A power instability monitoring and early warning method based on deep learning

Abstract

The present invention discloses a method for monitoring and early warning of power instability based on deep learning, comprising: identifying power reverse flow events and potential reverse flow instability events based on power flow data of the power system; constructing a power instability pattern recognition model to identify the instability pattern and instability event risk level of potential reverse flow instability events; identifying key nodes of the power grid with excessive reverse flow impact load based on the reverse flow impact factor of the node; determining the amount of load that needs to be transferred based on the load deviation value of the key nodes of the power grid with excessive reverse flow impact load, and formulating a load control plan; predicting the time point when the instability precursor event turns into a reverse flow instability event based on the power instability pattern recognition model; evaluating the stability of the power grid system after implementing the load control plan, and optimizing the control parameters based on the control feedback data. The present invention significantly improves the power system's monitoring, early warning and dynamic control capabilities for reverse flow and instability events, effectively ensuring the safe and stable operation of the power grid.

Description

Technical Field

[0001] The present invention relates to the field of power monitoring, and in particular to a power instability monitoring and early warning method based on deep learning. Background Art

[0002] In modern power systems, the stability of power flow is crucial. Power flow refers to the flow of electricity from power generation sources to loads. The stability of current flow direction, frequency, and voltage are key indicators for the safe operation of power grids. However, in actual operation, power systems sometimes experience abnormal reverse flow, where the direction of current flow reverses, resulting in short-term, large-scale reverse flow and accompanied by fluctuations in frequency, phase, and voltage. Under certain conditions, these reverse flow events can further develop into reverse flow instability events. This means that if the reverse flow phenomenon is not alleviated, it will lead to persistent power system instability, with parameters such as frequency and phase deviating from normal levels, seriously affecting system operational safety. Reverse flow instability events pose a particular threat to the security of critical power grid nodes. If left unchecked, they can cause equipment damage, overload at critical nodes, and even widespread power outages. These events differ from common power instability events in that the former evolve from reverse flow, while the latter are often caused by sudden factors such as load surges or system failures. In complex power grid environments, reverse flow events and instability events can easily be confused, making power system monitoring challenging. Existing power system monitoring relies heavily on set thresholds and static analysis methods, using fixed thresholds such as current fluctuation amplitude and frequency deviation to determine system status. However, reverse flow events and instability events are influenced by multiple factors, making traditional threshold methods difficult to accurately distinguish between the two. Under complex grid conditions, fixed thresholds often lead to reverse flow events being mistaken for instability precursors, resulting in false alarms and missed alarms, which impact the grid's response speed and the effectiveness of emergency response. Furthermore, existing methods have limitations in identifying the risk of reverse flow events and load regulation. The lack of dynamic load regulation at key nodes makes it difficult to adjust loads in a timely manner during reverse flow events, making it prone to overload at key nodes and increasing the risk of system instability. Instability events typically progress through stages: precursors to instability, reverse flow instability, and complete instability. Traditional monitoring methods struggle to identify potential instability trends during the precursor stage, forcing the power system to respond passively only when events worsen, lacking early warning capabilities. Especially under extreme conditions, the reverse flow fluctuation frequency can resonate nonlinearly with the load variation frequency, causing a sharp increase in the load fluctuation amplitude at the node. Reverse flow can quickly evolve into an instability event, making it difficult for traditional monitoring methods to respond in a timely manner. Therefore, there is an urgent need for a power instability monitoring and early warning method based on intelligent technology to accurately identify power reverse flow events and instability events, provide effective early warnings, and thus improve the stability and emergency response capabilities of the power system. Summary of the Invention

[0003] To address the problems of the above-mentioned prior art, the present invention provides a method for monitoring and early warning of power instability based on deep learning, which mainly includes:

[0004] Based on the power flow data of the power system, the system identifies periods of time when abnormal fluctuation amplitude exceeds the limit, determines power reverse flow events, and identifies potential reverse flow instability events based on the frequency deviation amplitude, deviation recovery rate, and instantaneous change rate of power reverse flow events.

[0005] Based on the key parameters of historical power instability events, a power instability pattern recognition model is constructed to identify the instability patterns of potential reverse flow instability events and the risk level of reverse flow instability events where the instability pattern is power instability;

[0006] Based on the real-time load fluctuation data of key grid nodes in the event of reverse flow instability, combined with the reverse flow impact factor of the node, key grid nodes with excessive reverse flow impact load are identified;

[0007] Obtain real-time load data and historical load fluctuation patterns for key grid nodes where reverse flow impact load exceeds the limit, calculate the load deviation value for each node, determine the amount of load that needs to be transferred to each node, and formulate a load control plan;

[0008] Based on the historical key parameters of potential reverse current instability events where the instability mode is a precursor to instability, the key parameters of the instability precursor events within a preset time period in the future are predicted. In combination with the power instability pattern recognition model, the time point when the instability precursor event turns into a reverse current instability event is predicted, and a dynamic damping coefficient adjustment scheme and a resonance early warning scheme are implemented;

[0009] Obtain the load fluctuation response of key nodes in the power grid, evaluate the stability of the power grid system after implementing the load control scheme, damping coefficient dynamic adjustment scheme and resonance early warning scheme, and optimize the control parameters based on the control feedback data.

