Distributed clean energy data multi-process parallel collection method based on heterogeneous isolation and dynamic load balancing

Abstract

This invention discloses a distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing, including the following steps: S1: heterogeneous data source classification and resource pool construction; S2: dynamic weighted load assessment; S3: process-level task elastic scheduling; S4: dual-channel heterogeneous data encapsulation; S5: fault-driven resource rebalancing. This invention addresses the lack of heterogeneous isolation by constructing a protocol-feature-driven hard isolation mechanism for the resource pool; addresses the dynamic load response hysteresis by using a weighted heatmap-driven elastic scheduling algorithm; addresses the mixed channel design by using physically separated dual-path transmission links; and addresses the coarse-grained fault recovery by developing a context-inherited self-healing engine, thereby systematically overcoming the four major technical obstacles in clean energy data acquisition.

Description

Technical Field

[0001] This invention belongs to the field of multi-process parallel data acquisition, and more specifically, it relates to a method for multi-process parallel acquisition of distributed clean energy data based on heterogeneous isolation and dynamic load balancing. Background Technology

[0002] With the large-scale deployment of clean energy facilities such as distributed photovoltaic and wind power, their operational data acquisition systems face challenges such as a surge in the number of terminals, strong protocol heterogeneity, and large differences in sampling frequencies. Current mainstream solutions mainly adopt a centralized data aggregation architecture or a distributed acquisition framework based on a polling mechanism. This involves aggregating data from various energy terminals through a unified access gateway and then using backend computing nodes for parsing and processing. While this architecture achieves physical connectivity at the device access layer;

[0003] Existing mainstream solutions lack physical-level isolation mechanisms for heterogeneous protocols. Multiple protocol parsing tasks block each other within shared computing units, causing high-priority real-time data acquisition to be hampered by inefficient protocols. Secondly, load balancing relies on manually preset fixed weights, failing to dynamically respond to network jitter and CPU resource fluctuations, resulting in resource waste such as "some nodes overloaded and crashing, while idle nodes have no task scheduling." Thirdly, data encapsulation channels do not differentiate between real-time and historical needs, and streaming processing and batch storage share the same path, causing excessive real-time data latency. Finally, fault recovery mechanisms employ a global restart strategy, interrupting the acquisition process and failing to inherit context, thus disrupting the continuity of energy data. These deficiencies are particularly prominent in wind-solar-storage collaborative monitoring scenarios, severely restricting the reliability and timeliness of clean energy holographic perception. Therefore, we propose a distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing. This method constructs a resource pool hard isolation mechanism driven by protocol features, employs a weighted heatmap-driven elastic scheduling algorithm to address the dynamic load response lag problem, and designs a dual-path physically separated transmission link to address the channel hybridity problem. This systematically overcomes the technical obstacles in clean energy data acquisition.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A multi-process parallel acquisition method for distributed clean energy data based on heterogeneous isolation and dynamic load balancing includes the following steps:

[0007] S1: Heterogeneous data source classification and resource pool construction. Based on the differences in communication protocols, sampling frequencies and data types of distributed energy terminals, heterogeneous data sources are divided into mutually exclusive acquisition clusters. A dedicated physical resource pool is dynamically allocated to each cluster to achieve hardware-level isolation.

[0008] S2: Dynamic weighted load assessment, real-time monitoring of CPU utilization, network latency and task queue depth of each resource pool, and integration of the above indicators to generate dynamic weights; constructing a multi-dimensional load heatmap based on the weight values ​​to identify overloaded / idle resource pools.

[0009] S3: Process-level elastic task scheduling, based on a multi-dimensional load heatmap, breaks down newly arriving data collection requests into atomic task units; and dynamically migrates atomic tasks to a low-load resource pool through a preemptive scheduler, ensuring that tasks within the same cluster are executed only by the bound process group.

[0010] S4: Dual-channel heterogeneous data encapsulation, performs protocol decoupling operation on successfully acquired energy data, stripping away the original communication protocol; injects the payload into two types of parallel channels, the real-time streaming channel uses memory mapping transmission, and the historical batch processing channel is written to an isolated solid-state storage pool.

[0011] S5: Fault-driven resource rebalancing. When a process exits abnormally or a device goes offline, the corresponding resource pool task queue is immediately frozen. The frozen tasks are redistributed based on the dynamic weights of the remaining online devices, and a daemon process is started in the new resource pool to inherit the collection context.

