A method for integrating and allocating computing resources based on large-scale models of power data

Abstract

This invention provides a method for integrating and allocating computing resources for large-scale models based on power data, comprising the following steps: acquiring basic data of the large-scale model, determining the total amount of computing resources of the large-scale model from the basic data, and constructing a core computing pool and an unnecessary computing pool based on the total amount of computing resources; querying the specific locations of converter stations monitored by the large-scale model, mapping converter station nodes around the core computing pool, and connecting the core computing pool and converter station nodes through data channels; when a converter station experiences an anomaly or failure, collecting power data of the converter station, and evaluating the viscosity value between the converter station and the core computing pool based on the power data; adjusting the data channels according to the viscosity value, and allocating the computing resources of the core computing pool and the unnecessary computing pool to the abnormal or failed converter station through the adjusted data channels. By dividing the core computing power into core and unnecessary computing power, the problem of chaotic allocation caused by mixed computing power can be avoided, and the quantified viscosity value can be used to rationally allocate computing resources.

Description

Technical Field

[0001] This invention relates to the field of computing power optimization technology, and in particular to a method for integrating and allocating large-scale computing power resources based on power data. Background Technology

[0002] With the rapid development of new power systems, converter stations, as the core hubs for AC / DC power conversion, are directly related to the overall reliability of the power grid through their safe and stable operation. In recent years, large-scale model technology has been widely used in converter station status monitoring, fault diagnosis and other scenarios. By analyzing massive amounts of power data, accurate early warning can be achieved, which greatly improves operation and maintenance efficiency. However, with the continuous increase in the number of converter stations and the increasing complexity of equipment types, power data is experiencing explosive growth, which places extremely high demands on the computing power support of large-scale models.

[0003] Currently, there are still significant bottlenecks in the integration and allocation of large-scale computing resources: on the one hand, computing resources are mostly deployed in different physical nodes and lack a unified pooling management mechanism, which makes it difficult to quickly call up idle computing resources when a faulty converter station urgently needs computing power. On the other hand, computing power allocation strategies are out of sync with the power operation characteristics of converter stations, and existing technologies have failed to establish quantitative correlation indicators that reflect the urgency of the computing power demand of converter stations, often resulting in situations of excess computing power wasted or insufficient computing power delaying processing. Summary of the Invention

[0004] In view of this, the present invention proposes a method for integrating and allocating computing resources for large-scale models based on power data. This method can allocate computing power by quantifying viscosity values, thereby avoiding situations of excessive or insufficient computing power.

[0005] The technical solution of this invention is implemented as follows: A method for integrating and allocating computing resources for large-scale power models based on power data includes the following steps: Step S1: Obtain the basic data of the large model, determine the total computing power resources of the large model from the basic data, and construct the core computing power pool and the non-essential computing power pool based on the total computing power resources. Step S2: Query the specific location of the converter station monitored by the large model, map the converter station nodes around the core computing power pool, and connect the core computing power pool and the converter station nodes through the data flow channel. Step S3: When an abnormality or fault occurs at the converter station, collect the power data of the converter station and evaluate the stickiness value between the converter station and the core computing pool based on the power data. Step S4: Adjust the data flow channel according to the viscosity value, and allocate the computing resources of the core computing pool and the non-essential computing pool to the abnormal or faulty converter station through the adjusted data flow channel.

[0006] Preferably, step S1 includes the following steps: Step S11: Obtain the computing power hardware parameters of the large model, and use the computing power hardware parameters to calculate the theoretical computing power limit of the large model; Step S12: Obtain the real-time load data of the large model, determine the occupied computing power based on the real-time load data, and deduct the occupied computing power from the theoretical computing power limit to obtain the total amount of allocable computing power resources. Step S13: Obtain the minimum computing power requirement for converter station fault diagnosis and convert the minimum computing power requirement into a division ratio; Step S14: Divide the total computing power resources according to the division ratio, and construct the core computing power pool and the unnecessary computing power pool.

[0007] Preferably, the specific steps of step S13 are as follows: obtain the historical fault handling log of the converter station, extract typical faults and the computing power consumption data corresponding to the typical faults from the historical fault handling log, calculate the average computing power requirement of a single typical fault as the minimum computing power for a single fault, simulate the scenario of multiple typical faults occurring simultaneously, and calculate the average computing power requirement under the superposition of multiple typical faults as the minimum computing power for multiple faults, take the larger value between the minimum computing power for a single fault and the minimum computing power for multiple faults as the minimum computing power requirement, and take the ratio of the minimum computing power requirement to the total amount of allocable computing power resources as the division ratio.