[0010] Furthermore, the method of identifying the period of time when the abnormal fluctuation amplitude exceeds the limit based on the power flow data of the power system, determining the power reverse flow event, and identifying the potential reverse flow instability event based on the frequency deviation amplitude, deviation recovery rate and instantaneous change rate of the power reverse flow event includes:

[0011] According to the power data interface of the power system, power flow data is obtained, including current direction, voltage fluctuation and frequency change; the data preprocessing algorithm is used to clean the power flow data, remove noise and abnormal points, and store the flow data in the data cache in timestamp order; the short-time Fourier transform algorithm is used to convert the current and voltage data into frequency domain and time domain signals, and the power characteristic parameters are extracted, including the change amplitude of the current direction, the change value of the voltage fluctuation amplitude and the frequency fluctuation rate; according to the power characteristic parameters, the isolation forest algorithm is used to identify the time period when the abnormal fluctuation amplitude exceeds the limit, and the power reverse flow event is determined; according to the power flow data of the power reverse flow event, the frequency deviation amplitude and deviation recovery rate of the power reverse flow event are statistically calculated, and the instantaneous change rate of the frequency deviation is calculated using the sliding average method; according to the frequency deviation amplitude, deviation recovery rate and instantaneous change rate of the power reverse flow event, the decision tree algorithm is used for model training to construct a potential reverse flow instability event identification model to identify potential reverse flow instability events.

[0012] Furthermore, the power instability pattern recognition model is constructed based on the key parameters of historical power instability events to identify the instability pattern of potential reverse flow instability events and the risk level of reverse flow instability events where the instability pattern is power instability, including:

[0013] Through the power instability feature database, the key parameters of historical power instability events are obtained, including the amplitude, duration and frequency characteristics of current fluctuations, and the power instability event stages and power instability event risk levels are marked. The random forest algorithm is used for model training to construct a power instability pattern recognition model. The power instability event stages include instability precursors and power instability. The power instability event risk levels include normal, warning, and dangerous. The instability modes include normal reverse flow, instability precursors, and power instability. According to the key parameters of potential reverse flow instability events, the power instability pattern recognition model is used to identify the instability mode of potential reverse flow instability events, and identify the risk level of reverse flow instability events whose instability mode is power instability. Potential reverse flow instability events whose instability mode is instability precursors are marked as instability precursor events, and potential reverse flow instability events whose instability mode is power instability are marked as reverse flow instability events. The reverse flow instability events with risk level labels are transmitted to the power instability event monitoring database, and a multi-level early warning classification mechanism is used to determine the response level of the reverse flow instability events.

[0014] Furthermore, the real-time load fluctuation data of the key nodes of the power grid caused by the reverse flow instability event, combined with the reverse flow impact factor of the node, identifies the key nodes of the power grid where the reverse flow impact load exceeds the limit, including:

[0015] Obtain real-time load fluctuation data of key nodes of the power grid during reverse flow instability events. Key nodes of the power grid include but are not limited to load concentration area connection points, trunk line nodes, power generation hub nodes, regional hub nodes and backup power supply nodes. Load fluctuation data include the load change rate, fluctuation amplitude, frequency and duration of the nodes. Based on the load fluctuation data of key nodes of the power grid, combined with the reverse flow influencing factors of the nodes, including the current reverse ratio and the phase angle of the load change, use the time-varying load reverse resonance factor evaluation formula The time-varying load reverse resonance factor Q of the key nodes of the power grid under the impact of reverse flow is calculated, where P1 is the benchmark load of the key nodes of the power grid, R is the current reverse ratio when the reverse flow occurs, indicating the intensity of the current reverse flow, θ is the phase angle of the load change, f is the reverse flow fluctuation frequency, g is the load change frequency, and ζ is the damping coefficient, indicating the attenuation ability of the power system to resonance, which is obtained by fitting historical data; based on the time-varying load reverse resonance factor and the benchmark load of the key nodes of the power grid, the load fluctuation amplitude of the key nodes of the power grid under the impact of reverse flow is determined; based on the preset node load fluctuation tolerance threshold and the calculated load fluctuation amplitude of the key nodes of the power grid under the impact of reverse flow, the key nodes of the power grid with excessive load due to reverse flow are identified; the reverse flow fluctuation frequency, load change frequency and the rising rate of the variable load reverse resonance factor are monitored in real time, and a dynamic adjustment plan for the damping coefficient and a resonance early warning plan are formulated.

[0016] Among them, real-time monitoring of the reverse flow fluctuation frequency, load change frequency and the rising rate of the variable load reverse resonance factor is carried out to formulate a dynamic adjustment plan for the damping coefficient and a resonance early warning plan, including:

[0017] Real-time monitoring of the reverse flow fluctuation frequency f and load change frequency g, if the frequency ratio is detected If the ratio is greater than the preset ratio threshold, a dynamic adjustment plan for the damping coefficient is formulated and implemented, and the damping coefficient ζ is dynamically adjusted to the preset critical damping coefficient ζ0; the rising rate of the time-varying load reverse resonance factor is calculated and recorded in real time. If the rising rate exceeds the preset rate threshold, a resonance early warning plan is formulated and implemented, including controlling the load flow by limiting the load at key nodes of the power grid or cutting off designated nodes, and transferring part of the load of the current node to adjacent nodes or backup power supply nodes whose load is lower than the preset load threshold.

[0018] Furthermore, the real-time load data and historical load fluctuation patterns of key nodes of the power grid where the reverse flow impact load exceeds the limit are obtained, the load deviation value of each node is calculated, the load amount that needs to be transferred or adjusted at each node is determined, and a load control plan is formulated, including:

[0019] Acquire real-time load data of key nodes of the power grid where the reverse impact load exceeds the limit, including load fluctuation amplitude, change rate and fluctuation frequency; extract historical load fluctuation patterns of key nodes of the power grid where the reverse impact load exceeds the limit through the power instability feature database, including load peak, average load level and fluctuation period; compare real-time load data with historical data, calculate the load deviation value of each node, and obtain the real-time-historical load deviation characteristics of key nodes of the power grid; utilize adaptive control algorithm to dynamically adjust control parameters according to the real-time-historical load deviation characteristics of key nodes of the power grid, determine the amount of load that needs to be transferred to each node, and obtain an adaptive control parameter set; formulate a load control plan based on the adaptive control parameter set to transfer the load of key nodes of the power grid where the reverse impact load exceeds the limit, and give priority to allocating the excess load to adjacent nodes or backup power supply nodes whose load is lower than the preset load threshold.