[0012] Preferably, the specific operation of dividing the acquisition cluster in S1 includes: dividing the RS-485 / Modbus terminals into a first cluster according to the physical layer transmission protocol, and dividing the IEC 61850 / MMS terminals into a second cluster according to the application layer protocol; configuring process group binding rules for resource pools independently for each cluster, and each resource pool exclusively using the CPU core group and network bandwidth channel to achieve hardware resource isolation.

[0013] Preferably, the specific logic for generating dynamic weights in S2 includes: setting the normalized threshold ratios for CPU utilization, network latency, and task queue depth to 0.4:0.3:0.3 respectively; when the real-time metric of any resource pool exceeds the threshold, an overload weight greater than 1 is generated proportionally, and a light load weight less than 1 is generated when it is below the threshold, with the weights updated every 200ms.

[0014] Preferably, the multi-dimensional load heatmap construction method of S2 is as follows: the physical address of the resource pool is used as the horizontal axis and the dynamic weight is used as the vertical axis. When the weight fluctuation exceeds ±15% for three consecutive times, the heatmap reconstruction is triggered. The overloaded resource pool is marked as a red block and the idle resource pool is marked as a green block, and the information is broadcast to all scheduling nodes through the message middleware.

[0015] Preferably, the atomic task unit decomposition rules of S3 include: dividing the acquisition request into the smallest data packet unit according to the data packet slot boundary, and attaching the source terminal ID, timestamp and check code to each unit; when the preemptive scheduler migrates the data packet unit across resource pools according to the heat map, it needs to verify that the process group of the target resource pool matches the source terminal cluster before allowing execution.

[0016] Preferably, the protocol decoupling operation of S4 specifically includes: stripping the function code and CRC check bit from the Modbus data frame through a preset protocol filter, or stripping the ASN.1 encoding header from the IEC 61850 message; the payload retains the original data value and timestamp, and is encapsulated into a unified JSON format for dual-channel transmission.

[0017] Preferably, the detection logic for abnormal process exit in S5 includes: the resource pool daemon process monitors the heartbeat packets of the working process, and when three consecutive heartbeats are lost and the process status word returns a SIGKILL signal, it is determined to be an abnormal exit; the device offline detection is based on the physical layer link state machine, and an alarm is triggered when the PHY layer carrier sensing fails for a continuous timeout of 2 seconds.

[0018] Preferably, the mathematical expression for generating the dynamic weights is:

[0019]

[0020] in, The dynamic weights at the current moment. >1 indicates overload. <1 indicates a light load; α, β, γ are normalization coefficients set for process-level task elastic scheduling; Real-time CPU utilization This is the preset CPU critical threshold. Due to network latency, As a reference delay, Maximum tolerable delay; For task queue depth, This is the queue capacity threshold; As a burst traffic correction factor, when When the instantaneous growth exceeds 50%, take 1.5.

[0021] Preferably, the real-time streaming channel operates as follows: a circular buffer is created in the local memory of the resource pool, and the JSON-formatted payload is transmitted to the streaming engine in the form of a memory-mapped file; at the same time, a zero-copy mechanism is enabled, and the buffer data is read directly by the message queue, with the transmission delay controlled within 5ms;

[0022] The transmission delay constraint for the real-time streaming channel is:

[0023] ;

[0024] ;

[0025] in, For memory mapping setup time; The message queue wait time; For data copy time; Size of the JSON payload; For memory bandwidth; Enable the zero-copy flag; when enabled... =1, otherwise =0; This is the acceleration coefficient; The maximum delay is 5ms.

[0026] The technical effects and advantages of this invention are as follows: Compared with the prior art, the distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing provided by this invention strictly divides the distributed energy terminals into mutually exclusive acquisition clusters according to the physical layer / application layer protocol standards, and independently configures each cluster with a dedicated CPU core group, network bandwidth channel and process binding rules; this design isolates the parsing process of different protocols at the physical layer, so that high real-time protocol terminals are not subject to resource preemption by inefficient protocol parsing, thus eliminating the message processing blocking problem from the source;

[0027] By integrating heterogeneous metrics such as CPU utilization, network latency, and task queue depth to generate dynamic weights in real time, and innovatively establishing a weight fluctuation threshold to trigger a heatmap reconstruction mechanism, this solution enables the preemptive scheduler to accurately identify the spatiotemporal distribution characteristics of overloaded / idle resource pools, realize cross-pool migration of atomic task units under minute-level load mutation conditions, and transform the "tidal imbalance" of traditional manual weight preset into dynamic elastic scheduling.

