A cloud edge fusion railway emergency power supply scheduling method

Abstract

The application discloses a cloud edge fusion railway emergency power supply scheduling method, relates to the technical field of railway power supply and power dispatching control, and comprises the following steps: S1, collecting real-time power data and voltage signals in the process of emergency power supply access through an edge node sensor of a railway power supply system, and classifying and identifying the collected data based on a support vector machine algorithm to determine an initial power disturbance mode corresponding to the emergency power supply access and generate a power disturbance feature vector for scheduling decision; the cloud edge fusion railway emergency power supply scheduling method improves the safety, reliability and continuous operation capacity of the railway emergency power supply system under complex working conditions as a whole.

Description

Technical Field

[0001] This invention relates to the field of railway power supply and power dispatch control technology, specifically to a cloud-edge fusion railway emergency power dispatch method. Background Technology

[0002] In the railway transportation system, the traction power supply system is a crucial infrastructure for ensuring the safe operation of trains and transportation efficiency. To cope with abnormal operating conditions such as sudden power outages, natural disasters, or critical equipment failures, the railway system is usually equipped with emergency power supplies to maintain the continuous operation of traction power and related loads when the main power supply is limited. With the expansion of the railway network and the increase in operational density, the rapid access and effective dispatch of emergency power supplies have become key factors affecting the safety and stability of railway power supply.

[0003] Existing railway emergency power dispatching methods mainly include two modes: centralized cloud dispatching and edge node independent control. While centralized cloud dispatching possesses global resource coordination capabilities, it is susceptible to communication latency and network reliability issues, making it difficult to respond promptly to rapid power disturbances generated during emergency power supply access. Edge independent control, limited by local information, lacks comprehensive awareness of adjacent traction power supply networks and the overall power resource status, easily leading to incoordination in dispatching decisions. In a cloud-edge fusion architecture, edge nodes are responsible for the nearest access and execution control of emergency power supplies, while the cloud provides global resource information and dispatching strategy support. However, when emergency power supplies access the traction power supply network, they cause instantaneous changes in power flow, voltage levels, and system frequency, characterized by strong suddenness and complex dynamic features. If the power disturbance patterns during the access process cannot be effectively identified and analyzed, it can easily lead to amplified voltage fluctuations, short-term overloads, or malfunctions of protection devices, thus affecting normal train operation. Furthermore, in complex scenarios with multiple loads and multiple emergency power supplies coexisting, the load status of the local traction power supply network is constantly changing. If the dispatching system cannot dynamically adjust dispatching parameters based on global power resource data from the cloud, it will be difficult to balance local stability with overall coordinated operation. Meanwhile, existing scheduling methods have limited means of assessing the effectiveness of scheduling execution and the risk of malfunctions, making it difficult to promptly determine the effectiveness of scheduling paths and the level of power supply continuity. Summary of the Invention

[0004] The purpose of this invention is to provide a cloud-edge fusion railway emergency power dispatching method to solve the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a cloud-edge fusion railway emergency power supply scheduling method, comprising: S1, collecting real-time power data and voltage signals during the emergency power supply access process through edge node sensors of the railway power supply system, and classifying and identifying the collected data based on a support vector machine algorithm to determine the initial power disturbance mode corresponding to the emergency power supply access, and generating a power disturbance feature vector for scheduling decisions; S2, obtaining frequency change information from adjacent traction power supply networks based on the power disturbance feature vector, and when the frequency change exceeds a preset scheduling threshold, performing Kalman filtering smoothing on the frequency signal to determine the emergency power supply scheduling threshold. S3. Based on the voltage fluctuation amplitude and the corresponding disturbance residual signal, dynamically schedule and adjust the load distribution status of the local traction power supply network according to the voltage fluctuation amplitude and the disturbance residual signal, and integrate the global power supply resource data in the cloud through the internal computing module of the edge node to generate the reverse power amplitude setting parameters and pulse width adjustment control parameters required for emergency power dispatch; S4. Obtain the reverse power amplitude setting parameters and pulse width adjustment control parameters, and download the pre-established emergency power dispatch model from the cloud edge fusion channel, and generate the corresponding power disturbance compensation signal sequence and the timing synchronization requirements for reverse power pulse injection based on the dispatch model.

[0006] Preferably, step S1 includes collecting real-time power data and voltage signals during the emergency power supply access process through edge node sensors of the railway power supply system to obtain an initial data set; classifying and identifying the initial data set using a support vector machine algorithm, where the input of the support vector machine algorithm is the feature values ​​of the real-time power data and voltage signals in the initial data set, and the output is a classification label to determine the power disturbance type; determining the initial power disturbance mode corresponding to the emergency power supply access based on the power disturbance type to obtain a disturbance mode description; extracting key features from the disturbance mode description, including power fluctuation amplitude and voltage change rate, to generate a power disturbance feature vector; and generating a feature representation for scheduling decisions based on the power disturbance feature vector, where the feature representation is a vector combination of key features.

[0007] Preferably, step S2 includes obtaining frequency change information from adjacent traction power supply networks based on the power disturbance feature vector to obtain a frequency change sequence; if the frequency change sequence exceeds a preset scheduling threshold, the frequency signal is smoothed by Kalman filtering. The Kalman filter calculates the predicted value and uncertainty of the frequency signal through a prediction step, and adjusts the predicted value by fusing the measured value in an update step to obtain a smoothed frequency signal; the voltage fluctuation amplitude caused by emergency power dispatch is determined by the smoothed frequency signal to obtain a voltage fluctuation description; a disturbance residual signal is extracted from the voltage fluctuation description, and the disturbance residual signal is calculated based on the difference between the smoothed frequency signal and the original frequency change sequence to obtain a residual signal sequence; and a disturbance adjustment vector for load balancing of traction substations is generated for the residual signal sequence.

[0008] Preferably, step S3 includes: obtaining the load distribution status from the local traction power supply network based on the voltage fluctuation amplitude and disturbance residual signal; locating unevenly distributed areas by comparing amplitude and residual signals to obtain a distribution status description; performing dynamic scheduling adjustments based on the distribution status description, transferring some power from load areas above a preset threshold to load areas below a preset threshold to obtain adjusted load balancing indicators; fusing global power supply resource data from the cloud and adjusted load balancing indicators through the internal computing module of the edge node, matching the resource data and indicators to the corresponding backup node locations, and extracting backup node monitoring information; determining the reverse power amplitude setting parameters required for emergency power dispatch from the backup node monitoring information, assigning values ​​to the power gap portion in the monitoring information, and generating an amplitude parameter vector; and using the amplitude parameter vector to adjust the pulse width control parameters, mapping the amplitude values ​​in the vector to the width range to obtain pulse width adjustment control parameters.

[0009] Preferably, step S4 includes downloading a pre-established emergency power scheduling model from the cloud edge fusion channel; determining model parameters by fusing global power supply data through a preset resource matching method to obtain the emergency power scheduling model; fusing reverse power amplitude setting parameters and pulse width adjustment control parameters through the emergency power scheduling model; generating a compensation signal sequence corresponding to the power disturbance using a parameter vector matching method; extracting reverse power generation information based on the compensation signal sequence; determining the timing synchronization requirements for pulse injection using an information comparison method; mapping the timing synchronization requirements to a preset parameter vector; adjusting the vector interval to obtain the timing synchronization judgment result; and integrating backup node resource matching based on the timing synchronization judgment result to generate a signal sequence extraction output.

[0010] Preferably, the method further includes S5: based on the compensation signal sequence and the timing synchronization requirements of the reverse power pulse injection, controlling the inverter drive at the edge node to execute the scheduling command for injecting reverse power pulses, and continuously adjusting the pulse energy precision metering and the injection channel isolation protection status through a real-time power feedback closed loop. Specifically, this includes obtaining the timing synchronization requirements from the compensation signal sequence, generating scheduling commands for edge node control, executing the injection of reverse power pulses through the inverter drive, and obtaining the initial execution result of the scheduling command; integrating real-time power feedback based on the initial execution result of the scheduling command, using a closed-loop adjustment mechanism to calculate the deviation by comparing the feedback value with the set value and iteratively adjusting the injection parameters to determine the pulse energy metering parameters; extracting injection channel isolation data from the pulse energy metering parameters, obtaining protection status monitoring information, determining whether the isolation protection status is stable by comparing the monitoring information with a preset stability threshold, and obtaining the isolation protection status adjustment sequence; integrating power disturbance compensation through the isolation protection status adjustment sequence, using a multi-node resource coordination method to optimize emergency response integration by synchronizing resource data through the allocation node communication protocol, and obtaining the compensated channel isolation configuration; mapping the compensated channel isolation configuration to emergency response integration, integrating backup power data to generate the final pulse energy precision metering result and the injection channel isolation protection status.

[0011] Preferably, the method further includes S6: performing matching analysis on the compensation signal sequence based on real-time monitoring data after reverse power pulse injection; determining that the local power supply response delay estimation is complete when the waveform correlation calculation result exceeds a preset scheduling matching threshold, and determining the corresponding residual signal amplitude analysis result. Specifically, this includes obtaining the compensation signal sequence from the real-time monitoring data after reverse power pulse injection; calculating the correlation value between the compensation signal sequence and the standard waveform template using the Pearson correlation coefficient to obtain the waveform correlation calculation result; comparing the waveform correlation calculation result with the preset scheduling matching threshold; determining that the local power supply response delay estimation is complete if the correlation value exceeds the threshold, and identifying a delay estimation completion indicator; extracting the response time parameter based on the delay estimation completion indicator; obtaining the initial value of the residual signal amplitude by calculating the difference between the actual response time and the theoretical response time; calculating the degree of deviation between the initial value of the residual signal amplitude and the preset standard amplitude range; determining the compensation deviation correction sequence by evaluating the degree of deviation; the compensation deviation correction sequence includes an amplitude adjustment factor and a phase adjustment factor; and performing linear correction processing on the residual signal using the amplitude adjustment factor and phase adjustment factor in the compensation deviation correction sequence to obtain the final residual signal amplitude analysis result and the local power supply response delay estimation state.