[0020] Furthermore, the historical key parameters of the potential reverse power instability event based on the instability mode as the instability precursor are predicted, the key parameters of the instability precursor event in the future preset time period are predicted, and the time point when the instability precursor event turns into the reverse power instability event is predicted in combination with the power instability pattern recognition model, and the damping coefficient dynamic adjustment scheme and resonance early warning scheme are implemented, including:

[0021] Through the power instability characteristic database, the historical key parameters of the instability precursor events are obtained and arranged in time series to construct the time series data of the historical key parameters of the instability precursor events; based on the time series data of the historical key parameters of the instability precursor events, the long short-term memory network is used to train the model to predict the key parameters of the instability precursor events in the future preset time period; based on the prediction results of the key parameters of the instability precursor events in the future preset time period, combined with the power instability pattern recognition model, the time point when the instability precursor event turns into the reverse current instability event and the risk level of the reverse current instability event are predicted; based on the predicted time point when the instability precursor event turns into the reverse current instability event, a reverse current instability event warning is sent, and a damping coefficient dynamic adjustment scheme and a resonance warning scheme are implemented.

[0022] Furthermore, the load fluctuation response of key nodes of the power grid is obtained, the stability of the power grid system after the load control scheme, the damping coefficient dynamic adjustment scheme and the resonance early warning scheme are evaluated, and the control parameters are optimized based on the control feedback data, including:

[0023] Continuously monitor the real-time load data of the power system after the implementation of the load control scheme, the damping coefficient dynamic adjustment scheme and the resonance warning scheme, obtain the load fluctuation response and dynamic change data of the damping coefficient of the key nodes of the power grid, and evaluate the stability of the power grid system after the implementation of each scheme; record the feedback data in the load control, damping coefficient adjustment and resonance warning response, including load changes, the ratio of reverse flow fluctuation frequency to load change frequency and the resonance warning effect, and judge the power instability mitigation effect of each scheme; based on the power instability mitigation effect, optimize the load control scheme, the damping coefficient dynamic adjustment scheme and the resonance warning scheme, adjust the load distribution amount, control direction and critical damping coefficient; re-verify the optimized load control scheme, damping coefficient dynamic adjustment scheme and resonance warning scheme, and record the load adjustment amount, critical damping coefficient, warning execution time and load distribution direction of each key node of the power grid. If the power instability mitigation effect of each scheme meets the expectations, the optimized load control scheme, damping coefficient dynamic adjustment scheme and resonance warning scheme will be stored in the power system control database.

[0024] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0025] The present invention provides a method for monitoring and early warning of power instability based on deep learning. By monitoring power flow data in real time and analyzing frequency deviation, recovery rate and instantaneous rate of change, the present invention can accurately identify abnormal reverse flow events and effectively distinguish between instability precursors and reverse flow instability, significantly reducing false alarms and missed reports. Based on historical data analysis, the present invention can identify different types of instability event patterns and evaluate their risk levels, helping the power grid to take more targeted early warning and response measures for different risk levels. By dynamically regulating the load of key nodes and quickly responding to the impact of reverse flow instability on nodes, the risk of overload at key nodes is avoided and the stability of the power grid is enhanced. The prediction and early warning functions of the present invention enable timely intervention in instability precursor events before they evolve into reverse flow instability, effectively preventing further deterioration of instability events. The present invention significantly improves the power system's monitoring, early warning and dynamic regulation capabilities for reverse flow and instability events, effectively ensuring the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a power instability monitoring and early warning method based on deep learning of the present invention;

[0027] Figure 2 Schematic diagram of a power instability monitoring and early warning method based on deep learning of the present invention;

[0028] Figure 3 This is another schematic diagram of a power instability monitoring and early warning method based on deep learning of the present invention. DETAILED DESCRIPTION

[0029] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] like Figure 1-3 In this embodiment, a power instability monitoring and early warning method based on deep learning may specifically include:

[0031] Step S101: Based on the power flow data of the power system, identify the period when the abnormal fluctuation amplitude exceeds the limit, determine the power reverse flow event, and identify potential reverse flow instability events based on the frequency deviation amplitude, deviation recovery rate and instantaneous change rate of the power reverse flow event.

[0032] Power flow data, including current direction, voltage fluctuations, and frequency changes, is acquired from the power system's power data interface. A data preprocessing algorithm is used to clean the power flow data, removing noise and outliers. The data is then stored in a data buffer in timestamp order. A short-time Fourier transform algorithm is used to convert current and voltage data into frequency and time domain signals, extracting power characteristic parameters, including the amplitude of current direction changes, the amplitude of voltage fluctuations, and the rate of frequency fluctuation. Based on these characteristic parameters, an isolation forest algorithm is used to identify periods of abnormal fluctuation amplitude exceeding the limit, thus identifying power reverse flow events. Based on the power flow data of power reverse flow events, the frequency deviation amplitude and deviation recovery rate are calculated, and the instantaneous rate of change of the frequency deviation is calculated using the sliding average method. A decision tree algorithm is used to train a model based on the frequency deviation amplitude, deviation recovery rate, and instantaneous rate of change of power reverse flow events to construct a potential reverse flow instability event identification model.