[0028] The protocol filter strips the communication protocol metadata and encapsulates the payload into a pure data format before splitting the transmission: the real-time streaming channel ensures the timeliness of streaming processing through memory mapping and zero-copy mechanism, while the historical batch processing channel uses a dedicated solid-state storage pool to achieve high-speed sequential writing; the physical channel separation design prevents the two types of data from interfering with each other, solving the buffer congestion problem of the mixed transmission mode.

[0029] When an abnormal process exit or device offline is detected, the atomic task unit is immediately frozen and the protocol context is fully preserved; the daemon process started in the new resource pool inherits the frozen task and historical state synchronously, and the collection progress is seamlessly connected through the command retransmission mechanism; it replaces the original recovery mode of global restart, ensuring that the data flow continuity between terminal and data center is maintained during resource rebalancing.

[0030] Based on the weight mean change trend of the multidimensional load heatmap, the creation of virtualization resource pool instances and the synchronization of protocol templates are automatically triggered; the overloaded terminal cluster is accessed in real time through software-defined networking, forming a closed loop of elastic expansion capability; the system resource scale is no longer a static preset value, but a dynamic variable that adapts to load fluctuations, solving the risk of cluster-level congestion under sudden traffic surges. Attached Figure Description

[0031] Figure 1 This is a flowchart of the distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0033] This invention provides, for example Figure 1 The distributed clean energy data multi-process parallel acquisition method shown is based on heterogeneous isolation and dynamic load balancing. It constructs a resource pool hard isolation mechanism driven by protocol features to address the lack of heterogeneous isolation; it uses an elastic scheduling algorithm driven by a weighted heatmap to address the hysteresis of dynamic load response; it designs a dual-path physically separated transmission link to address channel heterogeneity; and it develops a context inheritance self-healing engine for fault recovery. Thus, it systematically overcomes the four major technical obstacles of clean energy data acquisition.

[0034] The specific plan is as follows:

[0035] A multi-process parallel acquisition method for distributed clean energy data based on heterogeneous isolation and dynamic load balancing includes the following steps:

[0036] S1: Heterogeneous Data Source Classification and Resource Pool Construction. Based on the differences in communication protocols, sampling frequencies, and data types of distributed energy terminals, heterogeneous data sources are divided into mutually exclusive acquisition clusters. A dedicated physical resource pool is dynamically allocated to each cluster to achieve hardware-level isolation. The specific operations for dividing acquisition clusters in S1 include: dividing RS-485 / Modbus terminals into the first cluster according to the physical layer transmission protocol, and dividing IEC 61850 / MMS terminals into the second cluster according to the application layer protocol; configuring process group binding rules for resource pools independently for each cluster, and ensuring that each resource pool exclusively uses CPU core groups and network bandwidth channels to achieve hardware resource isolation.

[0037] S2: Dynamic weighted load assessment, real-time monitoring of CPU utilization, network latency, and task queue depth of each resource pool, and generation of dynamic weights by integrating these indicators; a multi-dimensional load heatmap is constructed based on the weight values ​​to identify overloaded / idle resource pools; the specific logic for generating dynamic weights in S2 includes: setting the normalized threshold ratios for CPU utilization, network latency, and task queue depth to 0.4:0.3:0.3 respectively; when the real-time indicators of any resource pool exceed the threshold, an overload weight greater than 1 is generated proportionally, and a light load weight less than 1 is generated when the indicators are below the threshold. The weights are updated every 200ms.

[0038] The S2 multidimensional load heatmap is constructed as follows: the physical address of the resource pool is used as the horizontal axis and the dynamic weight is used as the vertical axis. When the weight fluctuates by more than ±15% for three consecutive times, the heatmap is reconstructed. Overloaded resource pools are marked as red blocks and idle resource pools are marked as green blocks. The heatmap is broadcast to all scheduling nodes through the message middleware.