[0008] Preferably, the computing power hardware parameters include the model, quantity, and computing power of the GPU / CPU / FPGA.

[0009] Preferably, step S2 includes the following specific steps: Step S21: Query the physical coordinates of the converter stations monitored by the large model through the power dispatch topology database; Step S22: Map the core computing power pool and unnecessary computing power pool on the visualization interface, and map the converter station nodes around the core computing power pool based on the physical coordinates of the converter station. Step S23: Query the communication link information between the converter station and the large model, convert the communication link into a data channel, and connect the converter station node and the core computing power pool.

[0010] Preferably, the specific steps of step S23 are as follows: query the communication link type between the converter station and the large model from the power system database, select shielded fiber optic links for delay and packet loss rate screening, convert the physical logic of the selected shielded fiber optic links into data channels, and establish the mapping relationship between the data channels and the converter station nodes and the core computing pool.

[0011] Preferably, step S3 includes the following specific steps: Step S31: When the large model detects an alarm signal from the converter station, it triggers a power data acquisition command. Step S32: Extract fault urgency, data generation rate, and equipment importance parameters from the collected power data; Step S33: After assigning weights to the fault urgency, data generation rate and equipment importance parameters respectively, add them together to obtain the viscosity value between the converter station and the core computing pool.

[0012] Preferably, the specific steps of step S32 are as follows: Fault types are extracted from power data, and the fault types are compared with preset status evaluation standards to obtain a quantitative fault urgency. Record the number of real-time data frames within a period of time after the fault occurs, and calculate the data generation rate based on the number of real-time data frames and the acquisition time. Identify the equipment corresponding to the fault from power data, quantify the importance of the equipment based on its significance, and obtain equipment importance parameters.

[0013] Preferably, step S4 includes the following specific steps: Step S41: Compare the viscosity value with the preset threshold to determine the computing power demand status, and increase the data flow channel and / or increase the bandwidth according to the computing power demand status. Step S42: Identify pre-training tasks that are idle in the non-essential computing power pool, pause such tasks, and reclaim the computing power resources they occupy as redundant computing power. Step S43: Prioritize the transmission of computing power from the core computing power pool to the corresponding converter station through the adjusted data flow channel, and then supplement the transmission with redundant computing power recovered from the unnecessary computing power pool.

[0014] Preferred options also include: Step S5: Dynamically visualize the core computing power pool, non-essential computing power pool, converter station nodes, and data channels.

[0015] Compared with the prior art, the beneficial effects of the present invention are: The present invention provides a method for integrating and allocating computing resources for large-scale power models based on power data. This method divides the total computing resources of large-scale models into a core computing pool and an unnecessary computing pool according to the actual situation. The core computing pool focuses on key needs such as converter station fault diagnosis and reserves computing power in advance to match the minimum needs of fault handling. This avoids unnecessary tasks crowding out core resources and ensures that basic computing power is not lost when a fault occurs. The unnecessary computing pool carries non-urgent tasks such as model pre-training and historical data backup, making full use of idle computing power. This method not only improves the overall utilization rate of computing resources, but also avoids non-urgent tasks interfering with emergency fault handling and solves the problem of chaotic resource allocation caused by mixed computing power. After visualizing the core computing power pool and the non-essential computing power pool, converter station nodes are deployed around the core computing power pool based on the location of the converter stations. The core computing power pool and the converter station nodes are connected through data channels. In the event of a fault, a power data acquisition command can be triggered, and the viscosity value between the converter station and the core computing power pool can be evaluated based on the acquired power data. Based on the quantified viscosity value, the computing power resources of the core computing power pool and the non-essential computing power pool can be integrated and allocated, improving the real-time performance of fault handling and the economy of computing power resources. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the method for integrating and allocating computing resources based on large-scale power models using power data, as described in this invention. Figure 2 This is a flowchart of step S1 of the large-scale computing resource integration and allocation method based on power data of the present invention; Figure 3 This is a flowchart of step S2 of the large-scale computing resource integration and allocation method based on power data of the present invention; Figure 4 This is a flowchart of step S3 of the large-scale computing resource integration and allocation method based on power data of the present invention; Figure 5 This is a flowchart of step S4 of the large-scale computing resource integration and allocation method based on power data of the present invention. Detailed Implementation