[0033] For example, in an electric power system, the flow data obtained in real time through the electric power data interface include: the current direction changes from 100A to 125A in a short period of time, resulting in a flow direction change amplitude of 25A; the voltage rises from 220V to 230V during this period, resulting in a voltage fluctuation of 10V; the frequency fluctuates slightly from 50Hz to 49.9Hz, resulting in a frequency deviation of 0.1Hz. To ensure the accuracy of the data, the data is first cleaned to remove noise and anomalies, and the cleaned data is stored in the data cache in timestamp order. The short-time Fourier transform is used to convert the current and voltage data from the time domain to the frequency domain, and the power characteristic parameters are extracted, including a current flow direction change amplitude of 25A, a voltage fluctuation amplitude of 10V, and a frequency fluctuation rate of 0.02Hz / s. The isolation forest algorithm was used to detect abnormal fluctuations. If the power system thresholds were set at current flow direction changes of no more than 20A, voltage fluctuations of no more than 8V, and frequency fluctuation rates of no more than 0.015Hz / s, the detection results during this time period showed that both the current flow direction change of 25A and the frequency fluctuation rate of 0.02Hz / s exceeded the thresholds, thus identifying this period as a power reverse flow event. After identifying the reverse flow event, further statistics showed that the frequency deviation amplitude of the event was 0.1Hz, and the deviation recovery rate was 0.05Hz per minute. Using the sliding average method with a 5s period, the instantaneous rate of change was calculated to be 0.02Hz / s. Using the frequency deviation amplitude, deviation recovery rate, and instantaneous rate of change of the power reverse flow event, a decision tree algorithm was used for training and model building to identify the power reverse flow event as a potential reverse flow instability event.

[0034] Step S102 : constructing a power instability pattern recognition model based on key parameters of historical power instability events to identify the instability pattern of potential reverse current instability events and the risk level of reverse current instability events where the instability pattern is power instability.

[0035] Using the power instability feature database, key parameters of historical power instability events, including the amplitude, duration, and frequency characteristics of current fluctuations, are obtained. The power instability event stages and risk levels are annotated. A random forest algorithm is used for model training to construct a power instability pattern recognition model. The power instability event stages include instability precursors and power instability, and the power instability event risk levels include normal, warning, and dangerous. Instability modes include normal reverse flow, instability precursors, and power instability. Based on the key parameters of potential reverse flow instability events, the power instability pattern recognition model is used to identify the instability mode of potential reverse flow instability events and the risk level of reverse flow instability events where the instability mode is power instability. Potential reverse flow instability events where the instability mode is a precursor to instability are labeled as precursor events, and potential reverse flow instability events where the instability mode is power instability are labeled as reverse flow instability events. Reverse flow instability events with risk level labels are transmitted to the power instability event monitoring database, and a multi-level early warning classification mechanism is used to determine the response level for reverse flow instability events.

[0036] For example, the key parameters of multiple historical power instability events were obtained from the power instability feature database. Among them, the current fluctuation amplitude was 35A, the duration was 20s, and the frequency characteristics showed that the frequency briefly dropped from 50Hz to 49.3Hz during the event. This event was labeled as the power instability stage and the risk level was set to warning. Event 2, the current fluctuation amplitude was 42A, the duration was 25s, and the frequency dropped from 50Hz to 49.0Hz. This event was identified as the power instability stage and was rated as dangerous due to its large frequency fluctuation and longer duration. Event 3, the current fluctuation amplitude was 30A, the duration was 18s, and the frequency fluctuated between 50Hz and 49.5Hz. Although the frequency fluctuation was controlled, the fluctuation amplitude and duration met the criteria for the instability stage, so it was labeled as the power instability stage and the risk level was set to normal. Based on a large number of similar events, a power instability pattern recognition model was constructed through training using a random forest algorithm to identify the instability patterns of potential reverse current instability events. If the key parameters of a potential reverse instability event are a current fluctuation of 30A, a duration of 15s, and a frequency drop to 49.4Hz, the power instability pattern recognition model identifies it as a precursory mode. Therefore, the potential reverse instability event is not assigned a risk level. If the key parameters of a reverse instability event are a current fluctuation of 40A, a duration of 25s, and a frequency drop to 49.2Hz, the power instability pattern recognition model determines that it has entered the power instability phase and thus marks the event as a dangerous risk level. Potential reverse instability events with an instability pattern of precursory mode are marked as instability precursor events, and potential reverse instability events with an instability pattern of power instability are marked as reverse instability events. For events in the power instability phase and assigned a risk level, their classification labels with risk levels are stored in the power instability event monitoring database. Based on the dangerous risk level of the event, a three-level warning mechanism is triggered, alerting power management personnel to take immediate measures to address the event, ensuring power system stability and preventing further deterioration.

[0037] Step S103 , identifying key grid nodes with excessive reverse flow impact load based on real-time load fluctuation data of key grid nodes in the reverse flow instability event and the reverse flow impact factor of the node.

[0038] Obtain real-time load fluctuation data of key grid nodes that are affected by reverse flow instability events. Key grid nodes include but are not limited to load concentration area connection points, trunk line nodes, power generation hub nodes, regional hub nodes, and backup power supply nodes. Load fluctuation data includes the load change rate, fluctuation amplitude, frequency, and duration of the node. Based on the load fluctuation data of key grid nodes and the reverse flow influencing factors of the nodes, including the current reverse ratio and the phase angle of the load change, the time-varying load reverse resonance factor evaluation formula is used. The time-varying load reverse resonance factor Q of key grid nodes subjected to reverse flow is calculated, where P1 is the baseline load of the key grid node, R is the current reverse ratio when reverse flow occurs, indicating the intensity of the current reverse flow, θ is the phase angle of the load change, f is the reverse flow fluctuation frequency, g is the load change frequency, and ζ is the damping coefficient, representing the power system's ability to attenuate resonance. This factor is obtained by fitting historical data. Based on the time-varying load reverse resonance factor and the baseline load of the key grid node, the load fluctuation amplitude of the key grid node subjected to reverse flow is determined. Based on the preset node load fluctuation tolerance threshold and the calculated load fluctuation amplitude of the key grid node subjected to reverse flow, key grid nodes with excessive load due to reverse flow are identified. The reverse flow fluctuation frequency, load change frequency, and the rate of increase of the variable load reverse resonance factor are monitored in real time to develop a dynamic damping coefficient adjustment plan and a resonance early warning plan.