[0039] It should be noted that the mathematical expression for generating dynamic weights is:

[0040]

[0041] in, The dynamic weights at the current moment. >1 indicates overload. <1 indicates a light load; α, β, γ are normalization coefficients set for process-level task elastic scheduling; Real-time CPU utilization This is the preset CPU critical threshold. Due to network latency, As a reference delay, Maximum tolerable delay; For task queue depth, This is the queue capacity threshold; As a burst traffic correction factor, when When the instantaneous growth exceeds 50%, take 1.5;

[0042] The triggering condition for heatmap reconstruction is:

[0043]

[0044]

[0045] Where T is the sampling window, and three consecutive samples are taken; It is a historical dynamic weight sequence; This represents the variance of the weight sequence within the TTT window; Represents the arithmetic mean of the weight sequence; The set volatility is 15%. The signal-to-noise ratio coefficient is used to filter high-frequency interference fluctuations.

[0046] S3: Process-level elastic task scheduling, based on a multi-dimensional load heatmap, breaks down newly arriving data acquisition requests into atomic task units; a preemptive scheduler dynamically migrates atomic tasks to low-load resource pools, ensuring that tasks within the same cluster are executed only by the bound process group; S3's atomic task unit decomposition rules include: dividing acquisition requests into the smallest data packet units according to the data packet slot boundaries, with each unit appended with the source terminal ID, timestamp, and checksum; when the preemptive scheduler migrates data packet units across resource pools according to the heatmap, it must verify that the process group of the target resource pool matches the source terminal cluster before allowing execution;

[0047] In addition, the process group binding mechanism specifically involves preloading a protocol-adaptive dynamic link library for each resource pool and injecting a parsing engine for the corresponding communication protocol during process group initialization. When an atomic task unit migrates to a new resource pool, the bound process group calls the parsing engine of that resource pool to perform data acquisition, thus avoiding cross-cluster protocol conflicts.

[0048] It should be noted that the validity verification of atomic task unit migration must meet the following requirements:

[0049]

[0050] in, The set of protocol features required by the source terminal cluster; The set of protocol features supported by the target resource pool process group; Indicates the number of source protocol features; The protocol matching threshold is set to 0.95 to ensure basic compatibility.

[0051] S4: Dual-channel heterogeneous data encapsulation performs protocol decoupling operations on successfully acquired energy data, stripping away the original communication protocol; the payload is injected into two types of parallel channels, with the real-time streaming channel using memory-mapped transmission and the historical batch processing channel written to an isolated solid-state storage pool; the S4 protocol decoupling operation specifically includes: stripping the function code and CRC check bit from the Modbus data frame through a preset protocol filter, or stripping the ASN.1 encoded header from the IEC 61850 message; the payload retains the original data value and timestamp, and is encapsulated into a unified JSON format for dual-channel transmission;

[0052] Furthermore, the real-time streaming channel operates as follows: a circular buffer is created in the local memory of the resource pool, and the JSON-formatted payload is transmitted to the streaming engine as a memory-mapped file; at the same time, a zero-copy mechanism is enabled, and the buffer data is read directly by the message queue, with the transmission latency controlled within 5ms;

[0053] The transmission delay constraint for the real-time streaming channel is:

[0054] ;

[0055] ;

[0056] in, For memory mapping setup time; The message queue wait time; For data copy time; Size of the JSON payload; For memory bandwidth; Enable the zero-copy flag; when enabled... =1, otherwise =0; This is the acceleration coefficient; Maximum latency is 5ms;

[0057] Furthermore, the storage strategy for the historical batch processing channel includes allocating an independent NVMe solid-state storage pool for each resource pool, storing JSON data in shards based on the terminal IDHash value, adopting an append-only write mode and a CRC32 check mechanism, triggering a compression and archiving operation every morning, and uploading the compressed data blocks to the distributed file system.

[0058] S5: Fault-driven resource rebalancing. When a process exits abnormally or a device goes offline, the corresponding resource pool task queue is immediately frozen. The frozen tasks are redistributed based on the dynamic weights of the remaining online devices, and a daemon process is started in the new resource pool to inherit the acquisition context. The abnormal process exit detection logic in S5 includes: the resource pool daemon process monitors the heartbeat packets of the working process. When three consecutive heartbeats are lost and the process status word returns a SIGKILL signal, it is determined to be an abnormal exit. Device offline detection is based on the physical layer link state machine. When the PHY layer carrier sensing fails and the timeout lasts for 2 seconds, an alarm is triggered.