[0012] Preferably, the method further includes S7: calculating the quantification value of the risk level of malfunction caused by scheduling actions based on the multi-channel fusion of residual signal amplitude analysis results and monitoring data; when the quantification value of the risk level of malfunction is lower than a preset threshold, recording and obtaining the operation status log of the railway power supply system to determine the activation status of the emergency power dispatch path; specifically, obtaining multi-channel monitoring data from the residual signal amplitude analysis results, obtaining an integrated signal sequence through weighted average fusion processing; calculating the quantification value of the risk level of malfunction caused by scheduling actions for the integrated signal sequence, obtaining the risk quantification value through statistical deviation summation; if the risk quantification value is lower than the preset threshold, recording the operation status log of the railway power supply system to obtain a log data set; determining the activation status of the emergency power dispatch path based on the log data set, and determining the activation status result.

[0013] Preferably, it also includes S8: matching local power supply response sensing records within the confirmed time window based on the comparison results of the operation status log, and determining the railway power supply continuity index after emergency power dispatch. Specifically, this includes obtaining the comparison result confirmation from the operation status log, obtaining the local power supply response sequence by summing the time series comparison deviations, matching sensing records within the time window for the local power supply response sequence, and obtaining the integrated response set by weighted fusion of power supply load balancing records.

[0014] Preferably, step S8 further includes calculating the activation degree of the emergency power dispatch path based on the integrated response set, obtaining the activation degree value by accumulating statistical deviations, and determining the continuous path of railway power supply if the activation degree is higher than a preset threshold; obtaining the continuity index through the continuous path of railway power supply, and using the average load deviation to calculate and judge the power supply stability level after emergency power dispatch.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects: This cloud-edge fusion railway emergency power dispatching method achieves accurate identification of power disturbance patterns and dynamic generation of dispatching parameters by collecting power and voltage data in real time from edge nodes during emergency power access and combining this data with global power supply resource information from the cloud for collaborative analysis. This enables effective prediction and control of potential voltage and frequency disturbances from the initial stage of emergency power access. Simultaneously, by introducing a disturbance-characteristic-based dispatching model and a reverse power pulse injection mechanism, it achieves rapid execution and closed-loop feedback of dispatching commands while ensuring the stability of the local traction power supply network, effectively reducing the probability of voltage fluctuation amplification, short-term overload, and malfunction of protection devices. Furthermore, this invention quantifies the dispatching execution effect and malfunction risk, and determines the activation status of the dispatching path and power supply continuity indicators by combining operational status logs. This makes the emergency power dispatching process verifiable, evaluable, and traceable, thereby comprehensively improving the safety, reliability, and continuous operation capability of the railway emergency power supply system under complex operating conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart of the railway emergency power dispatching method based on cloud-edge fusion of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1As shown, this invention provides a technical solution: a cloud-edge fusion railway emergency power supply scheduling method, including: S1, collecting real-time power data and voltage signals during the emergency power supply access process through edge node sensors of the railway power supply system, classifying and identifying the collected data based on the support vector machine algorithm, determining the initial power disturbance mode corresponding to the emergency power supply access, and generating a power disturbance feature vector for scheduling decisions; S2, obtaining frequency change information from adjacent traction power supply networks based on the power disturbance feature vector, and performing Kalman filtering smoothing on the frequency signal when the frequency change exceeds a preset scheduling threshold to determine the voltage fluctuation amplitude and corresponding disturbance residual signal caused by emergency power supply scheduling; S3, dynamically adjusting the load distribution status of the local traction power supply network based on the voltage fluctuation amplitude and disturbance residual signal, and fusing global power supply resource data from the cloud through the internal computing module of the edge node to generate the reverse power amplitude setting parameters and pulse width adjustment control parameters required for emergency power supply scheduling; S4, obtaining the reverse power amplitude setting parameters and pulse width adjustment control parameters, and downloading the pre-established emergency power supply scheduling model from the cloud-edge fusion channel. The following steps are taken: S5. Based on the scheduling model, a compensation signal sequence for the corresponding power disturbance and the timing synchronization requirements for the injection of reverse power pulses are generated; S6. According to the compensation signal sequence and the timing synchronization requirements for the injection of reverse power pulses, the inverter drive at the edge node is controlled to execute the scheduling command for injecting reverse power pulses, and the pulse energy is continuously adjusted for precise metering and the isolation protection status of the injection channel through a real-time power feedback closed loop; S7. Based on the real-time monitoring data after the injection of reverse power pulses, the compensation signal sequence is matched and analyzed. When the correlation calculation result of the sequence waveform exceeds the preset scheduling matching threshold, it is determined that the local power supply response delay estimation is completed, and the corresponding residual signal amplitude analysis result is determined; S8. Based on the multi-channel fusion of the residual signal amplitude analysis result and the monitoring data, the quantification value of the malfunction risk level caused by the scheduling action is calculated. When the quantification value of the malfunction risk level is lower than the preset threshold, the operation status log of the railway power supply system is recorded and obtained to determine the activation status of the emergency power supply scheduling path; S9. According to the comparison result of the operation status log, the local power supply response perception record is matched within the confirmed time window to determine the railway power supply continuity index after emergency power supply scheduling.

[0019] In the above implementation, the method is based on a cloud-edge collaborative architecture, which integrates the real-time sensing capabilities of the railway power supply system with the global scheduling capabilities of the cloud. Edge nodes acquire transient data of emergency power supply access through high-frequency sampling power and voltage sensors, and use support vector machine algorithms to classify and identify power disturbance patterns, thereby mapping complex and nonlinear access disturbances into power disturbance feature vectors that can be used for scheduling decisions.

[0020] Based on this, potential system-level disturbances are jointly assessed by associating frequency change information with adjacent traction power supply networks. When the frequency change exceeds a preset threshold, Kalman filtering is used to smooth the frequency signal, suppressing measurement noise and extracting the true voltage fluctuation components caused by emergency power dispatch. Furthermore, based on the voltage fluctuation amplitude and disturbance residual signal, the load distribution status of the local traction power supply network is dynamically evaluated. The edge node computing module integrates global power supply resource data from the cloud to form reverse power amplitude setting parameters and pulse width adjustment control parameters. Subsequently, combined with the pre-established emergency power dispatch model in the cloud-edge fusion channel, a compensation signal sequence and the timing synchronization requirements for reverse power pulse injection are generated. Through precise control of the edge node inverter, controlled injection of reverse power pulses is achieved. Relying on a real-time power feedback closed-loop mechanism, the pulse energy metering and injection channel isolation protection status are continuously corrected, thereby ensuring the stability and safety of the dispatch process. Finally, through sequence matching and multi-channel fusion analysis of the injected monitoring data, a comprehensive assessment of power supply response delay, residual signal, and malfunction risk is achieved.

[0021] By introducing support vector machines and Kalman filtering algorithms at edge nodes, rapid identification and accurate modeling of railway emergency power supply access disturbances were achieved, significantly improving the response speed and accuracy of scheduling decisions to transient power disturbances. Through a cloud-edge fusion mechanism, local real-time perception was combined with global power supply resource information from the cloud, effectively avoiding the risk of misjudgment from a single-node scheduling perspective. A combination of reverse power pulse injection and compensation signal sequences was used to actively suppress voltage fluctuations caused by emergency power supply scheduling, improving the stability and continuity of the traction power supply system in emergency scenarios. Simultaneously, through the quantification of malfunction risk levels and the recording of operational status logs, a traceable and quantifiable data foundation was provided for determining the activation status of scheduling paths and power supply continuity indicators, thereby enhancing the safety and reliability of the railway power supply system.

[0022] S1 includes collecting real-time power data and voltage signals during the emergency power supply access process through edge node sensors of the railway power supply system to obtain an initial data set; using a support vector machine (SVM) algorithm to classify and identify the initial data set, with the input of the SVM algorithm being the feature values ​​of the real-time power data and voltage signals in the initial data set, and the output being a classification label to determine the power disturbance type; determining the initial power disturbance mode corresponding to the emergency power supply access based on the power disturbance type to obtain a disturbance mode description; extracting key features from the disturbance mode description, including power fluctuation amplitude and voltage change rate, to generate a power disturbance feature vector; and generating a feature representation for scheduling decisions based on the power disturbance feature vector, with the feature representation being a vector combination of key features.