[0039] For example, in an electric power system, the real-time load fluctuation data of a key node of the power grid in a reverse flow instability event is obtained. The key node is located in a load concentration area. The recorded load fluctuation data includes a load change rate of 20kW / s, a fluctuation amplitude of 80kW, a frequency of 49Hz, and a duration of 40s. Among them, the key nodes of the power grid include load concentration area connection points, trunk line nodes, power generation hub nodes, regional hub nodes and backup power supply nodes, etc., and the benchmark load P1 of the key node is obtained to be 300kW, and the phase angle θ of the load change is 45°. In some reverse flow instability events, the current reverse ratio R and the reverse flow fluctuation frequency f will produce a nonlinear resonance effect with the load change frequency g, causing the load fluctuation amplitude of the node to increase sharply, further increasing the risk of the key node. According to the reverse flow impact factor of the key node of the power grid, the reverse flow impact load fluctuation amplitude formula is used. The time-varying load reverse resonance factor Q at a key grid node subjected to a reverse current shock was calculated. The current reverse ratio R was 1.1, the fluctuation frequency f was 49 Hz, the load variation phase angle θ was 45° (sin(θ) = 0.707), and the load variation frequency g was 50 Hz. The damping coefficient ζ, obtained by fitting historical data, was 0.05, resulting in a calculated Q of approximately 1.95. Based on the time-varying load reverse resonance factor Q and the node's baseline load P1, the amplified load fluctuation amplitude at this node under the reverse current shock was determined to be 585 kW. Given the grid's preset load fluctuation tolerance threshold of 500 kW, a fluctuation amplitude of 585 kW exceeds the tolerance range, identifying the risk of excessive load at this key grid node due to a reverse current shock. This result triggers the power system's early warning mechanism, prompting relevant departments to take timely regulatory measures to reduce load pressure and ensure power system stability. To further control risks, the system developed a dynamic adjustment plan for the damping coefficient and a resonance early warning plan, and began real-time monitoring of the rising rate of the reverse flow fluctuation frequency f, the load change frequency g, and the resonance factor Q to alleviate load fluctuations and prevent the resonance effect from spreading to other nodes in the power grid.

[0040] Among them, the countercurrent fluctuation frequency, load change frequency and the rising rate of the variable load reverse resonance factor are monitored in real time, and a dynamic adjustment plan for the damping coefficient and a resonance early warning plan are formulated.

[0041] Real-time monitoring of the reverse flow fluctuation frequency f and load change frequency g, if the frequency ratio is detected If the ratio is greater than the preset ratio threshold, a dynamic damping coefficient adjustment plan is formulated and implemented to dynamically adjust the damping coefficient ζ to the preset critical damping coefficient ζ0. The rising rate of the time-varying load reverse resonance factor is calculated and recorded in real time. If the rising rate exceeds the preset rate threshold, a resonance early warning plan is formulated and implemented, including limiting the load at key nodes of the power grid or cutting off designated nodes to control load flow and transfer part of the load of the current node to adjacent nodes or backup power supply nodes with loads below the preset load threshold.

[0042] For example, in the power system, a reverse flow instability event is detected in real time at a key node of the power grid, and the reverse flow fluctuation frequency and load change frequency of the node are continuously recorded during the process. If the reverse flow fluctuation frequency f is 49.2Hz and the load change frequency g is 50Hz. The preset frequency ratio threshold is 0.98, and the frequency ratio at this time is The ratio is 0.984, exceeding the preset ratio threshold. Upon detecting that this ratio exceeds the threshold, a dynamic damping coefficient adjustment scheme is immediately initiated, gradually increasing the damping coefficient from its original value to the preset critical damping coefficient of 0.15 to suppress the system's resonance effect. This adjustment effectively mitigates the amplification of load fluctuations caused by reverse flow instability, helping to keep the power system within a controllable range. Simultaneously, the rising rate of the time-varying load reverse resonance factor is calculated and recorded in real time. If the current rising rate is 0.05 / s, while the system's preset rate threshold is 0.04 / s, this indicates that the time-varying load reverse resonance factor is increasing too rapidly, potentially leading to a risk of overload. Therefore, a resonance warning scheme is immediately developed and implemented. This scheme includes intervention measures such as load limiting and load shifting. First, the load at this critical node is limited to below 400kW to prevent further fluctuation amplification. Then, the system shifts part of the node's load, such as 50kW, to a nearby backup power supply node or another node with a load below the preset threshold, ensuring that other nodes in the grid are not affected by the transmission of the resonance effect.

[0043] Step S104, obtaining real-time load data and historical load fluctuation patterns of key nodes of the power grid where the reverse flow impact load exceeds the limit, calculating the load deviation value of each node, determining the load amount that needs to be transferred to each node, and formulating a load control plan.

[0044] Real-time load data is obtained for key grid nodes where reverse flow impact load exceeds the limit, including load fluctuation amplitude, change rate, and fluctuation frequency. The historical load fluctuation patterns of key grid nodes where reverse flow impact load exceeds the limit are extracted using the power instability feature database, including load peak, average load level, and fluctuation period. Real-time load data is compared with historical data, and the load deviation value of each node is calculated to obtain the real-time-historical load deviation characteristics of key grid nodes. An adaptive control algorithm is used to dynamically adjust control parameters based on the real-time-historical load deviation characteristics of key grid nodes, determine the amount of load that needs to be transferred to each node, and obtain an adaptive control parameter set. Based on the adaptive control parameter set, a load control plan is formulated to transfer the load of key grid nodes where reverse flow impact load exceeds the limit, and preferentially distribute the excess load to adjacent nodes or backup power supply nodes whose load is below the preset load threshold.