[0018] To better understand the technical content of this invention, a specific embodiment is provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0019] See Figures 1 to 5 The present invention provides a method for integrating and allocating large-scale computing resources based on power data, comprising the following steps: Step S1: Obtain the basic data of the large model, determine the total computing power resources of the large model from the basic data, and construct the core computing power pool and the non-essential computing power pool based on the total computing power resources. Step S2: Query the specific location of the converter station monitored by the large model, map the converter station nodes around the core computing power pool, and connect the core computing power pool and the converter station nodes through the data flow channel. Step S3: When an abnormality or fault occurs at the converter station, collect the power data of the converter station and evaluate the stickiness value between the converter station and the core computing pool based on the power data. Step S4: Adjust the data flow channel according to the viscosity value, and allocate the computing resources of the core computing pool and the non-essential computing pool to the abnormal or faulty converter station through the adjusted data flow channel.

[0020] In the management of converter stations in power systems, large-scale models are used for fault monitoring and timely alarms to facilitate timely fault location and repair / replacement, ensuring the reliable operation of the converter stations. To ensure full utilization of the large-scale model's computing power, a single model typically monitors multiple converter stations. If a converter station experiences a serious fault, the large-scale model needs to allocate computing resources to the faulty converter station to ensure fault detection speed and accuracy. However, current computing resource integration is challenging, with some idle computing power. Most systems simply allocate core computing power directly to the faulty converter station without quantifying the required computing resources or selecting the appropriate type of computing power for allocation, which can easily lead to problems. Regarding resource allocation issues, this invention first obtains the basic data of a large model, then determines the total computing power resources of the large model from the basic data, and then divides the total computing power resources into a core computing power pool and an unnecessary computing power pool according to the corresponding division ratio. The core computing power pool focuses on key needs such as converter station fault diagnosis, and reserves computing power in advance to match the minimum needs of fault handling, avoiding unnecessary tasks from crowding out core resources and ensuring that basic computing power is not lost when a fault occurs. The unnecessary computing power pool carries non-urgent tasks such as model pre-training and historical data backup, making full use of idle computing power. This not only improves the overall computing power resource utilization rate, but also avoids non-urgent tasks from interfering with emergency fault handling, solving the problem of chaotic resource allocation caused by the co-location of computing power in traditional systems.

[0021] After dividing the computing power pool into a core computing power pool and a non-essential computing power pool, the location of all converter stations monitored by the large model is queried, and converter station nodes are mapped around the core computing power pool. Then, a data channel connects the converter station nodes and the core computing power pool, allowing the data channel to deliver computing resources from the core computing power pool to the converter station nodes. Visualization provides an intuitive view of the computing resources currently used by each converter station node. When a converter station experiences an anomaly or failure, more computing resources need to be allocated for fault monitoring. At this time, power data from the converter station is collected, and the stickiness value between the converter station and the core computing power pool is assessed based on the real-time power data. A higher viscosity value indicates that the converter station requires more computing resources. The data flow is then adjusted based on the viscosity value. Adjustment measures include increasing the number of data flow channels or increasing bandwidth, so that more computing resources flow to the faulty converter station, which facilitates rapid fault prediction and location. By setting the viscosity value, the computing resources required by the converter station can be quantified. When allocating computing resources, they are allocated from the core computing pool and the non-essential computing pool respectively, avoiding the concentration of computing resources from either the core computing pool or the non-essential computing pool. This ensures that the fault monitoring of the core computing power and the backup and training functions of the non-essential computing power can be carried out normally.

[0022] Preferably, step S1 includes the following steps: Step S11: Obtain the computing power hardware parameters of the large model. The computing power hardware parameters include the model, quantity, and computing power of the GPU / CPU / FPGA. Calculate the theoretical computing power limit of the large model based on the computing power hardware parameters. Step S12: Obtain the real-time load data of the large model, determine the occupied computing power based on the real-time load data, and deduct the occupied computing power from the theoretical computing power limit to obtain the total amount of allocable computing power resources. Step S13: Obtain the minimum computing power requirement for converter station fault diagnosis and convert the minimum computing power requirement into a division ratio; Step S14: Divide the total computing power resources according to the division ratio, and construct the core computing power pool and the unnecessary computing power pool.