[0023] In this embodiment, after the emergency power supply connection process begins, the edge node sensors of the railway power supply system synchronously collect real-time power data and voltage signals at the same time, and write one record obtained from each sampling into an initial data set. The initial data set consists of multiple records arranged in timestamp order, where the real-time power data is taken from the power value output by the sensor, the voltage signal is taken from the voltage value output by the sensor, and the timestamp is taken from the recording time of the edge node for this sampling, to ensure a one-to-one correspondence between power and voltage at the same sampling point. Subsequently, classification and recognition are performed on the initial data set. Specifically, the real-time power data value and voltage signal value in each record are used as feature values ​​of that record and input into the support vector machine algorithm. The support vector machine algorithm calls a pre-trained and solidified classification parameter set to perform discrimination calculation on the input feature values. The classification parameter set is obtained by training historical initial data set samples and their known power disturbance type labels. During the training process, the interval maximization segmentation boundary is solved for samples of different power disturbance types, and the solution is solidified into a discriminant parameter set that can be directly called, so that a classification label is output for each record in the classification stage. The classification label is represented by an integer number and is associated with the power disturbance. A one-to-one correspondence is established between the categories. This correspondence remains fixed after training and is stored along with the classification parameter set. The process for determining the classification label is as follows: After performing discrimination calculations for each record and obtaining the discrimination value, the discrimination value is compared with a threshold of 0. The threshold of 0 is taken as the discrimination benchmark value of the segmentation boundary to ensure that the output label is consistent with the segmentation boundary. When the power perturbation type is multi-class, pairwise discrimination is performed on the same record, and the output labels are counted. The label with the largest count is taken as the final classification label of that record. After obtaining the classification label of each record in the initial dataset, the process for determining the power perturbation type is to perform pairwise discrimination on all records. The final classification labels are counted, and the power disturbance type corresponding to the label with the largest count is taken as the power disturbance type of this emergency power supply access. The initial power disturbance mode is determined based on this power disturbance type, and a disturbance mode description is generated. The disturbance mode description uses the power disturbance type as the mode category identifier, and writes the key features calculated from the initial data set. The calculation process of the key features is as follows: the power fluctuation amplitude is obtained by subtracting the minimum value from the maximum value of all real-time power data in the initial data set. The maximum and minimum values ​​are obtained by iterating through the initial data set one by one and updating the current maximum power and the current minimum power, respectively.The voltage change rate is determined by taking the maximum value from the rate sequence obtained by dividing the absolute value of the voltage difference between adjacent records in the initial dataset by the difference in the timestamps of adjacent records. Specifically, the voltage values ​​of two adjacent records are taken one by one, the difference is calculated, and the absolute value is taken. Then, the time interval is obtained by subtracting the timestamp of the previous record from the timestamp of the next record. This division is performed to obtain the rate of the adjacent interval. All adjacent intervals are traversed, and the maximum rate is retained as the voltage change rate, thus ensuring that the voltage change rate reflects the most drastic voltage change segment during the access process. After obtaining the power fluctuation amplitude and voltage change rate, the power fluctuation amplitude is written first, followed by the voltage change rate, in a fixed order to form a power disturbance feature vector. Subsequently, the power disturbance feature vector is used to generate a feature representation for scheduling decisions. The generation process of the feature representation involves combining the power disturbance feature vector with the integer number of the classification label corresponding to the power disturbance type. The combination method is to append the integer number to the end of the feature vector to form a data item of fixed length, so that the feature representation is consistent with the data type. The system carries information on disturbance intensity and disturbance category. Among the parameters mentioned above, real-time power data and voltage signals are directly acquired by edge node sensors. The initial data set is accumulated from all sampling records during the access process according to timestamps. Feature values ​​are directly composed of the power and voltage values ​​in each sampling record. Classification labels are output by the support vector machine algorithm after discriminative calculation of the feature values ​​and encoded with fixed integer numbers. The power disturbance type is determined by the largest number of classification labels in the initial data set. The initial power disturbance mode is directly determined by the power disturbance type. The disturbance mode description is composed of mode category identifiers and key features. The power fluctuation amplitude and voltage change rate are determined by the process of traversing to find the extreme value and taking the maximum rate of adjacent differences, respectively. The power disturbance feature vector is written into the key features in a fixed order. The feature representation is determined by adding the classification label integer number to the vector combination. The only step involving a threshold is the classification discrimination threshold 0, which is determined by the discrimination benchmark of the segmentation boundary and completed during training.

[0024] S2 includes obtaining frequency change information from adjacent traction power supply networks based on the power disturbance feature vector to obtain a frequency change sequence; if the frequency change sequence exceeds a preset scheduling threshold, the frequency signal is smoothed by Kalman filtering. The Kalman filter calculates the predicted value and uncertainty of the frequency signal through a prediction step, and adjusts the predicted value by fusing the measured value in an update step to obtain a smoothed frequency signal; the voltage fluctuation amplitude caused by emergency power dispatch is determined by the smoothed frequency signal to obtain a voltage fluctuation description; the disturbance residual signal is extracted from the voltage fluctuation description, and the disturbance residual signal is calculated based on the difference between the smoothed frequency signal and the original frequency change sequence to obtain a residual signal sequence; and a disturbance adjustment vector for load balancing of traction substations is generated for the residual signal sequence.

[0025] In this embodiment, the frequency change information acquisition process of adjacent traction power supply networks is initiated based on the obtained power disturbance feature vector. Specifically, the edge node continuously reads the frequency measurement results of adjacent traction power supply networks according to a predetermined sampling period, and stores the frequency values ​​corresponding to each sampling moment in a frequency change sequence in chronological order. The frequency change sequence consists of multiple frequency values ​​obtained from continuous sampling, and the order remains unchanged. The preset scheduling threshold is determined through a fixed process before the system is put into operation. The determination process involves collecting multiple frequency change sequences under the condition that emergency power dispatch is not performed and the power supply system is in normal operation, calculating the absolute value of the frequency change for each sequence point by point, summarizing them, and taking the maximum value as the threshold. A preset scheduling threshold is set and written into the threshold configuration item of the edge node as a basis for subsequent judgment, thereby ensuring that smoothing processing is only initiated when a frequency change exceeds the maximum normal operating range in the frequency change sequence. When the frequency change sequence is determined to exceed the preset scheduling threshold, Kalman filtering smoothing processing is performed on the frequency signal. The Kalman filtering is performed point-by-point, and each sampling point strictly executes prediction and update steps. Before the arrival of the current sampling point, the prediction step determines the predicted value of the frequency signal at the current sampling point based on the smoothed frequency signal at the previous sampling point, and generates the uncertainty of the current sampling point based on the uncertainty of the previous sampling point. The uncertainty is used to characterize the reliability of the predicted value relative to the true value. The confidence level is determined and the predicted value is entered into the update step along with the predicted value after the prediction step is completed. The update step reads the measured value of the current sampling point after it arrives. The measured value is taken from the frequency value of the frequency change sequence at the same sampling point. Then, the measured value and the predicted value are fused and a smoothed frequency signal is output. The fusion process determines the influence ratio of the predicted value and the measured value based on uncertainty, so that the side with less uncertainty receives a higher influence ratio. Simultaneously, the uncertainty is updated for use in the next sampling point, thus obtaining a smoothed frequency signal covering the entire frequency change sequence. The process of determining the voltage fluctuation amplitude caused by emergency power dispatch based on the smoothed frequency signal is carried out within the same analysis window. The system iterates through the smoothed frequency signal point by point and records the maximum and minimum smoothed frequency values. The difference between the maximum and minimum smoothed frequency values ​​is taken as the voltage fluctuation amplitude. The voltage fluctuation amplitude and its corresponding analysis window together form a voltage fluctuation description for subsequent residual extraction. The process of extracting the disturbance residual signal from the voltage fluctuation description is as follows: for each sampling point in the analysis window, the smoothed frequency signal value and the original frequency change sequence value are read, the difference is calculated point by point, and the residual signal sequence is stored in the sampling order. The disturbance residual signal is characterized by the residual signal sequence and its value is completely determined by the difference between the smoothed frequency signal and the original frequency change sequence at the same sampling point.The process of generating a disturbance adjustment vector for load balancing in traction substations based on the residual signal sequence involves iterating through the residual signal sequence point by point and extracting the variation information of the residual amplitude over time. This variation information is then organized into an ordered list of values ​​according to the number of components required for load balancing, serving as the disturbance adjustment vector. The number of components is determined by the number of loads participating in load balancing at the traction substation and remains fixed within the substation. The component values ​​are generated consistently based on the residual amplitude variation within the same analysis window of the residual signal sequence, ensuring that the disturbance adjustment vector directly reflects the demand of the disturbance residue for load balancing adjustment. Among the parameters, the power disturbance feature vector is output from the previous step and serves as the trigger and correlation basis for this step. The frequency variation information is obtained from adjacent traction power supply networks and is determined by frequency variation. The data is stored in sequence format. The preset scheduling threshold is determined and fixed by the maximum absolute value of the normal operating frequency change. The predicted value is determined by the smoothed frequency signal of the previous sampling point. Uncertainty is generated by the prediction step and used in the update step to determine the fusion ratio. The measured value is determined by the frequency value of the current sampling point in the frequency change sequence. The smoothed frequency signal is output by the update step. The voltage fluctuation amplitude is determined by the difference between the maximum and minimum values ​​of the smoothed frequency signal within the analysis window. The voltage fluctuation description consists of the voltage fluctuation amplitude and its corresponding analysis window. The disturbance residual signal is determined by the point-by-point difference between the smoothed frequency signal and the original frequency change sequence and stored as a residual signal sequence. The disturbance adjustment vector is determined by organizing the residual amplitude change information of the residual signal sequence within the same analysis window according to the number of components required for load balancing.

[0026] S3 includes: obtaining the load distribution status from the local traction power supply network based on voltage fluctuation amplitude and disturbance residual signal; locating uneven distribution areas by comparing amplitude and residual to obtain a distribution status description; performing dynamic scheduling adjustments based on the distribution status description, transferring some power from load areas above a preset threshold to load areas below a preset threshold to obtain adjusted load balancing indicators; fusing global power supply resource data from the cloud and adjusted load balancing indicators through the internal computing module of the edge node, matching the resource data and indicators to the corresponding backup node locations, and extracting backup node monitoring information; determining the reverse power amplitude setting parameters required for emergency power dispatch from the backup node monitoring information, assigning values ​​to the power gap portion in the monitoring information, and generating an amplitude parameter vector; and using the amplitude parameter vector to adjust the pulse width control parameters, mapping the amplitude values ​​in the vector to the width range to obtain pulse width adjustment control parameters.