[0045] For example, in the power grid system, a key node of the power grid exceeds the load limit due to a reverse flow shock. Real-time data shows that the load fluctuation amplitude of the node is 380kW, the change rate is 12kW / s, and the fluctuation frequency is 48Hz. Through the power instability feature database, the historical load fluctuation pattern of the node is extracted, which records a load peak of 350kW, an average load level of 320kW, and a fluctuation period of 60s. The real-time data is compared with the historical data, and the load deviation value of the node is calculated to be 30kW by subtracting the historical load peak of 350kW from the real-time load fluctuation amplitude of 380kW, and the real-time-historical load deviation feature of the node is obtained. Due to the large load deviation, the adaptive control algorithm is started to dynamically adjust the control parameters according to the degree of load deviation and the change rate, and it is calculated that the load that needs to be transferred to the node is 30kW. Based on the adaptive control parameter set, a load control plan is formulated to prioritize the 30 kW excess load to neighboring nodes or backup power supply nodes with loads below a preset threshold. For example, 15 kW of the load is transferred to neighboring node M, where the load is lower, and the remaining 15 kW is transferred to backup power supply node N. This control plan effectively reduces the load pressure on this critical node by rationally distributing the excess load, thereby enhancing the stability and security of the power grid.

[0046] Step S105, based on the historical key parameters of the potential reverse current instability event whose instability mode is a precursor to instability, predict the key parameters of the instability precursor event in the future preset time period, and combine with the power instability pattern recognition model to predict the time point when the instability precursor event turns into a reverse current instability event, and implement the damping coefficient dynamic adjustment plan and the resonance early warning plan.

[0047] The historical key parameters of instability precursor events are obtained from the power instability characteristic database and arranged in time series to construct time series data of the historical key parameters of instability precursor events. Based on the time series data of the historical key parameters of instability precursor events, a long short-term memory network is used for model training to predict the key parameters of instability precursor events within a preset future time period. Based on the predicted key parameters of instability precursor events within the preset future time period, combined with the power instability pattern recognition model, the time point when the instability precursor event will transform into a reverse current instability event and the risk level of the reverse current instability event are predicted. Based on the predicted time point when the instability precursor event transforms into a reverse current instability event, a reverse current instability event warning is issued, and a dynamic damping coefficient adjustment scheme and a resonance warning scheme are implemented.

[0048] For example, historical data for a precursory event for instability is obtained from the power instability feature database. The parameters for the past three precursory events are current fluctuations of 15A, 20A, and 25A, and event durations of 10s, 15s, and 20s, respectively. Frequency characteristics show a decrease from 50Hz to 49.7Hz, 49.5Hz, and 49.3Hz, respectively. This data is arranged chronologically to construct a time series of precursory events for this node. A long-short-term memory (LSTM) network model is trained on this time series data to predict key parameters of the precursory event within the next 10 minutes. The predicted current fluctuations are expected to reach 30A, the duration is expected to extend to 25s, and the frequency may drop to 49.2Hz. Combined with analysis from the power instability pattern recognition model, the LSTM model predicts that the precursory event for this node will transition to a reverse current instability event in 8 minutes, with a risk level of dangerous. Based on this prediction, an immediate reverse current instability event alert is issued, indicating that a reverse current instability event is likely to occur at this node within the next 8 minutes. In order to reduce the load and prevent the instability from worsening, the dynamic adjustment scheme of the damping coefficient and the resonance early warning scheme were started in advance, and real-time monitoring of the reverse flow fluctuation frequency, load change frequency and the rising rate of the resonance factor was started to alleviate the load fluctuation and prevent the resonance effect from spreading to other nodes in the power grid.

[0049] Step S106, obtaining the load fluctuation response of key nodes of the power grid, evaluating the stability of the power grid system after implementing the load control scheme, the damping coefficient dynamic adjustment scheme and the resonance early warning scheme, and optimizing the control parameters based on the control feedback data.

[0050] Continuously monitor real-time power system load data after implementing load control schemes, dynamic damping coefficient adjustment schemes, and resonance warning schemes. Obtain data on load fluctuation responses and dynamic changes in damping coefficients at key grid nodes to evaluate grid system stability after implementation of each scheme. Record feedback data from load control, damping coefficient adjustment, and resonance warning responses, including load changes, the ratio of reverse flow fluctuation frequency to load change frequency, and resonance warning effectiveness, to determine the effectiveness of each scheme in mitigating power instability. Based on the effectiveness of power instability mitigation, optimize the load control scheme, dynamic damping coefficient adjustment scheme, and resonance warning scheme, adjusting the load distribution, control direction, and critical damping coefficient. Revalidate the optimized load control scheme, dynamic damping coefficient adjustment scheme, and resonance warning scheme, and record the load adjustment, critical damping coefficient, warning execution time, and load distribution direction at each key grid node. If the power instability mitigation effectiveness of each scheme meets expectations, store the optimized load control scheme, dynamic damping coefficient adjustment scheme, and resonance warning scheme in the power system control database.

