A substation centralized relay protection system based on offline GPU supercomputing and tripled independent ring network

Abstract

This invention discloses a centralized relay protection system for substations based on offline GPU supercomputing and a triple-isolated independent ring network, relating to the field of power system relay protection technology. The system includes a triple-isolated independent sampling unit, three sets of physically isolated ring networks, an offline GPU supercomputing unit, a nanosecond-level hardware timescale generation unit, a transformer fault identification and early warning unit, a maintenance adaptive degradation unit, a station-wide protection timing coordination module, a pure hardware 3-out-of-2 output matrix, and a circuit breaker tripping execution unit. This invention employs a triple-layered physical isolation communication architecture with independent power supply, independent optical fiber, and independent switch, achieving full-link redundancy in sampling and transmission. The GPU unit is deployed in the power safety zone I and configured with a unidirectional optical gate for physical isolation, eliminating external network access and remote control risks. It achieves synchronization accuracy of ≤5ns, timekeeping accuracy of ±20ns / h, and timekeeping capability of ≥24 hours, relying on hardware time stamp and time backoff alignment technology, completely independent of GPS / BeiDou. It uses a power-loss-safe pure hardware AND / OR gate array to implement a 2-out-of-3 output logic with an action time of ≤3ms, without software or CPU involvement. The system supports ≤10ms non-disruptive switching for maintenance adaptive degradation and uninterrupted expansion of new bays with dual-zone isolation, possessing advantages such as no protection blind spots, uninterrupted maintenance, and configuration without affecting real-time operation. This invention can significantly improve the reliability, real-time performance, security, and operational flexibility of substation relay protection, and is suitable for centralized protection and control scenarios in high-reliability smart substations of various voltage levels.

Description

Technical Field

[0001] This invention relates to the fields of power system relay protection, smart substations, and power system automation. Specifically, it relates to a centralized relay protection system for substations based on offline GPU supercomputing parallel processing, triple-redundant physical ring network communication, hardware time-stamped synchronization, and pure hardware power failure safety exit. It is applicable to centralized protection and control of the entire substation, triple-redundant operation, maintenance without interruption, and interval expansion without power interruption in substations with voltage levels of 110kV and above. Background Technology

[0002] Current substation relay protection systems are developing towards centralization, station-localization, and collaboration, but existing technologies generally suffer from the following technical bottlenecks: Traditional centralized protection systems often use a single / dual network architecture, lacking triple physical isolation capabilities. They are prone to network-wide communication interruptions due to single-point faults or common-mode faults, resulting in insufficient protection reliability. Existing protection devices rely on GPS / BeiDou satellite timing. After timing is interrupted, the synchronization accuracy drops sharply and there is no ability to keep up with high accuracy for a long time. Protection operations mostly use a CPU serial architecture, which has insufficient parallel computing capabilities and is difficult to support the real-time processing of massive electrical quantities across multiple intervals throughout the entire station. Most protection exits use software logic or CPU to control, which poses risks such as program crashes, communication blockages, computational delays, and malfunctions due to power failure. Maintenance and expansion of protection devices require the removal of protection and power outages for construction, which creates protection blind spots and affects the reliability of power supply. Some centralized systems lack security partitioning and physical isolation, posing security risks such as network attacks and unauthorized access. Summary of the Invention