[0027] In this embodiment, the voltage fluctuation amplitude and disturbance residual signal are directly taken from the output of the previous step. The voltage fluctuation amplitude is the amplitude representation value of the voltage fluctuation within the same analysis window, and the disturbance residual signal is the residual signal sequence obtained by point-by-point difference between the smoothing result and the original frequency change sequence within the same analysis window. When obtaining the load distribution status of the local traction power supply network based on the voltage fluctuation amplitude and disturbance residual signal, the edge node reads the load records of each load area from the local traction power supply network within the same time window as the previous step, and records the load amount of each load area within that time window. The load quantities are arranged in chronological order to form a load sequence for each load region. All load sequences from different load regions together constitute the load distribution state. When locating unevenly distributed areas, the edge node first sums the load quantities of all load regions within the same time window at each sampling moment and divides this sum by the number of load regions to obtain the average load quantity of the entire network at that sampling moment. Then, for each load region, the difference between its load quantity and the average load quantity of the entire network is calculated, and the absolute value is taken to obtain the load deviation sequence of that load region at each sampling moment. Simultaneously, the absolute value of the disturbance residual signal sequence is taken point by point to obtain the residual intensity sequence, and this sequence is then determined within the time window. The maximum residual strength and its occurrence time are determined. Then, for each load area, the load deviation value of the load area is read at the time when the maximum residual strength occurs, and the load deviation value is compared with the voltage fluctuation amplitude. When the load deviation value is greater than the deviation criterion corresponding to the voltage fluctuation amplitude, the load area is identified as an uneven distribution area and written into the distribution state description. The deviation criterion uses the voltage fluctuation amplitude itself as the comparison benchmark to ensure that the positioning basis is entirely derived from the limited voltage fluctuation amplitude and residual signal. The distribution state description consists of the identified uneven distribution area identifier, the load sequence of the load area in the time window, the maximum residual strength and its occurrence time, and serves as the input for subsequent scheduling adjustment. When performing dynamic scheduling adjustment based on the distribution state description, a preset threshold is first determined. The preset threshold is calculated offline and solidified through a fixed process before the system is put into operation. Specifically, the load distribution state is collected within one consecutive operating cycle, the average load of the entire network is calculated at each sampling time, and the maximum value of the average load of the entire network within the operating cycle is taken as the preset threshold and written into the threshold configuration, so that the threshold corresponds to the upper limit of the average load of the entire network during normal operation.During scheduling adjustments, the average load sequence of each load region within the current time window is taken as the current load of that load region. The current load is then compared with a preset threshold. Load regions with a current load greater than the preset threshold are identified as load regions above the preset threshold, and load regions with a current load less than the preset threshold are identified as load regions below the preset threshold. Power transfer calculations and updates are then performed. Power transfer is implemented using a deterministic iterative approach: for each load region above the preset threshold, its excess is calculated as the difference between the current load of that load region and the preset threshold; for each load region below the preset threshold, its excess is calculated as follows: Calculate the shortfall amount, which is the difference between a preset threshold and the current load amount of the load area. Then, process the high-load area and low-load area sequentially according to the load area number. The amount of power transferred from the high-load area to the low-load area is the smaller of the remaining excess power in the current high-load area and the remaining shortfall amount in the current low-load area in each transfer. After the transfer, the excess power in the high-load area is reduced simultaneously, and the shortfall amount in the low-load area is reduced simultaneously, until the excess power in all high-load areas is 0 or the shortfall amount in all low-load areas is 0. After completion, the adjusted load amount of each load area is obtained, and the adjusted load balancing index is obtained accordingly. The load deviation is calculated and summarized for each load area within a time window. Specifically, for each load area, the difference between the adjusted load and a preset threshold is taken as the absolute value as the load balance deviation value for that load area. The load balance deviation values ​​of all load areas are then added together to obtain the load balance index. This index value represents the degree of balance and is written into the scheduling results. Subsequently, when the edge node's internal computing module integrates the global power supply resource data from the cloud and the adjusted load balance index, the edge node's internal computing module reads the global power supply resource data from the cloud. The global power supply resource data provides the location identifier of each standby node, its available power information, and the range identifier of the load area that the standby node can support in the form of resource entries. The edge node's internal computing module matches the low load area gap information corresponding to the adjusted load balance index with the range identifier of the load area that can support in the resource entries one by one. The matching rule is that when the range identifier of the load area that can support in the resource entry contains the low load area identifier with a gap, the location of the standby node corresponding to the resource entry is determined as the standby node location and recorded. Then, the standby node monitoring information is extracted from the resource entry corresponding to the recorded standby node location. The standby node monitoring information consists of the available power information of the standby node and its current load-related information.When determining the reverse power amplitude setting parameters based on backup node monitoring information, the edge node first calculates the corresponding power gap for each backup node. The power gap is the sum of the gaps that the backup node can support in all low-load areas that are still not filled after adjustment. The gap amount comes directly from the remaining gap amount in the low-load area after the aforementioned power transfer iteration. Then, the power gap is compared with the power information that the backup node can provide. When the power gap is greater than the power that can be provided, the reverse power amplitude setting is taken as the power that can be provided. When the power gap is not greater than the power that can be provided, the reverse power amplitude setting is taken as the power gap. The reverse power amplitude setting corresponding to each backup node is written sequentially according to the recording order of the backup node position to form an amplitude parameter vector. Each amplitude value in the amplitude parameter vector completes the assignment of the power gap part in the monitoring information. When adjusting the pulse width control parameter using the amplitude parameter vector, the edge node's internal calculation module pre-stores the minimum pulse width. The system includes allowable width values, maximum allowable width values, and a width interval boundary table. The minimum and maximum allowable width values ​​are determined and fixed by scheduling control parameters before system commissioning. The width interval boundary table is arranged in order from the minimum to the maximum allowable width values ​​and remains unchanged. The mapping process is performed item by item on the amplitude parameter vector: first, the maximum amplitude value is determined in the amplitude parameter vector and used as the upper bound of the mapping; then, the relative position between the current amplitude value and the upper bound is calculated, and the corresponding width interval is selected accordingly. When the current amplitude value is 0, the minimum allowable width value is selected; when the current amplitude value equals the upper bound of the mapping, the maximum allowable width value is selected; when the current amplitude value is between the two, the width value corresponding to the width interval into which its relative position falls is selected. The obtained width values ​​are then written into the pulse width adjustment control parameters according to the original order of the amplitude parameter vector, thus ensuring that the pulse width adjustment control parameters are determined by the amplitude parameter vector and are consistent with the preset width interval.

[0028] S4 includes downloading a pre-established emergency power dispatch model from the cloud edge fusion channel, determining model parameters by fusing global power supply data through a preset resource matching method, and obtaining the emergency power dispatch model; fusing reverse power amplitude setting parameters and pulse width adjustment control parameters through the emergency power dispatch model, and generating a compensation signal sequence for the corresponding power disturbance using a parameter vector matching method; extracting reverse power generation information from the compensation signal sequence, and determining the timing synchronization requirements for pulse injection using an information comparison method; mapping the timing synchronization requirements to a preset parameter vector, and adjusting the vector interval to obtain the timing synchronization judgment result; integrating backup node resource matching based on the timing synchronization judgment result, and generating a signal sequence extraction output.

