preset coupling threshold is as follows: If the broadband mode coupling resonance index of the power grid is less than the preset coupling threshold, the prediction result is determined to be reliable, and intelligent power grid dispatch is carried out based on the load prediction result and the new energy output prediction result. If the broadband mode coupling resonance index of the power grid is not less than the preset coupling threshold, the prediction result is determined to be unreliable, a prediction reliability insufficient prompt is sent, and the corresponding broadband mode coupling resonance index of the power grid is fed back to the power grid intelligent dispatch center.

Images