[0023] Before constructing the core computing power pool and the non-essential computing power pool, it is first necessary to determine the total computing power resources of the large model. Based on the hardware parameters of the large model during deployment, the theoretical computing power limit of the large model can be roughly estimated. The computing power hardware parameters include the model and quantity of the corresponding controllers, such as GPU / CPU / FPGA, etc. Different models of controllers have different computing power, and the computing power of a single device also varies. The theoretical computing power limit of the large model can be obtained by statistically analyzing the aforementioned computing power hardware parameters. After the large model is deployed, there will be some load consumption, which will occupy computing power resources. This occupation is necessary. Therefore, after obtaining the real-time load data of the large model, the occupied computing power can be determined through the real-time load data. Then, the theoretical computing power limit is subtracted from the occupied computing power to obtain the total allocable computing power resources. When determining the allocation ratio, it is necessary to obtain the minimum computing power requirement for converter station fault diagnosis. Based on this, the allocation ratio is determined. Finally, the total computing power resources can be allocated according to the allocation ratio to construct the core computing power pool and the non-essential computing power pool.

[0024] Preferably, the specific steps of step S13 are as follows: obtain the historical fault handling log of the converter station, extract typical faults and the computing power consumption data corresponding to the typical faults from the historical fault handling log, calculate the average computing power requirement of a single typical fault as the minimum computing power for a single fault, simulate the scenario of multiple typical faults occurring simultaneously, and calculate the average computing power requirement under the superposition of multiple typical faults as the minimum computing power for multiple faults, take the larger value between the minimum computing power for a single fault and the minimum computing power for multiple faults as the minimum computing power requirement, and take the ratio of the minimum computing power requirement to the total amount of allocable computing power resources as the division ratio.

[0025] When determining the minimum computing power requirement for converter station fault diagnosis, a comprehensive assessment is necessary. This is because when a converter station experiences an anomaly, it may not be caused by a single fault. The computing power resources required for a single fault and multiple faults are also different. First, the historical fault handling logs of the converter station are obtained. Then, typical faults and the computing power consumption data corresponding to each typical fault are extracted from them. A single typical fault may have occurred multiple times in the historical fault handling logs. Therefore, by averaging the computing power consumption data corresponding to a single typical fault, the minimum computing power for a single fault can be obtained. Then, a scenario in which 2-3 typical faults occur simultaneously is simulated, and the average computing power requirement under the superposition of multiple typical faults is calculated as the minimum computing power requirement for multiple faults. Finally, the minimum computing power requirement for a single fault and the minimum computing power requirement for multiple faults are compared, and the maximum value of the two is taken as the minimum computing power requirement, ensuring that the allocated core computing power pool can support both single-fault and multi-fault scenarios.

[0026] Preferably, step S2 includes the following specific steps: Step S21: Query the physical coordinates of the converter stations monitored by the large model through the power dispatch topology database; Step S22: Map the core computing power pool and unnecessary computing power pool on the visualization interface, and map the converter station nodes around the core computing power pool based on the physical coordinates of the converter station. Step S23: Query the communication link information between the converter station and the large model, convert the communication link into a data channel, and connect the converter station node and the core computing power pool.

[0027] After constructing the core computing power pool and the non-essential computing power pool, the converter stations need to be mapped to the vicinity of the core computing power pool. First, the power dispatch topology database is accessed to query the physical coordinates of the converter stations monitored by the large model to determine their locations. Then, the core computing power pool and the non-essential computing power pool are visualized, and the non-essential computing power pool and the core computing power pool can communicate with each other. Then, based on the orientation and specific distance of the converter stations, they are mapped to the periphery of the core computing power pool to form converter station nodes. Based on the communication link information between the converter stations and the large model, the physical logic of the communication link is converted into a virtual data channel, and then the converter station nodes and the core computing power pool are connected through the data channel.

[0028] Preferably, the specific steps of step S23 are as follows: query the communication link type between the converter station and the large model from the power system database, select shielded fiber optic links for delay and packet loss rate screening, convert the physical logic of the selected shielded fiber optic links into data channels, and establish the mapping relationship between the data channels and the converter station nodes and the core computing pool.