[0029] In this implementation, a pre-established emergency power scheduling model is first sent from the cloud-edge fusion channel to the edge nodes. Upon receiving the model, the edge nodes perform an integrity check on the model file. This check includes file length consistency verification and checksum consistency verification. The checksum is provided synchronously by the cloud-edge fusion channel during the sending process and is recalculated and compared on the edge node side according to the same verification rules. If the check passes, the model is loaded as a callable emergency power scheduling model. Subsequently, global power supply data is fused using a preset resource matching method to determine the model parameters. This global power supply data is sent from the cloud-edge fusion channel and includes the available power capacity corresponding to each backup node location. The power and availability records are matched using a pre-defined resource matching method fixed before system commissioning: a one-to-one correspondence between the standby node location and the corresponding record in the global power supply data. The matching process involves edge nodes sequentially searching for records with the same location in the global power supply data, following the order of standby node locations. If a record is found, its available power and availability status are extracted and written into the model parameters. If not found, the corresponding model parameter is set to an unavailable state and written into the model. This results in a set of model parameters, which is then injected into the emergency power dispatch model, ensuring that the model operation is strictly constrained by the current global power supply data. Then, the reverse power is fused through the emergency power dispatch model. The power amplitude setting parameters and pulse width adjustment control parameters generate a compensation signal sequence. The reverse power amplitude setting parameters are taken from the amplitude parameter vector output from the previous step. This amplitude parameter vector consists of multiple amplitude values ​​arranged in the order of the backup node positions, with each amplitude value determined by the power gap portion in the monitoring information of the corresponding backup node. The pulse width adjustment control parameters are taken from the width parameter vector output from the previous step. This width parameter vector consists of multiple width values ​​arranged in the same order as the amplitude parameter vector, with each width value determined by mapping the corresponding amplitude value to a pre-configured width range. The parameter vector matching method is fixed as "phase" before system commissioning. The combination rule is "the amplitude value and width value with the same sequence number form a control pair". The combination process is to start from the first sequence number, and pair the amplitude parameter vector and the width parameter vector with the same sequence number to form a control pair sequence while keeping the sequence number order unchanged. Then, the control pair sequence is input into the emergency power dispatch model. The model calculates the compensation control result for each control pair in combination with the injected model parameter set. The output content includes the compensation amplitude and compensation duration corresponding to the segment. The edge nodes write each segment output into the compensation signal sequence according to the order of the control pair sequence, so that the compensation signal sequence corresponds to the compensation process required for this power disturbance segment by segment.Next, the reverse power generation information is extracted from the compensation signal sequence, and the timing synchronization requirements for pulse injection are determined. The extraction process involves reading the compensation amplitude and duration of each segment of the compensation signal sequence and recording them as the reverse power target information for that segment. Simultaneously, the sequential position of the segment in the sequence is recorded as the basis for comparison and positioning. The information comparison method is fixed as a two-condition comparison rule: "comparing the compensation amplitude with the amplitude parameter vector to match the amplitude value of the sequence number, and comparing the compensation duration with the width parameter vector to match the duration corresponding to the width value of the sequence number." The comparison process is performed segment by segment number, and the pulse is only transferred when both comparisons are true. A segment is identified as an executable synchronization segment, and its start and end times are written into the timing synchronization requirements. The start time is obtained by accumulating the duration of each preceding segment from the first segment of the compensation signal sequence, and the end time is obtained by accumulating the duration of the current segment based on the start time. This ensures that the timing synchronization requirements are calculated deterministically by accumulating the compensation signal sequence and are consistent with the parameters. Subsequently, the timing synchronization requirements are mapped to a preset parameter vector to obtain the timing synchronization judgment result. The mapping process uses the start and end times of each executable synchronization segment in the timing synchronization requirements as time boundaries, and the control sequence is sequentially mapped according to time. The sequence is divided into multiple vector intervals, each corresponding to a synchronization segment. The adjustment of these vector intervals involves sequentially assigning execution times to the control pairs within each interval. The execution time of the first control pair is equal to the start execution time of the synchronization segment. Subsequently, the execution time of each control pair is equal to the execution time of the previous control pair plus its corresponding compensation duration. After assignment, it is checked whether the end execution time of the last control pair within the interval is less than or equal to the end execution time of the synchronization segment. If so, the interval is marked as synchronized; otherwise, it is marked as not synchronized. All interval markings are summarized to form the timing synchronization judgment result. Finally, based on the timing synchronization judgment results, the backup node resource matching is integrated to generate a signal sequence extraction output. The integration process is to perform backup node resource matching consistency check only on the vector intervals that have passed synchronization. The check rule is fixed as "the compensation amplitude of each segment in the interval is not greater than the available power of the same backup node position in the model parameters and the available status of the position is available". The check process compares the compensation amplitude and available power segment by segment according to the segment number in the interval and checks the available status at the same time. When all segments meet the requirements, the compensation signal sequence segment corresponding to the interval and the timing synchronization requirement segment corresponding to the interval are extracted and spliced ​​in the original time order to form the signal sequence extraction output.Among the parameters mentioned above, the emergency power dispatch model is issued by the cloud-edge fusion channel and determined after integrity verification. Global power supply data is issued by the cloud-edge fusion channel and serves as the data source for the model parameters. Resource matching methods, parameter vector matching methods, information comparison methods, vector interval segmentation rules, and interval adjustment rules are all configured before system commissioning and remain unchanged during operation. The model parameter set is extracted by matching the backup node location with the corresponding records of the global power supply data. The reverse power amplitude setting parameter is determined by assigning values ​​to the power gap portion in the backup node monitoring information to form the amplitude parameter vector. The pulse width... The degree adjustment control parameters are determined by mapping the amplitude value to a preset width range and forming a width parameter vector. The compensation signal sequence is determined by the model calculating the control sequence segment by segment and writing it in sequence. The reverse power generation information is determined by reading the compensation signal sequence segment by segment, determining the compensation amplitude and compensation duration. The timing synchronization requirement is determined by sequentially accumulating the compensation duration and writing it into the comparison segment. The timing synchronization judgment result is determined by the deterministic comparison between the end execution time after interval adjustment and the end execution time of the synchronization segment. The signal sequence extraction output is determined by splicing the sequence segments that have passed synchronization and resource verification in chronological order.

[0030] S5 includes obtaining timing synchronization requirements from the compensation signal sequence, generating scheduling instructions for edge node control, injecting reverse power pulses through inverter drive execution, and obtaining the initial execution result of the scheduling instructions; based on the initial execution result of the scheduling instructions, integrating real-time power feedback, and using a closed-loop regulation mechanism to calculate the deviation by comparing the feedback value with the set value and iteratively adjusting the injection parameters to determine the pulse energy metering parameters; extracting injection channel isolation data from the pulse energy metering parameters, obtaining protection status monitoring information, and determining whether the isolation protection status is stable by comparing the monitoring information with the preset stability threshold to obtain the adjustment sequence of the isolation protection status; integrating power disturbance compensation through the adjustment sequence of the isolation protection status, and using a multi-node resource coordination method to optimize emergency response integration by synchronizing resource data through the allocation node communication protocol to obtain the compensated channel isolation configuration; mapping the compensated channel isolation configuration to emergency response integration, and integrating backup power data to generate the final accurate pulse energy metering result and the injected channel isolation protection status.

[0031] In this embodiment, the edge node first reads the start and end times of each segment from the compensation signal sequence, and determines the set of these time-ordered start and end times as the timing synchronization requirement. Then, based on the timing synchronization requirement, it generates scheduling instructions for edge node control. The generation process of the scheduling instructions involves establishing a one-to-one correspondence between each segment of the compensation signal sequence, using the start and end times of each segment as the time constraint for that segment's instruction, and simultaneously writing the corresponding reverse power pulse execution information into that segment's instruction. These instructions are then concatenated according to the segment number to form a complete instruction stream. The edge node sends the complete instruction stream to the inverter driver and triggers the injection of reverse power pulses according to the instruction stream sequence. The inverter driver initiates injection at the start of each segment and stops injection at the end of each segment, thus obtaining the initial execution result of the scheduling command. The initial execution result consists of the actual start time of each segment, the actual end time of each segment, and the sampling sequence of real-time power feedback during the execution of each segment, and is archived according to the segment number. Subsequently, the edge node fuses the real-time power feedback based on the initial execution result and performs closed-loop regulation. The real-time power feedback is continuously acquired by the edge node at a fixed sampling period during the injection of each segment and aligned with the setpoint of that segment point by point. The setpoint is taken from the target reverse power information corresponding to that segment in the compensation signal sequence and remains unchanged throughout the entire duration of that segment. After alignment, the deviation is calculated for each sampling point. The calculation rule for the deviation is fixed as subtracting the feedback value from the set value to obtain the deviation value and recording the deviation direction. Then, the injection parameters are iteratively adjusted. The iterative adjustment process is to process the sampling points sequentially within the same segment. When the deviation of a certain sampling point is not 0, the injection parameters are immediately corrected in the next sampling cycle, and new feedback values ​​are collected to form a new deviation record. The injection parameters are taken from the injection control quantity corresponding to that segment in the scheduling instruction, and only this control quantity is continuously corrected during the iteration process. The correction determination rule is fixed as follows: when the feedback value is less than the set value, the injection parameters are corrected in the direction of increasing the injection intensity; when the feedback value is greater than the set value, the injection... The parameters are corrected in the direction of reducing the injection intensity until the end of the segment, forming a complete set of deviation convergence records and corresponding injection parameter correction records. Based on this, the pulse energy measurement parameters are determined. The process of determining the pulse energy measurement parameters is as follows: for each sampling point in the segment, the feedback power value of the sampling point is calculated by adding energy to the duration corresponding to the sampling period to obtain the cumulative injection energy value of the segment. At the same time, the set value of the segment and the same duration are added according to the same accumulation rule to obtain the set injection energy value of the segment. Then, the energy deviation value of the segment is obtained by subtracting the cumulative injection energy value from the set injection energy value. The cumulative injection energy value, the set injection energy value, and the energy deviation value are used together as the pulse energy measurement parameters of the segment.Subsequently, edge nodes extract injection channel isolation data based on pulse energy metering parameters to obtain protection status monitoring information. The extraction process involves reading injection channel isolation data in chronological order consistent with power sampling before, during, and after each injection segment to form a monitoring sequence. This monitoring sequence is then identified as protection status monitoring information. Next, it is determined whether the isolation protection status has stabilized. This determination process involves comparing the value of each sampling point in the protection status monitoring information with a preset stability threshold point by point and generating a stability judgment mark sequence. When a sampling point does not meet the stability condition corresponding to the threshold, it is marked as an unstable point, and its time position and segment number are recorded. All unstable points are summarized in chronological order to obtain the isolation protection status. The adjustment sequence for the isolation protection state; the preset stability threshold is determined through a fixed process before system commissioning. The determination process involves collecting multiple segments of injection channel isolation data under the condition that reverse power pulse injection is not performed and the injection channel isolation is in a stable protection state. The fluctuation range of the collected isolation data is statistically analyzed point by point, and the upper limit of the fluctuation range of the isolation data in the stable state is taken as the stability threshold and written into the threshold configuration. Any fluctuation exceeding this upper limit during operation is judged as unstable. Subsequently, the edge nodes integrate power disturbance compensation through the adjustment sequence of the isolation protection state and optimize emergency response integration using a multi-node resource coordination method. The integration process involves mapping the time position of unstable points in the adjustment sequence to the compensation signal. The sequence is mapped to corresponding segments and time slices, which are then identified as compensation intervals requiring priority coordination. Participating nodes synchronize resource data within each compensation interval according to the inter-node communication protocol. This resource data is taken from each node's currently available resource records and exchanged and acknowledged within the synchronization period specified in the communication protocol. Edge nodes update the resource allocation for compensation execution based on the synchronized resource data, ensuring that compensation execution and isolation protection adjustment requirements are simultaneously met within the same time slice. This yields the compensated channel isolation configuration, which consists of the isolation state configuration results for each compensation interval and their effective time slices, and is fixed in chronological order. Finally, the edge nodes... The compensated channel isolation configuration is mapped to the emergency response integration and fused with backup power data to generate the final result. The fusion process involves aligning the backup power data with time slices and segment numbers and writing it into the emergency response integration record. The compensated channel isolation configuration is used as the injection safety constraint to finally confirm the pulse energy metering parameters. The confirmation rule is that only when the channel isolation configuration of the corresponding time slice of a segment is at a stable pass mark is the cumulative injected energy value and energy deviation value of that segment written into the final summary; otherwise, the segment is marked as requiring priority isolation protection adjustment and written into the final summary. This generates the final accurate pulse energy metering result and outputs the injection channel isolation protection status corresponding one-to-one with the final metering result of each segment.