[0051] For example, during power system operation, a load control scheme, a dynamic damping coefficient adjustment scheme, and a resonance warning scheme were implemented at a key grid node in a load-concentrated area to mitigate the risk of load fluctuations at this node due to reverse flow instability. Real-time monitoring data showed that the initial load fluctuation amplitude at this node was 75 kW, with a load change rate of 20 kW / s, a reverse flow fluctuation frequency of 49 Hz, and a load change frequency of 50 Hz. The frequency ratio was 0.98, reaching the preset ratio threshold. Therefore, the dynamic damping coefficient adjustment scheme was immediately implemented, directly setting the damping coefficient of this node to the preset critical damping coefficient of 0.1 to suppress the resonance effect. After implementation of the scheme, load fluctuation response data was continuously monitored and recorded. The load fluctuation amplitude was observed to have dropped to 55 kW, indicating that the load control scheme achieved some mitigation effect. Furthermore, the resonance warning scheme effectively controlled the further spread of load fluctuations, ensuring system stability. The control feedback data recorded a load adjustment of 25 kW at this node, with some load shifted to a lower-load backup power supply node. The damping coefficient adjustment feedback indicated that the critical damping coefficient had been successfully set to 0.1, ensuring the system's robust attenuation capability. The resonance warning scheme feedback recorded an alert execution time of 12 minutes. By limiting load flow at key nodes and distributing the load, the power system effectively mitigated fluctuations. Based on this feedback data, the load control scheme, damping coefficient dynamic adjustment scheme, and resonance warning scheme were optimized. The optimized scheme adjusted the load distribution to 20 kW and replanned the control direction to balance the load distribution. The critical damping coefficient was verified to be 0.09 to enhance system stability in the event of a similar incident. The optimized schemes were re-verified. If all achieved the expected power instability mitigation effect, the optimized load control scheme, damping coefficient dynamic adjustment scheme, and resonance warning scheme, along with related data, were stored in the power system control database for reference in subsequent incident responses. The relevant data included the load adjustment at key nodes, the set critical damping coefficient, the warning execution time, and the load distribution direction.

[0052] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the concept of this application. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A power instability monitoring and early warning method based on deep learning, characterized in that: The method comprises: Based on the power flow data of the power system, the system identifies periods of time when abnormal fluctuation amplitude exceeds the limit, determines power reverse flow events, and identifies potential reverse flow instability events based on the frequency deviation amplitude, deviation recovery rate, and instantaneous change rate of power reverse flow events. Based on the key parameters of historical power instability events, a power instability pattern recognition model is constructed to identify the instability patterns of potential reverse flow instability events and the risk level of reverse flow instability events where the instability pattern is power instability; Based on the real-time load fluctuation data of key grid nodes in the event of reverse flow instability, combined with the reverse flow impact factor of the node, key grid nodes with excessive reverse flow impact load are identified; Obtain real-time load data and historical load fluctuation patterns for key grid nodes where reverse flow impact load exceeds the limit, calculate the load deviation value for each node, determine the amount of load that needs to be transferred to each node, and formulate a load control plan; Based on the historical key parameters of potential reverse current instability events where the instability mode is a precursor to instability, the key parameters of the instability precursor events within a preset time period in the future are predicted. In combination with the power instability pattern recognition model, the time point when the instability precursor event turns into a reverse current instability event is predicted, and a dynamic damping coefficient adjustment scheme and a resonance early warning scheme are implemented; Obtain the load fluctuation response of key nodes in the power grid, evaluate the stability of the power grid system after implementing load control schemes, damping coefficient dynamic adjustment schemes, and resonance early warning schemes, and optimize control parameters based on control feedback data; The method of identifying key grid nodes with excessive reverse flow impact loads based on real-time load fluctuation data of key grid nodes caused by reverse flow instability events, combined with the reverse flow impact factors of the nodes, includes: Obtain real-time load fluctuation data of key nodes of the power grid during reverse flow instability events. Key nodes of the power grid include but are not limited to load concentration area connection points, trunk line nodes, power generation hub nodes, regional hub nodes and backup power supply nodes. Load fluctuation data include the load change rate, fluctuation amplitude, frequency and duration of the nodes. Based on the load fluctuation data of key nodes of the power grid, combined with the reverse flow influencing factors of the nodes, including the current reverse ratio and the phase angle of the load change, use the time-varying load reverse resonance factor evaluation formula , calculate the time-varying load reverse resonance factor of key nodes of the power grid under the impact of reverse flow ,in is the benchmark load of the key nodes of the power grid, R is the reverse ratio of the current when the reverse flow occurs, indicating the reverse flow intensity of the current. is the phase angle of load change, is the countercurrent fluctuation frequency, is the load change frequency, It is the damping coefficient, which represents the attenuation ability of the power system against resonance; according to the time-varying load reverse resonance factor and the benchmark load of the key nodes of the power grid, the load fluctuation amplitude of the key nodes of the power grid under the impact of reverse flow is determined; according to the preset node load fluctuation tolerance threshold and the calculated load fluctuation amplitude of the key nodes of the power grid under the impact of reverse flow, the key nodes of the power grid with excessive reverse flow impact load are identified; the reverse flow fluctuation frequency, load change frequency and the rising rate of the variable load reverse resonance factor are monitored in real time, and a dynamic adjustment plan for the damping coefficient and a resonance early warning plan are formulated.

2. The method according to claim 1, wherein The method of identifying, based on the power flow data of the power system, a period of time when the abnormal fluctuation amplitude exceeds the limit, determining a power reverse flow event, and identifying a potential reverse flow instability event based on the frequency deviation amplitude, deviation recovery rate, and instantaneous change rate of the power reverse flow event includes: According to the power data interface of the power system, power flow data is obtained, including current direction, voltage fluctuation and frequency change; the data preprocessing algorithm is used to clean the power flow data, remove noise and abnormal points, and store the flow data in the data cache in timestamp order; the short-time Fourier transform algorithm is used to convert the current and voltage data into frequency domain and time domain signals, and the power characteristic parameters are extracted, including the change amplitude of the current direction, the change value of the voltage fluctuation amplitude and the frequency fluctuation rate; according to the power characteristic parameters, the isolation forest algorithm is used to identify the time period when the abnormal fluctuation amplitude exceeds the limit, and the power reverse flow event is determined; according to the power flow data of the power reverse flow event, the frequency deviation amplitude and deviation recovery rate of the power reverse flow event are statistically calculated, and the instantaneous change rate of the frequency deviation is calculated using the sliding average method; according to the frequency deviation amplitude, deviation recovery rate and instantaneous change rate of the power reverse flow event, the decision tree algorithm is used for model training to construct a potential reverse flow instability event identification model to identify potential reverse flow instability events.