[0029] When constructing the data channel, the power system database is read to query the communication link type between the converter station and the large model, including shielded fiber optic links and 5G communication links. Among them, the shielded fiber optic link has better shielding performance, so it is selected as the main communication link, while the 5G communication link is used as the backup communication link. The latency and packet loss rate of multiple shielded fiber optic links are screened. The physical logic of shielded fiber optic links with latency and packet loss rate below the preset threshold is converted into virtual data channels and then connected to the converter station node and the core computing power pool.

[0030] Preferably, step S3 includes the following specific steps: Step S31: When the large model detects an alarm signal from the converter station, it triggers a power data acquisition command. Step S32: Extract fault urgency, data generation rate, and equipment importance parameters from the collected power data; Step S33: After assigning weights to the fault urgency, data generation rate and equipment importance parameters respectively, add them together to obtain the viscosity value between the converter station and the core computing pool.

[0031] The large model can monitor the converter station in real time. If an alarm signal is detected from the converter station, it can trigger the data acquisition command of the converter station. The acquisition equipment in the converter station can collect power data. By analyzing the power data, parameters such as fault urgency, data generation rate and equipment importance can be extracted. Finally, a weighted summation method is used to sum the fault urgency, data generation rate and equipment importance parameters separately to obtain a quantitative value. This quantitative value is recorded as the viscosity value between the converter station and the core computing power pool. The higher the viscosity value, the more computing power resources are required for fault monitoring of the converter station.

[0032] Preferably, the specific steps of step S32 are as follows: Fault types are extracted from power data, and the fault types are compared with preset status evaluation standards to obtain a quantitative fault urgency. Record the number of real-time data frames within a period of time after the fault occurs, and calculate the data generation rate based on the number of real-time data frames and the acquisition time. Identify the equipment corresponding to the fault from power data, quantify the importance of the equipment based on its significance, and obtain equipment importance parameters.

[0033] The fault type is extracted from the alarm signals collected by the converter station's SCADA system, such as abnormal converter valve triggering or low GIS air chamber pressure. The fault type is mapped to a fault urgency level of 1-10 by comparing it with the preset status evaluation standard. The number of real-time data frames within 5 minutes after the fault occurs is recorded by the PMU device. The data generation rate is calculated as the number of real-time data frames / acquisition time. The equipment corresponding to the fault is determined from the equipment ledger of the converter station. Different equipment has different importance. Main equipment, such as converter valves, converter transformers, and valve cooling systems, has higher importance, while auxiliary equipment, including security monitoring and ventilation equipment, has lower importance. The importance is then converted into a quantitative equipment importance parameter.

[0034] Preferably, step S4 includes the following specific steps: Step S41: Compare the viscosity value with the preset threshold to determine the computing power demand status, and increase the data flow channel and / or increase the bandwidth according to the computing power demand status. Step S42: Identify pre-training tasks that are idle in the non-essential computing power pool, pause such tasks, and reclaim the computing power resources they occupy as redundant computing power. Step S43: Prioritize the transmission of computing power from the core computing power pool to the corresponding converter station through the adjusted data flow channel, and then supplement the transmission with redundant computing power recovered from the unnecessary computing power pool.

[0035] When the viscosity value is high, more computing power resources are required. By comparing the viscosity value with a preset threshold, the computing power demand status can be determined. When the computing power demand is high, the data flow channel can be increased or the bandwidth can be improved. The increased computing power resources can be obtained from the core computing power pool or from the non-essential computing power pool. First, it is necessary to identify the pre-training tasks that are idle in the non-essential computing power pool. After pausing the pre-training task, the corresponding occupied computing power resources are recovered as redundant computing power. When integrating and allocating computing power resources, the computing power in the core computing power pool is given priority to be sent to the converter station, and then the redundant computing power recovered from the non-essential computing power pool is allocated.

[0036] Preferred options also include: Step S5: Dynamically visualize the core computing power pool, non-essential computing power pool, converter station nodes, and data channels.