[0032] S6 includes acquiring a compensation signal sequence from real-time monitoring data after reverse power pulse injection, calculating the correlation value between the compensation signal sequence and the standard waveform template using the Pearson correlation coefficient, and obtaining the waveform correlation calculation result; comparing the waveform correlation calculation result with a preset scheduling matching threshold, and determining that the local power supply response delay estimation is complete if the correlation value exceeds the threshold, and identifying a delay estimation completion marker; extracting response time parameters based on the delay estimation completion marker, and obtaining the initial value of the residual signal amplitude by calculating the difference between the actual response time and the theoretical response time; calculating the degree of deviation of the initial value of the residual signal amplitude from the preset standard amplitude range, and determining a compensation deviation correction sequence by evaluating the degree of deviation, which includes an amplitude adjustment factor and a phase adjustment factor; and performing linear correction processing on the residual signal using the amplitude adjustment factor and phase adjustment factor in the compensation deviation correction sequence to obtain the final residual signal amplitude analysis result and the local power supply response delay estimation status.

[0033] In this embodiment, after the reverse power pulse is injected, the edge node immediately extracts the sampling sequence corresponding to the same time window as the compensation signal sequence from the real-time monitoring data after the injection. The sampling sequence is then truncated and spliced ​​according to the segment number and the start and end times of each segment of the compensation signal sequence to obtain the compensation signal sequence used for matching analysis. The real-time monitoring data consists of continuous power or voltage sampling records after injection, the time window is determined by the start and end times of each segment given by the compensation signal sequence, and the segment number is determined by the writing order of the compensation signal sequence. Subsequently, the comparison between the compensation signal sequence and the standard waveform template is calculated. Waveform correlation is determined by using a standard waveform template based on multiple typical response sequences collected under stable operating conditions for the same type of scheduling action before system commissioning. The determination process involves time alignment and amplitude normalization of the multiple typical response sequences, followed by taking the average shape of each sample point, and then solidifying this average shape as the standard waveform template. The correlation value is calculated using the Pearson correlation coefficient. Specifically, the compensated signal sequence and the standard waveform template are first aligned point-by-point along the same number of sample points. The alignment rule is to use the first sample point of the compensated signal sequence as the alignment starting point and truncate the same number of consecutive sample points as the standard waveform template. Subsequently, the correlation coefficients are calculated... The average value of the compensated signal sequence within the alignment interval and the average value of the standard waveform template within the alignment interval are calculated. Then, for each sampling point, the difference between the value at that sampling point and the average value of the sequence is calculated, forming two sets of difference sequences. These two sets of difference sequences are then multiplied point-by-point, and the products are accumulated point-by-point to obtain a cumulative value of coordinated change. Simultaneously, the two sets of difference sequences are squared point-by-point, and each is accumulated to obtain two sets of cumulative energy values. Finally, the cumulative value of coordinated change is normalized by multiplying the square roots of the two sets of cumulative energy values ​​to obtain a correlation value. This correlation value is the waveform correlation calculation result. A preset scheduling matching threshold is set before system commissioning. The correlation is determined by matching and statistically analyzing the same standard waveform template. The determination process involves using the standard waveform template to calculate multiple sets of correlation values ​​with typical response sequences under multiple stable operating conditions using the same calculation process described above. The minimum correlation value is then taken as the preset scheduling matching threshold and written into the threshold configuration, so that the threshold represents the minimum consistency requirement that can still be met under stable response conditions. During operation, the waveform correlation calculation result is compared with the preset scheduling matching threshold. When the correlation value is strictly greater than the preset scheduling matching threshold, a delay estimation completion flag is immediately generated and set to the completion state, thereby determining that the local power supply response delay estimation is complete.When the delay estimation completion status is marked as complete, the edge node extracts the response time parameter and calculates the initial value of the residual signal amplitude. The response time parameter consists of the actual response time and the theoretical response time. The extraction process of the actual response time involves determining the response start time and the response arrival time within the corresponding time window of the real-time monitoring data. The response start time is taken as the start time of the injection segment, and the response arrival time is taken as the moment when the real-time monitoring data first reaches and continuously satisfies the corresponding arrival condition of the standard waveform template. The arrival condition is fixed by the standard waveform template as "the position of the arrival point in the template and the amplitude threshold corresponding to that position" before commissioning. The extraction process of the theoretical response time involves determining the theoretical start time and the theoretical arrival time within the standard waveform template. The theoretical arrival time and theoretical starting time are taken as the first sampling point time of the template, and the theoretical arrival time is taken as the sampling point time corresponding to the template's arrival condition. Then, the actual response time is obtained by subtracting the response starting time from the response arrival time, and the theoretical response time is obtained by subtracting the theoretical starting time from the theoretical arrival time. Finally, the initial value of the residual signal amplitude is obtained by subtracting the theoretical response time from the actual response time, ensuring a one-to-one correspondence between this initial value and the time difference of the response delay. Next, the deviation between the initial value of the residual signal amplitude and the preset standard amplitude range is calculated, and a compensation deviation correction sequence is determined. The preset standard amplitude range is determined through stable operating condition statistics before system commissioning. The determination process involves collecting data under the condition that the delay estimation is completed and stable determination is achieved. Multiple initial values ​​for the residual signal amplitude are generated. These initial values ​​are sorted by numerical value, and the lower bound is taken as the minimum value and the upper bound as the maximum value, which are then fixed as the upper and lower bounds of the standard amplitude range. During operation, the initial value of the residual signal amplitude is compared with these upper and lower bounds. When the initial value falls between the upper and lower bounds, the deviation is set to 0. When the initial value is less than the lower bound, the deviation is set to the difference between the lower bound and the initial value. When the initial value is greater than the upper bound, the deviation is set to the difference between the initial value and the upper bound. Based on the deviation assessment, a compensation deviation correction sequence is generated. The correction sequence is written in segment sequence, and each segment contains an amplitude adjustment factor and a phase adjustment factor. The rule for determining the amplitude adjustment factor is that it is required to correct the initial value of the residual signal amplitude to the nearest boundary within the standard amplitude range. The proportional adjustment amount and the phase adjustment factor are determined by converting the time offset between the actual response arrival time and the theoretical arrival time into the time correction amount required for phase alignment. Finally, linear correction processing is performed on the residual signal and the result is output. The linear correction processing is implemented segment by segment according to the segment number. First, the amplitude adjustment factor is used to scale the initial value of the residual signal amplitude of the segment to obtain the correction amplitude value. Then, the phase adjustment factor is used to perform time alignment correction on the corresponding response time parameter of the segment to obtain the corrected response time difference. The correction amplitude value and the corrected response time difference are jointly determined as the final residual signal amplitude analysis result. At the same time, the delay estimation completion flag is bound to the correction result and output as the local power supply response delay estimation status.Among the above parameters, the compensation signal sequence is generated by the preceding steps and solidified according to the segment number and start and end times; the real-time monitoring data is obtained by continuous sampling after injection and sorted by timestamp; the standard waveform template is determined by aligning and averaging the typical response sequence under stable operating conditions; the waveform correlation calculation result is determined by the average, difference, multiplicative accumulation, square accumulation and normalization processing within the alignment interval; the preset scheduling matching threshold is determined and solidified by the minimum correlation value obtained by matching the typical response sequence; the delay estimation completion indicator is determined by the comparison result between the correlation value and the threshold; the actual response time and the theoretical response time are determined by the difference between the start time and the arrival time in the real-time monitoring data and the standard waveform template, respectively; the initial value of the residual signal amplitude is determined by the difference between the actual response time and the theoretical response time; the preset standard amplitude range is determined and solidified by the minimum and maximum values ​​of multiple initial values; the deviation degree is determined by the difference between the initial value and the upper and lower bounds of the standard amplitude range; the compensation deviation correction sequence is obtained by mapping the deviation degree and includes amplitude adjustment factor and phase adjustment factor; and the final residual signal amplitude analysis result and the local power supply response delay estimation state are jointly determined by the correction amplitude value after linear correction processing, the difference between the corrected response time and the completion indicator. ;

[0034] S7 includes obtaining multi-channel monitoring data from residual signal amplitude analysis results, and obtaining an integrated signal sequence through weighted average fusion processing; calculating the quantification value of the risk level of malfunctions caused by scheduling actions for the integrated signal sequence, and obtaining the risk quantification value by summing statistical deviations; if the risk quantification value is lower than a preset threshold, recording the operation status log of the railway power supply system and obtaining the log data set; judging the activation status of the emergency power dispatch path based on the log data set and determining the activation status result.