[0003] 1. Purpose of the invention To address the problems of existing technologies, such as reliance on satellite synchronization, insufficient timekeeping capability, insufficient communication redundancy, limited computing power, low export reliability, blind spots in maintenance, power outages required for expansion, and weak network security, this invention provides a centralized relay protection system that is highly reliable, highly accurate, highly secure, can operate offline, and supports adaptive maintenance downgrades and uninterrupted expansion. 2. Technical Solution A centralized relay protection system for substations based on offline GPU supercomputing and a triple independent ring network comprises: a triple independent sampling unit, three sets of physically isolated ring networks, an offline GPU supercomputing unit, a nanosecond-level hardware time stamp generation unit, a transformer fault identification and early warning unit, a maintenance adaptive degradation unit, a station-wide protection timing coordination module, a pure hardware 3-out-of-2 output matrix, a circuit breaker tripping execution unit, and offline configuration, offline simulation verification, and online one-click loading units. The triple independent sampling unit consists of sampling units A, B, and C. The three sampling channels are physically isolated from each other, independently powered, and independently grounded. They respectively collect analog voltage and current signals from each bay of the substation, realizing redundancy and fault isolation in the sampling process. Three sets of physically isolated ring networks are connected one-to-one with the triple independent sampling units. Each ring network is designed with independent optical fiber, independent switch and independent power supply to realize triple redundant transmission of sampling data and avoid the impact of single ring network failure, lightning strike and short circuit on the overall operation of the system. The offline GPU supercomputing unit is deployed in the power safety zone I. It is equipped with a one-way optical gate to achieve physical one-way isolation. It is not connected to the external network, so there is no risk of remote control or unauthorized upgrades. It adopts a multi-GPU parallel computing architecture to realize parallel operation of electrical quantities, fault diagnosis, logical decision-making and data interaction of multiple intervals throughout the station. The nanosecond-level hardware time stamp generation unit assigns a hardware-level nanosecond timestamp to each sampled data stream. Combined with dynamic delay compensation and time back-off alignment algorithms, it achieves a synchronization accuracy of ≤5ns, a timekeeping accuracy of ±20ns / h, and a timekeeping capability of ≥24 hours, completely independent of GPS / BeiDou satellite time synchronization. The instrument transformer fault detection and early warning unit monitors the validity of the sampled data in real time, identifies faults such as instrument transformer abnormalities, open circuits, and saturation, and issues early warning signals in advance to improve the system's fault tolerance. The maintenance adaptive degradation unit automatically switches the working mode according to the system operating status, with a switching time of ≤10ms for seamless switching: three normal sets → two out of three to operate; one abnormal set → one out of two to operate; two abnormal sets → single set to operate + alarm, with no protection blind spots throughout the process and maintenance without interruption. The station-wide protection timing coordination module uniformly schedules the timing of protection actions in all bays of the station, enabling cross-bay fault linkage, multi-level protection coordination, and station-wide collaborative control. The pure hardware 3-out-of-2 output matrix uses a power-loss safe hardware AND-OR gate array to implement the logic Y=AB+AC+BC. There is no software, CPU or communication involved. The action response time is ≤3ms. It remains locked in the power-loss state and does not malfunction, ensuring the absolute reliability of the protection output. The system adopts a dual-zone isolation architecture of offline configuration zone and online operation zone, supports offline configuration of new intervals, offline simulation verification, and online one-click loading and deployment, and enables rapid access of primary equipment without power interruption. The access time for a single interval is ≤4 hours, and the configuration process does not occupy real-time task resources or affect the original protection operation. 3. Beneficial effects Triple physical isolation ring network: independent power supply, independent optical fiber, and independent switch, completely eliminating the risk of single-point failure and common-mode failure in single / dual network communication; Offline GPU + Secure I-zone + One-way optical shutter: High computing power, no external network attacks, no risk of remote control, and data security level meets the highest requirements of the power grid; Hardware time stamping + time backoff + long-term timekeeping: ≤5ns synchronization accuracy, timekeeping ≥24 hours, completely eliminating dependence on satellite timekeeping; Pure hardware power failure safety exit: action ≤3ms, no false activation during power failure, no risk of software failure, and the highest level of reliability for the protection exit; ≤10ms Non-disruptive maintenance degradation: Maintenance without interruption or blind spots, significantly improving the reliability of power grid supply; Dual-zone isolation uninterrupted power supply expansion: New bay access requires no power outage and does not affect real-time operation, adapting to the efficient operation and maintenance needs of smart substations; Triple redundancy across the entire link: redundancy throughout the sampling, transmission, computation, and output stages, significantly improving system availability and security stability. 