3. The method according to claim 1, wherein The power instability pattern recognition model is constructed based on the key parameters of historical power instability events to identify the instability pattern of potential reverse flow instability events and the risk level of reverse flow instability events where the instability pattern is power instability, including: Through the power instability feature database, the key parameters of historical power instability events are obtained, including the amplitude, duration and frequency characteristics of current fluctuations, and the power instability event stages and power instability event risk levels are marked. The random forest algorithm is used for model training to construct a power instability pattern recognition model. The power instability event stages include instability precursors and power instability. The power instability event risk levels include normal, warning, and dangerous. The instability modes include normal reverse flow, instability precursors, and power instability. Based on the key parameters of potential reverse flow instability events, the power instability pattern recognition model is used to identify the instability mode of potential reverse flow instability events and identify the reverse flow instability event risk level when the instability mode is power instability. The potential reverse current instability event whose instability mode is a precursor to instability is marked as a precursor to instability event, and the potential reverse current instability event whose instability mode is power instability is marked as a reverse current instability event; The reverse power instability events with risk level labels are transmitted to the power instability event monitoring database, and a multi-level early warning classification mechanism is used to determine the response level of the reverse power instability events.

4. The method according to claim 1, wherein The real-time monitoring of the reverse flow fluctuation frequency, the load change frequency and the rising rate of the variable load reverse resonance factor, and the formulation of a dynamic damping coefficient adjustment plan and a resonance early warning plan include: Real-time monitoring of the reverse flow fluctuation frequency f and load change frequency g, if the frequency ratio is detected If the ratio is greater than the preset threshold, a dynamic adjustment plan for the damping coefficient is formulated and implemented to adjust the damping coefficient Dynamically adjust to preset critical damping coefficient ; Calculate and record the rising rate of the time-varying load reverse resonance factor in real time. If the rising rate exceeds the preset rate threshold, formulate and implement a resonance early warning plan, including limiting the load at key nodes of the power grid or cutting off designated nodes to control the load flow and transfer part of the load of the current node to adjacent nodes or backup power supply nodes with a load lower than the preset load threshold.

5. The method according to claim 1, wherein The method of obtaining real-time load data and historical load fluctuation patterns of key nodes of the power grid where the reverse flow impact load exceeds the limit, calculating the load deviation value of each node, determining the load amount that needs to be transferred to each node, and formulating a load control plan includes: Acquire real-time load data of key nodes of the power grid where the reverse impact load exceeds the limit, including load fluctuation amplitude, change rate and fluctuation frequency; extract historical load fluctuation patterns of key nodes of the power grid where the reverse impact load exceeds the limit through the power instability feature database, including load peak, average load level and fluctuation period; compare real-time load data with historical data, calculate the load deviation value of each node, and obtain the real-time-historical load deviation characteristics of key nodes of the power grid; utilize adaptive control algorithm to dynamically adjust control parameters according to the real-time-historical load deviation characteristics of key nodes of the power grid, determine the amount of load that needs to be transferred to each node, and obtain an adaptive control parameter set; formulate a load control plan based on the adaptive control parameter set to transfer the load of key nodes of the power grid where the reverse impact load exceeds the limit, and give priority to allocating the excess load to adjacent nodes or backup power supply nodes whose load is lower than the preset load threshold.

6. The method according to claim 1, wherein The method predicts the key parameters of the instability precursor event in a preset future time period based on the historical key parameters of the potential reverse current instability event based on the instability mode as the instability precursor event, and predicts the time point when the instability precursor event turns into the reverse current instability event in combination with the power instability pattern recognition model, and implements the damping coefficient dynamic adjustment plan and the resonance early warning plan, including: Through the power instability characteristic database, the historical key parameters of the instability precursor events are obtained and arranged in time series to construct the time series data of the historical key parameters of the instability precursor events; Based on the time series data of historical key parameters of instability precursor events, long short-term memory networks are used for model training to predict the key parameters of instability precursor events within a preset time period in the future. Based on the prediction results of the key parameters of instability precursor events within the preset time period in the future, combined with the power instability pattern recognition model, the time point when the instability precursor event turns into a reverse power instability event and the risk level of the reverse power instability event are predicted. Based on the predicted time point when the instability precursor event turns into a reverse power instability event, a reverse power instability event warning is sent, and a dynamic adjustment plan for the damping coefficient and a resonance warning plan are implemented.

7. The method according to claim 1, wherein The method of obtaining load fluctuation responses of key nodes of the power grid, evaluating the stability of the power grid system after implementing the load control scheme, the damping coefficient dynamic adjustment scheme, and the resonance early warning scheme, and optimizing the control parameters based on the control feedback data includes: Continuously monitor the real-time load data of the power system after the implementation of the load control plan, the damping coefficient dynamic adjustment plan and the resonance early warning plan, obtain the load fluctuation response and dynamic change data of the damping coefficient at key nodes of the power grid, and evaluate the stability of the power grid system after the implementation of each plan; Record feedback data from load control, damping coefficient adjustment, and resonance warning responses, including load changes, the ratio of reverse flow fluctuation frequency to load change frequency, and resonance warning effects, to determine the effectiveness of each scheme in alleviating power instability. Based on the effectiveness of power instability mitigation, optimize the load control scheme, damping coefficient dynamic adjustment scheme, and resonance warning scheme, and adjust the load distribution, control direction, and critical damping coefficient. Re-verify the optimized load control scheme, damping coefficient dynamic adjustment scheme and resonance warning scheme, and record the load adjustment amount, critical damping coefficient, warning execution time and load distribution direction of each key node of the power grid. If the power instability mitigation effect of each scheme meets the expectations, the optimized load control scheme, damping coefficient dynamic adjustment scheme and resonance warning scheme will be stored in the power system control database.

Images