[0059] It should be noted that the operation of freezing the task queue is as follows: a copy of the frozen area is created on the resource pool management node to save the atomic task unit pointer and protocol context at the time of abnormal exit; at the same time, the access channel for new tasks in the resource pool is blocked until the fault recovery process is completed.

[0060] The daemon process inheritance mechanism includes starting a backup process group in the target resource pool, loading atomic task units and protocol contexts from the frozen area copy; if the protocol context contains incomplete transactions, the last valid instruction is resent to the source terminal, and the original collection progress is inherited and execution continues after a response is received.

[0061] This embodiment further includes a dynamic resource pool expansion mechanism. When the average dynamic weight of all online resource pools is greater than 1.2 for 1 minute, a virtualized resource pool instance is automatically created. The new instance inherits the configuration template of the overloaded cluster and accesses the corresponding terminal group through a software-defined network. The load data of the newly added instance is synchronized to the heat map.

[0062] The target fault recovery time is calculated as follows:

[0063]

[0064]

[0065] in, Freeze the task time; For context loading time; This is the handshake time for retransmitting instructions; The size of the last valid instruction; This represents the current network link bandwidth. This represents the number of retries. This is the lower limit of the terminal sampling period.

[0066] In summary, this invention strictly divides distributed energy terminals into mutually exclusive acquisition clusters according to physical layer / application layer protocol standards, and independently configures each cluster with a dedicated CPU core group, network bandwidth channel, and process binding rules. This design physically isolates the parsing process of different protocols, protecting high real-time protocol terminals from resource preemption caused by inefficient protocol parsing, thus eliminating the message processing blocking problem at the source.

[0067] By integrating heterogeneous metrics such as CPU utilization, network latency, and task queue depth to generate dynamic weights in real time, and innovatively establishing a weight fluctuation threshold to trigger a heatmap reconstruction mechanism, this solution enables the preemptive scheduler to accurately identify the spatiotemporal distribution characteristics of overloaded / idle resource pools, realize cross-pool migration of atomic task units under minute-level load mutation conditions, and transform the "tidal imbalance" of traditional manual weight preset into dynamic elastic scheduling.

[0068] The protocol filter strips the communication protocol metadata and encapsulates the payload into a pure data format before splitting the transmission: the real-time streaming channel ensures the timeliness of streaming processing through memory mapping and zero-copy mechanism, while the historical batch processing channel uses a dedicated solid-state storage pool to achieve high-speed sequential writing; the physical channel separation design prevents the two types of data from interfering with each other, solving the buffer congestion problem of the mixed transmission mode.

[0069] When an abnormal process exit or device offline is detected, the atomic task unit is immediately frozen and the protocol context is fully preserved; the daemon process started in the new resource pool inherits the frozen task and historical state synchronously, and the collection progress is seamlessly connected through the command retransmission mechanism; it replaces the original recovery mode of global restart, ensuring that the data flow continuity between terminal and data center is maintained during resource rebalancing.

[0070] Based on the weight mean change trend of the multidimensional load heatmap, the creation of virtualization resource pool instances and the synchronization of protocol templates are automatically triggered; the overloaded terminal cluster is accessed in real time through software-defined networking, forming a closed loop of elastic expansion capability; the system resource scale is no longer a static preset value, but a dynamic variable that adapts to load fluctuations, solving the risk of cluster-level congestion under sudden traffic surges.

[0071] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing, characterized in that, Includes the following steps: S1: Heterogeneous data source classification and resource pool construction. Based on the differences in communication protocols, sampling frequencies and data types of distributed energy terminals, heterogeneous data sources are divided into mutually exclusive acquisition clusters. A dedicated physical resource pool is dynamically allocated to each cluster to achieve hardware-level isolation. S2: Dynamic weighted load assessment, real-time monitoring of CPU utilization, network latency and task queue depth of each resource pool, and integration of the above indicators to generate dynamic weights; constructing a multi-dimensional load heatmap based on the weights to identify overloaded / idle resource pools; S3: Process-level elastic task scheduling, based on a multi-dimensional load heatmap, breaks down newly arriving data collection requests into atomic task units; and dynamically migrates atomic tasks to a low-load resource pool through a preemptive scheduler, ensuring that tasks within the same cluster are executed only by the bound process group; S4: Dual-channel heterogeneous data encapsulation, performs protocol decoupling operation on successfully acquired energy data, stripping away the original communication protocol; injects the payload into two types of parallel channels, the real-time streaming channel uses memory-mapped transmission, and the historical batch processing channel is written to an isolated solid-state storage pool; S5: Fault-driven resource rebalancing. When a process exits abnormally or a device goes offline, the corresponding resource pool task queue is immediately frozen. The frozen tasks are redistributed based on the dynamic weights of the remaining online devices, and a daemon process is started in the new resource pool to inherit the collection context.

2. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 1, characterized in that, The specific operations for dividing the acquisition clusters in S1 include: dividing RS-485 / Modbus terminals into the first cluster according to the physical layer transmission protocol, and dividing IEC 61850 / MMS terminals into the second cluster according to the application layer protocol; configuring process group binding rules for resource pools independently for each cluster, and ensuring that each resource pool exclusively occupies the CPU core group and network bandwidth channel to achieve hardware resource isolation.

3. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 1, characterized in that, The specific logic for generating dynamic weights in S2 includes: setting the normalized threshold ratios for CPU utilization, network latency, and task queue depth to 0.4:0.3:0.3 respectively; when the real-time metrics of any resource pool exceed the threshold, an overload weight greater than 1 is generated proportionally, and a light load weight less than 1 is generated when the metrics are below the threshold. The weights are updated every 200ms.

4. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 3, characterized in that, The multidimensional load heatmap construction method of S2 is as follows: the physical address of the resource pool is used as the horizontal axis and the dynamic weight is used as the vertical axis. When the weight fluctuation exceeds ±15% for three consecutive times, the heatmap reconstruction is triggered. Overloaded resource pools are marked as red blocks, and idle resource pools are marked as green blocks, and this is broadcast to all scheduling nodes via message middleware.

5. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 1, characterized in that, The atomic task unit decomposition rules of S3 include: dividing the acquisition request into the smallest data packet unit according to the data packet slot boundary, and attaching the source terminal ID, timestamp and check code to each unit; when the preemptive scheduler migrates the data packet unit across resource pools according to the heat map, it must verify that the process group of the target resource pool matches the source terminal cluster before allowing execution.

6. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 1, characterized in that, The protocol decoupling operation of S4 specifically includes: stripping the function code and CRC check bit from the Modbus data frame through a preset protocol filter, or stripping the ASN.1 encoded header from the IEC 61850 message; the payload retains the original data value and timestamp, and is encapsulated into a unified JSON format for dual-channel transmission.

7. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 1, characterized in that, The detection logic for abnormal process exit in S5 includes: the resource pool daemon process monitors the heartbeat packets of the working process, and when three consecutive heartbeats are lost and the process status word returns a SIGKILL signal, it is determined to be an abnormal exit; the device offline detection is based on the physical layer link state machine, and an alarm is triggered when the PHY layer carrier sensing fails for a continuous timeout of 2 seconds.

8. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 3, characterized in that, The mathematical expression for generating the dynamic weights is: in, The dynamic weights at the current moment. >1 indicates overload. <1 indicates a light load; α, β, γ are normalization coefficients set for process-level task elastic scheduling; Real-time CPU utilization This is the preset CPU critical threshold. Due to network latency, As a reference delay, Maximum tolerable delay; For task queue depth, This is the queue capacity threshold; As a burst traffic correction factor, when When the instantaneous growth exceeds 50%, take 1.

5.

9. The distributed clean energy data multi-process parallel acquisition method based on heterogeneous isolation and dynamic load balancing according to claim 6, characterized in that, The operation mode of the real-time streaming channel is as follows: a circular buffer is opened in the local memory of the resource pool, and the JSON format payload is transmitted to the streaming engine in the form of a memory-mapped file; at the same time, a zero-copy mechanism is enabled, and the buffer data is directly read by the message queue, and the transmission latency is controlled within 5ms. The transmission delay constraint for the real-time streaming channel is: ； ； in, For memory mapping setup time; The message queue wait time; For data copy time; Size of the JSON payload; For memory bandwidth; Enable the zero-copy flag; when enabled... =1, otherwise =0; This is the acceleration coefficient; The maximum delay is 5ms.

Images