4. Description of the attached drawings Figure 1 This is a block diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the hardware time stamp + GPU time back-off synchronization principle of the present invention; Figure 3 This is a schematic diagram of the three-out-of-two hard-contact output circuit of the present invention; Figure 4 This is a flowchart of the adaptive degradation logic for maintenance in this invention. Figure 5 This is a flowchart of the new bay expansion and uninterrupted power supply renovation process of the present invention. In the diagram: 101-Triple Independent Sampling Unit A, 102-Triple Independent Sampling Unit B, 103-Triple Independent Sampling Unit C, 104-Three sets of physically isolated ring networks, 105-Offline GPU supercomputing unit, 106-Instrument transformer fault identification and early warning unit, 107-Maintenance adaptive degradation unit, 108-Full-site protection timing coordination module, 109-Pure hardware 2 / 3 output matrix, 110-Circuit breaker trip execution unit. 201-ADC sampling unit, 202-Nanosecond-level hardware time stamp generation unit, 203-Three-ring network transmission channel, 204-GPU receiving unit, 205-Delay measurement unit, 206-Delay dynamic compensation unit, 207-Time backoff alignment unit, 208-Synchronous sampling data unit, 209-Parallel protection calculation unit. 301 - Action Signal A, 302 - Action Signal B, 303 - Action Signal C, 304 - Pure Hardware AND / OR Gate Array, 305 - Trip Output Unit. 401 - Process Start-up Unit, 402 - Number of Running Units Judgment Unit, 403 - Two-out-of-three Running Mode Unit, 404 - One-out-of-two Running Mode Unit, 405 - Single Unit Operation + Alarm Unit, 406 - Protection Output Execution Unit. 501 - Offline Configuration Unit, 502 - Offline Simulation Verification Unit, 503 - Online One-Click Loading Unit, 504 - Put into Operation Unit. 5. Detailed Implementation 5.1 System Overall Architecture Implementation ( Figure 1 ) The triple independent sampling units 101, 102, and 103 collect voltage and current signals, which are then transmitted to the offline GPU supercomputing unit 105 via three sets of physically isolated ring networks 104. After parallel computation, the GPU unit outputs the signals to the transformer fault discrimination and early warning unit 106, the maintenance adaptive degradation unit 107, and the whole-station protection timing coordination module 108. Finally, after logical voting by the pure hardware 3-out-of-2 output matrix 109, the circuit breaker trip execution unit 110 is activated. The system operates entirely offline, with triple redundancy and physical isolation, providing the highest level of safety and reliability. 5.2 Hardware time stamping and time rollback are implemented simultaneously ( Figure 2 ) The ADC sampling unit 201 collects analog signals, which are then marked by the nanosecond-level hardware time stamp generation unit 202. The data is transmitted to the GPU receiving unit 204 via the three-ring network transmission channel 203. Precise data synchronization is achieved through delay measurement 205, dynamic compensation 206, and time back-off alignment 207. Finally, the data enters the parallel protection operation unit 209 to complete the protection calculation. This implementation has a synchronization accuracy of ≤5ns, a timekeeping accuracy of ±20ns / h, a timekeeping capability of ≥24 hours, and is completely independent of GPS / BeiDou. 5.3 Pure hardware three-out-of-two exit implementation ( Figure 3 ) Action signals A301, B302, and C303 are input to a pure hardware AND-OR gate array 304, which performs a two-out-of-three voting according to the logic Y=AB+AC+BC, and outputs to the trip output unit 305. The entire process involves no software, no CPU, and no communication, with an action time of ≤3ms. It does not malfunction when the power is lost and the interlocking is disabled, achieving the highest safety level of the power system. 5.4 Implementation of Adaptive Degradation for Maintenance ( Figure 4 ) After system startup, the running set judgment unit 402 monitors the channel status in real time: three sets are normal → execute 403 three-out-of-two mode; one set is abnormal → execute 404 two-out-of-one mode; two sets are abnormal → execute 405 single set operation + alarm; finally, the unified output is output by the protection output execution unit 406. The entire process switching time is ≤10ms, with seamless switching, no blind spots, and no interruption of operation, without affecting the continuous operation of protection. 5.5 Uninterrupted power supply expansion implementation ( Figure 5 ) The new bay access adopts the process of offline configuration 501 → offline simulation verification 502 → online one-click loading 503 → commissioning 504, which realizes safe and fast access without power interruption of equipment, single bay access ≤4 hours, and the dual-zone isolation architecture does not affect the real-time protection operation.