[0035] In this embodiment, the edge nodes first use the time window corresponding to the residual signal amplitude analysis results as a unified index. They then extract sampling sequences within the same time range from each monitoring channel according to the start and end times of this time window, forming multi-channel monitoring data. This multi-channel monitoring data consists of sampling sequences from multiple channels, with each sequence containing chronologically arranged sample values ​​and their corresponding time positions. Subsequently, time alignment is performed on the multi-channel monitoring data. This time alignment is achieved by determining the set of common sampling times within the time window and taking the corresponding sample value for each channel at each common sampling time. If a channel lacks a sample value at a certain common sampling time, then that channel's sample value is taken. The most recent sampled value before a given moment is used as the sampled value at that moment, thus obtaining a set of aligned data with multi-channel sampled values ​​at each common sampling moment. After alignment, a weighted average fusion process is performed to obtain an integrated signal sequence. The weights used in the weighted average fusion process are the weights of each monitoring channel. The channel weights are determined and fixed through stable operation statistics before the system is put into operation. The determination process involves collecting long-term series of each channel under stable operation of the railway power supply system, calculating the fluctuation range of each channel's sampled value relative to its own mean, and determining the weight based on the magnitude of the fluctuation range. Channels with smaller fluctuation ranges are assigned larger weights, and channels with larger fluctuation ranges are assigned smaller weights. The system normalizes all channel weights so that their sum equals 1, and this normalization ensures the normalization remains unchanged during operation. The integrated signal sequence is calculated by multiplying each channel's sampled value by its respective channel weight at each common sampling moment to obtain a weighted value. These weighted values ​​are then summed to obtain the fused value for that moment. The fused values ​​from all common sampling moments are then written into the integrated signal sequence in chronological order. Subsequently, the system calculates the quantification of the risk level of erroneous actions caused by scheduling actions based on the integrated signal sequence. This risk quantification is achieved through statistical deviation summation. First, a baseline value sequence for deviation calculation is determined. This baseline value sequence is obtained before system commissioning by adjusting the values ​​under stable operating conditions. The integrated signal sequence is statistically determined and solidified. The determination process involves generating multiple integrated signal sequences using the same channel extraction, alignment, and weighted average fusion process as during operation, under conditions of stable operation and without emergency power dispatch. The multiple integrated signal sequences are then averaged point by point at the same sampling point to obtain a baseline value sequence, which is then solidified. During operation, the deviation of the integrated signal sequence is calculated point by point. The deviation is the difference between the fused value at that sampling point and the corresponding value of the baseline value sequence at that sampling point, and the absolute value is taken. Then, the absolute deviation values ​​of all sampling points within the time window are accumulated point by point to obtain a risk quantification value. This risk quantification value reflects the overall deviation of the integrated signal from the baseline state within the time window.A preset threshold is used to determine whether the risk quantification value is within the allowable range. This preset threshold is determined and fixed through stable operation statistics before the system is put into operation. The determination process involves calculating multiple risk quantification values ​​under stable operation conditions using the same deviation summation method, sorting these risk quantification values ​​by size, and taking the maximum value as the preset threshold, which is then written into the threshold configuration. This preset threshold corresponds to the upper bound of the risk quantification value under stable operation conditions. During operation, the risk quantification value is compared with the preset threshold. When the risk quantification value is strictly less than the preset threshold, the operation status log is triggered. The recording process involves writing the operation status log of the railway power supply system into the log storage within this time window, forming a log data set. The log data set consists of consecutive log record entries within this time window, and each entry includes the recording time and operation status fields. Finally, the activation status of the emergency power dispatch path is determined based on the log data set, and the activation status result is confirmed. The determination process involves retrieving the status field record corresponding to the emergency power dispatch path in the log data set in chronological order and checking... If the status field changes from inactive to active within the time window and remains continuously active, the activation status result is determined as active; otherwise, it is determined as inactive. Among the parameters, the residual signal amplitude analysis result is output from the preceding steps and a time window is provided. The multi-channel monitoring data consists of the sampling sequences of each monitoring channel within the time window. The channel weights are determined and normalized by the fluctuation amplitudes from stable operation statistics. The integrated signal sequence is calculated point-by-point by summing the aligned multi-channel sampled values ​​according to their weights. The benchmark value sequence is determined by averaging multiple integrated signal sequences under stable operation conditions. The deviation is obtained by taking the absolute value of the difference between the fused value and the benchmark value at corresponding points. The risk quantification value is obtained by accumulating all absolute deviations point-by-point within the time window. The preset threshold is determined by the maximum value of multiple risk quantification values ​​under stable operation conditions. The log data set consists of the operation status log entries written when the threshold condition is met. The activation status result is determined by the change trajectory of the scheduling path status field in the log data set.

[0036] S8 includes obtaining comparison results from the operation status log, obtaining a local power supply response sequence by summing the time series comparison deviations; matching the local power supply response sequence with sensing records within a time window, and using power supply load balancing records for weighted fusion to obtain an integrated response set; calculating the activation degree of the emergency power dispatch path based on the integrated response set, obtaining the activation degree value by accumulating statistical deviations, and determining the continuous path of railway power supply if the activation degree is higher than a preset threshold; obtaining the continuity index through the continuous path of railway power supply, and using the average load deviation to calculate and judge the power supply stability level after emergency power dispatch.

[0037] In this embodiment, the edge node first obtains the comparison result confirmation from the operation status log and forms a local power supply response sequence. Specifically, this is achieved by determining a confirmation time window and extracting the status field time sequence from the operation status log within that time window. The status field time sequence consists of status field values ​​arranged chronologically. Simultaneously, a baseline log status time sequence is collected and fixed under the stable operation state of the railway power supply system before system commissioning. Subsequently, a time sequence comparison is performed between the two time sequences. The comparison process involves first aligning each time position within the time window and then reading the status field value of the operation status log at each time position and finding the baseline value of the baseline log status time sequence at the same time position. Next, calculate the difference value at each time position. The difference value is the absolute value of the difference between the value in the running status log and the baseline value. After calculating the difference values ​​for all time positions, write all the difference values ​​in chronological order to form a difference sequence. Then, sum the differences point by point in the difference sequence to obtain the deviation sum value. The deviation sum value serves as the numerical basis for confirming the comparison results, and the difference sequence is determined as the local power supply response sequence. Subsequently, match the sensing records within the same time window for the local power supply response sequence. The matching process involves searching for local power supply response sensing records point by point within the time window, requiring that the time position of the sensing record be completely consistent with the time position of the local power supply response sequence. The sensing records matched at each time position are then... A set of matching sensing records is assembled in chronological order. Then, a weighted fusion of power supply load balancing records is used to obtain an integrated response set. The weighted fusion process involves simultaneously reading the sensing record value from the matching sensing record set and the corresponding power supply load balancing record value at each time point, and configuring a weighting coefficient for the power supply load balancing record. This weighting coefficient is determined and fixed through stable operation statistics before system commissioning. The determination process involves collecting multiple segments of power supply load balancing records under stable operating conditions and calculating their fluctuation amplitude. Records with smaller fluctuation amplitudes are assigned larger weighting coefficients, and normalization is performed to make the weighting coefficients constant and reusable. During fusion, the sensing record value and the power supply load balancing record are compared at each time point. The values ​​are weighted according to fixed weights and summed to obtain the integrated response value at that time position. The integrated response values ​​of all time positions are arranged in chronological order to form an integrated response set. Then, the activation degree of the emergency power dispatch path is calculated based on the integrated response set to determine the continuous path of railway power supply. The activation degree is calculated by statistical deviation accumulation. Specifically, before the system is put into operation, a set of benchmark integrated response sets is collected and solidified under stable operating conditions. During operation, the integrated response set is aligned with the benchmark integrated response set at each time position. A deviation term is calculated at each time position. The deviation term is the difference between the running integrated response value and the benchmark integrated response value and the absolute value is taken. Then, the deviation terms of all time positions are accumulated point by point to obtain the activation degree value.A preset threshold is used to determine whether the activation value meets the conditions for a continuous path. This preset threshold is determined and fixed before the system is put into operation through statistical analysis of multiple known scheduling path activation scenarios. The determination process involves repeatedly calculating the activation value under multiple known activation scenarios, sorting the resulting set of activation values ​​by numerical value, and taking the minimum activation value as the preset threshold and writing it into the threshold configuration. This ensures that any activation value higher than this threshold corresponds to the activation characteristics under known activation scenarios. When the activation value is strictly greater than the preset threshold during operation, the continuous path of railway power supply is determined to be established, and its corresponding time window is recorded. Finally, the continuity index is obtained through the continuous path of railway power supply, and the power supply stability level is judged. The continuity index is calculated using the average load deviation. Specifically, within the time window corresponding to the establishment of a continuous railway power supply path, the power supply load record after emergency power dispatch and the balancing benchmark load record are extracted. The balancing benchmark load record is collected and fixed under stable operating conditions before the system is put into operation. Subsequently, the load deviation is calculated at each time position. The load deviation is taken as the difference between the load record value after dispatch and the balancing benchmark load record value, and the absolute value is taken. Then, the absolute values ​​of all load deviations within the time window are summed and divided by the number of time positions to obtain the average load deviation. The average load deviation is determined as the continuity index, and the value of the continuity index characterizes the power supply stability level after emergency power dispatch. In the parameters described above, the running status log is the set of log data generated in step S7; the confirmation time window is determined by the time range of the running status log and is used in this step; the baseline log status time series is collected and solidified from stable operation and used as the comparison benchmark; the difference value is obtained by taking the absolute value of the difference between the log value at each time position and the baseline value; the deviation summation value is obtained by accumulating the difference values ​​point by point and is used as the basis for confirming the comparison result; the local power supply response sequence is directly determined by the difference sequence; the sensing record is the local power supply response sensing record within the same time window and matched according to time position; the power supply load balancing record is the load balancing record at the same time position and participates in weighted fusion; the weighted fusion... The coefficients are determined and fixed by statistical analysis of stable operation fluctuations. The integrated response set is composed of weighted fusion results from each time point in chronological order. The benchmark integrated response set is fixed by stable operation data collection. The activation value is obtained by summing the absolute values ​​of the deviations between the integrated response set and the benchmark integrated response set point by point. The preset threshold is determined and fixed by the minimum value of the activation value set under known activation scenarios. The continuous path of railway power supply is determined by comparing the activation value with the preset threshold. The power supply load record is the load data after scheduling. The balanced benchmark load record is the fixed benchmark for stable operation. The average load deviation is obtained by summing the absolute values ​​of load deviations and dividing by the number of time points, and is output as a continuity indicator.

[0038] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud-edge fusion method for railway emergency power dispatching, characterized in that, include: S1. Real-time power data and voltage signals during the emergency power supply access process are collected through edge node sensors of the railway power supply system. The collected data are classified and identified based on the support vector machine algorithm to determine the initial power disturbance mode corresponding to the emergency power supply access and generate a power disturbance feature vector for scheduling decision-making. S2. Based on the power disturbance feature vector, obtain frequency change information from the adjacent traction power supply network. When the frequency change exceeds the preset scheduling threshold, perform Kalman filtering smoothing on the frequency signal to determine the voltage fluctuation amplitude and the corresponding disturbance residual signal caused by emergency power dispatch. S3. Based on the voltage fluctuation amplitude and disturbance residual signal, dynamically schedule and adjust the load distribution status of the local traction power supply network, and integrate global power supply resource data from the cloud through the internal computing module of the edge node to generate the reverse power amplitude setting parameters and pulse width adjustment control parameters required for emergency power dispatch. S4. Obtain the reverse power amplitude setting parameters and pulse width adjustment control parameters, and download the pre-established emergency power scheduling model from the cloud edge fusion channel. Based on the scheduling model, generate the corresponding power disturbance compensation signal sequence and the timing synchronization requirements for reverse power pulse injection.

2. The railway emergency power dispatching method based on cloud-edge fusion according to claim 1, characterized in that: S1 includes: Real-time power data and voltage signals during the emergency power supply connection process are collected by edge node sensors of the railway power supply system to obtain an initial data set; The support vector machine algorithm is used to classify and identify the initial dataset. The input of the support vector machine algorithm is the feature value of the real-time power data and voltage signal in the initial dataset, and the output is the classification label to determine the type of power disturbance. Determine the initial power disturbance mode corresponding to the emergency power supply connection based on the power disturbance type, and obtain the disturbance mode description; Key features are extracted from the disturbance mode description, including power fluctuation amplitude and voltage change rate, to generate a power disturbance feature vector; Feature representations for scheduling decisions are generated from power perturbation feature vectors, and these feature representations are based on vector combinations of key features.

3. The railway emergency power dispatching method based on cloud-edge fusion according to claim 1, characterized in that: S2 includes: Based on the power disturbance feature vector, frequency change information is obtained from adjacent traction power supply networks to obtain a frequency change sequence; If the frequency change sequence exceeds the preset scheduling threshold, the frequency signal is smoothed by Kalman filtering. Kalman filtering calculates the predicted value and uncertainty of the frequency signal through the prediction step, and the update step fuses the measured value to adjust the predicted value to obtain a smooth frequency signal. By smoothing the frequency signal, the voltage fluctuation amplitude caused by emergency power dispatch is determined, and a voltage fluctuation description is obtained; The disturbance residual signal is extracted from the voltage fluctuation description. The disturbance residual signal is calculated based on the difference between the smoothed frequency signal and the original frequency change sequence to obtain the residual signal sequence. For the residual signal sequence, a disturbance adjustment vector is generated for load balancing of the traction substation.

4. The railway emergency power dispatching method based on cloud-edge fusion according to claim 1, characterized in that: S3 includes: Based on the voltage fluctuation amplitude and disturbance residual signal, the load distribution status is obtained from the local traction power supply network. By comparing the amplitude and residual, the uneven distribution area is located, and the distribution status description is obtained. Dynamic scheduling adjustments are made based on the distribution status description, transferring some power from load areas with loads above a preset threshold to load areas with loads below a preset threshold, thereby obtaining the adjusted load balancing index. By integrating global power supply resource data from the cloud and adjusted load balancing indicators through the internal computing module of the edge node, the resource data and indicators are matched with the corresponding backup node locations, and the backup node monitoring information is extracted. The reverse power amplitude setting parameters required for emergency power dispatch are determined from the standby node monitoring information. Values ​​are assigned to the power gap part in the monitoring information to generate an amplitude parameter vector. The pulse width control parameter is adjusted by using an amplitude parameter vector. The amplitude value in the vector is mapped to the width range to obtain the pulse width adjustment control parameter.

5. The railway emergency power dispatching method based on cloud-edge fusion according to claim 1, characterized in that: S4 includes: Download the pre-established emergency power scheduling model from the cloud edge fusion channel, and determine the model parameters by merging global power supply data through a preset resource matching method to obtain the emergency power scheduling model; By integrating the reverse power amplitude setting parameters and pulse width modulation control parameters through the emergency power dispatch model, a compensation signal sequence for the corresponding power disturbance is generated using the parameter vector matching method. The reverse power generation information is extracted from the compensation signal sequence, and the timing synchronization requirements of pulse injection are determined by information comparison. The timing synchronization requirement is mapped to a preset parameter vector, and the timing synchronization judgment result is obtained by adjusting the vector range. Based on the timing synchronization judgment results, the backup node resources are integrated and matched to generate a signal sequence for extraction and output.

6. The railway emergency power dispatching method based on cloud-edge fusion according to claim 1, characterized in that, It also includes S5, which controls the inverter drive at the edge node to execute the scheduling command for injecting reverse power pulses based on the timing synchronization requirements of the compensation signal sequence and reverse power pulse injection, and continuously adjusts the pulse energy accuracy metering and injection channel isolation protection status through real-time power feedback closed loop, specifically including: The timing synchronization requirements are obtained from the compensation signal sequence, scheduling instructions are generated for edge node control, and the reverse power pulse is injected through the inverter drive to obtain the initial execution result of the scheduling instructions. Based on the initial execution results of the scheduling instructions and real-time power feedback, a closed-loop regulation mechanism is adopted to calculate the deviation by comparing the feedback value with the set value and iteratively adjusting the injection parameters to determine the pulse energy metering parameters. For pulse energy metering parameters, injection channel isolation data is extracted, protection status monitoring information is obtained, and it is determined whether the isolation protection status is stable. By comparing the monitoring information with the preset stability threshold, the adjustment sequence of the isolation protection status is obtained. By integrating power disturbance compensation through the adjustment sequence of isolation protection status, and by using a multi-node resource coordination method to optimize emergency response integration through the synchronization of resource data via the communication protocol between allocated nodes, a compensated channel isolation configuration is obtained. The emergency response is integrated by mapping the compensated channel isolation configuration, and the backup power data is merged to generate the final accurate pulse energy measurement result and the injected channel isolation protection status.

7. The railway emergency power dispatching method based on cloud-edge fusion according to claim 6, characterized in that, It also includes S6, which performs matching analysis on the compensation signal sequence based on real-time monitoring data after the reverse power pulse injection. When the correlation calculation result of the sequence waveform exceeds the preset scheduling matching threshold, it is determined that the local power supply response delay estimation is complete, and the corresponding residual signal amplitude analysis result is determined, specifically including: The compensation signal sequence is obtained from the real-time monitoring data after the reverse power pulse injection. The correlation between the compensation signal sequence and the standard waveform template is calculated using the Pearson correlation coefficient to obtain the waveform correlation calculation result. The waveform correlation calculation result is compared with the preset scheduling matching threshold. If the correlation value exceeds the threshold, the local power supply response delay estimation is determined to be completed, and the delay estimation completion indicator is determined. Based on the delay estimation, the response time parameter is extracted from the identifier, and the initial value of the residual signal amplitude is obtained by calculating the difference between the actual response time and the theoretical response time. The degree of deviation between the initial value of the residual signal amplitude and the preset standard amplitude range is calculated. The degree of deviation is evaluated to determine the compensation deviation correction sequence, which includes an amplitude adjustment factor and a phase adjustment factor. The residual signal is linearly corrected by using the amplitude adjustment factor and phase adjustment factor in the compensation deviation correction sequence to obtain the final residual signal amplitude analysis result and the local power supply response delay estimation state.

8. The railway emergency power dispatching method based on cloud-edge fusion according to claim 7, characterized in that, It also includes S7, which uses multi-channel fusion of residual signal amplitude analysis results and monitoring data to calculate the quantification value of the risk level of malfunctions caused by scheduling actions. When the quantification value of the malfunction risk level is lower than a preset threshold, the operation status log of the railway power supply system is recorded and obtained to determine the activation status of the emergency power dispatch path. Specifically, this includes: Multi-channel monitoring data are obtained from the residual signal amplitude analysis results, and a combined signal sequence is obtained through weighted average fusion processing. The risk level of erroneous actions caused by scheduling actions is quantified by calculating the integrated signal sequence, and the risk quantification value is obtained by summing the statistical deviations. If the risk quantification value is lower than the preset threshold, the operation status log of the railway power supply system is recorded, and the log data set is obtained. The activation status of the emergency power dispatch path is determined based on the log data set, and the activation status result is confirmed.

9. A railway emergency power dispatching method based on cloud-edge fusion according to claim 8, characterized in that, This also includes S8, which, based on the comparison results of the operation status log, matches the local power supply response perception records within the confirmed time window to determine the railway power supply continuity indicators after emergency power dispatch, specifically including: The comparison results are confirmed by obtaining the operation status log, and the local power supply response sequence is obtained by summing the time series comparison deviations. For local power supply response sequences, sensing records are matched within a time window, and a weighted fusion of power supply load balancing records is used to obtain an integrated response set.

10. A railway emergency power dispatching method based on cloud-edge fusion according to claim 9, characterized in that: S8 further includes: The activation degree of the emergency power dispatch path is calculated based on the integrated response set. The activation degree value is obtained by accumulating statistical deviations. If the activation degree is higher than the preset threshold, the continuous railway power supply path is determined. The continuity index is obtained by using the continuous path of railway power supply, and the stability level of power supply after emergency power dispatch is judged by calculating the average load deviation.

Images