Claims

1. A centralized relay protection system for substations based on offline GPU supercomputing and a triple independent ring network, characterized in that, include: The system comprises three independent sampling units: a physically isolated sampling unit A, a sampling unit B, and a sampling unit C, used to collect analog voltage and current signals from the substation; three physically isolated ring networks, each with independent optical fiber, independent switch, and independent grounding design, connected one-to-one with sampling units A, B, and C respectively, achieving complete electrical isolation and fault isolation transmission of each sampled signal; an offline GPU supercomputing unit deployed in the power safety zone I, configured with a unidirectional optical shutter for physical unidirectional isolation, without external network access or remote control capabilities, used to receive physically isolated sampled signals and perform offline parallel computation processing; and a nanosecond-level hardware time stamp generation unit associated with the three independent sampling units, used to generate data for the sampling signals. The sampling signal provides nanosecond-level high-precision time stamps, and achieves sampling data synchronization through time back-off alignment and dynamic delay compensation. The synchronization accuracy is ≤5ns, the timekeeping accuracy is ±20ns / h, and the timekeeping capability is ≥24 hours, without relying on GPS / BeiDou. The transformer fault discrimination and early warning unit, the maintenance adaptive degradation unit, and the whole-station protection timing coordination module are all connected to the offline GPU supercomputing unit to receive the operation output signal. The pure hardware 3-out-of-2 output matrix uses a power-down safe hardware AND-OR gate array to implement the 3-out-of-2 output logic. The logic satisfies Y=AB+AC+BC, the action time is ≤3ms, and the lockout is maintained in the power-down state without false operation. There is no software or CPU participation throughout the process. The circuit breaker trip execution unit is connected to the pure hardware 3-out-of-2 output matrix, receives the output logic signal, and executes the trip action.

2. The system according to claim 1, characterized in that, The three signal transmission paths of the triple independent sampling unit are completely independent, and any failure, short circuit or lightning strike of one path will not affect the normal operation of the other two paths.

3. The system according to claim 1, characterized in that, The offline GPU supercomputing unit includes at least three GPU computing cores and a parallel data bus interaction unit, which can realize parallel processing of electrical quantities and high-speed data interaction across multiple intervals throughout the station.

4. The system according to claim 1, characterized in that, The maintenance adaptive degradation unit supports ≤10ms seamless switching. When all three sets are normal, it adopts a three-out-of-two mode; when one set is abnormal, it switches to a two-out-of-one mode; when both sets are abnormal, it switches to a single-set operation + alarm mode. There are no protection blind spots throughout the process, and maintenance is uninterrupted.

5. The system according to claim 1, characterized in that, The system features a dual-zone isolation architecture with an offline configuration zone and an online operation zone. It supports offline configuration of new intervals, offline simulation verification, and one-click online loading. The single interval access time is ≤4 hours, and the device does not experience power outages or affect real-time protection operation.

6. The system according to claim 1, characterized in that, The current transformer fault detection and early warning unit can identify current transformer disconnection, saturation, and abnormal operating conditions in real time and issue early warning signals in advance.

7. The system according to claim 1, characterized in that, The whole-station protection timing coordination module realizes cross-interval fault linkage, multi-level protection coordination and whole-station coordinated control.

8. The system according to claim 1, characterized in that, The pure hardware 3-out-of-2 output matrix is ​​implemented independently using hard-contact circuits and is not affected by CPU crashes, communication interruptions, or program abnormalities.

9. The system according to claim 1, characterized in that, The system forms a triple-redundant architecture across the entire link of sampling, transmission, computation, and output, meeting the requirements for safe and stable operation of highly reliable intelligent substations.

Images