[0037] The adjustment of data flow channels and the flow of computing power will be dynamically visualized, so that personnel can view the flow of computing resources and the computing resource occupancy status of each converter station.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 method for integrating and deploying large model computing resources based on power data, characterized in that, The method comprises the following steps: Step S1, obtaining basic data of the large model, determining the total amount of computing resource of the large model from the basic data, and constructing a core computing resource pool and a non-essential computing resource pool according to the total amount of computing resource; Step S2, querying the specific location of the converter station monitored by the large model, and mapping the converter station node around the core computing resource pool, connecting the core computing resource pool and the converter station node through a data flow channel; Step S3, when the converter station has an abnormality or a fault, collecting power data of the converter station, and evaluating the viscosity value of the converter station and the core computing resource pool based on the power data; Step S4, adjusting the data flow channel according to the viscosity value, and allocating the computing resource of the core computing resource pool and the non-essential computing resource pool to the abnormal or faulty converter station through the adjusted data flow channel.

2. The method of claim 1, wherein, The specific steps of step S1 include: Step S11, obtaining computing hardware parameters of the large model, and counting the theoretical upper limit of computing power of the large model through the computing hardware parameters; Step S12, obtaining real-time load data of the large model, determining occupied computing power based on the real-time load data, and obtaining the total amount of allocable computing resource by deducting the occupied computing power from the theoretical upper limit of computing power; Step S13, obtaining the minimum computing power requirement of the converter station fault diagnosis, and converting the minimum computing power requirement into a division ratio; Step S14, dividing the total amount of computing resource based on the division ratio, and constructing the core computing resource pool and the non-essential computing resource pool.

3. The method of claim 2, wherein, The specific steps of step S13 are: obtaining the historical fault handling log of the converter station, extracting typical faults and computing power consumption data corresponding to the typical faults from the historical fault handling log, calculating the average computing power requirement of a single typical fault as the single-fault minimum computing power, simultaneously simulating the scenario of multiple typical faults occurring at the same time, and calculating the average computing power requirement under the superposition of multiple typical faults as the multi-fault minimum computing power, taking the larger value between the single-fault minimum computing power and the multi-fault minimum computing power as the minimum computing power requirement, and taking the ratio of the minimum computing power requirement to the total amount of allocable computing resource as the division ratio.

4. The method of claim 3, wherein, The computing hardware parameters include the model, quantity and single-device computing power of GPU / CPU / FPGA.

5. The method of claim 1, wherein, The specific steps of step S2 include: Step S21, querying the physical coordinates of the converter station monitored by the large model through the power dispatching topology database; Step S22, mapping the core computing resource pool and the non-essential computing resource pool on the visual interface, and mapping the converter station node around the core computing resource pool based on the physical coordinates of the converter station; Step S23, querying the communication link information between the converter station and the large model, converting the communication link into a data channel, and connecting the converter station node and the core computing resource pool.

6. The method of claim 5, wherein, The specific steps of step S23 are: querying the communication link type between the converter station and the large model from the power system database, selecting shielded optical fiber links for delay and packet loss rate screening, converting the physical logic of the screened shielded optical fiber links into data channels, and establishing a mapping relationship between the data channels and the converter station node and the core computing resource pool.

7. The method of claim 1, wherein, The specific steps of step S3 include: Step S31, when the large model detects that the converter station sends an alarm signal, triggering a power data collection instruction; Step S32, extracting the fault emergency degree, data generation rate and equipment importance parameter from the collected power data; Step S33, respectively weighting the fault emergency degree, data generation rate and equipment importance parameter and then adding them to obtain the viscosity value of the converter station and the core computing power pool.

8. The large model computing resource integration and deployment method based on power data according to claim 7, characterized in that, The specific steps of step S32 are: Extracting the fault type from the power data, comparing the fault type with the preset state evaluation standard, and obtaining the quantitative fault emergency degree; Recording the number of real-time data frames within a period of time after the fault occurs, and calculating the data generation rate according to the number of real-time data frames and the collection time length; Determining the corresponding equipment from the power data according to the importance of the equipment, and obtaining the equipment importance parameter by quantitative scoring.

9. The method of claim 1, wherein, The specific steps of step S4 include: Step S41, comparing the viscosity value with the preset threshold value to determine the computing power demand state, and increasing and / or improving the bandwidth of the data flow channel according to the computing power demand state; Step S42, identifying the pre-training tasks in the idle state in the unnecessary computing power pool, suspending such tasks and recycling the occupied computing power resources as redundant computing power; Step S43, preferentially delivering the computing power of the core computing power pool to the corresponding converter station through the adjusted data flow channel, and then supplementarily delivering the redundant computing power recycled from the unnecessary computing power pool.

10. The method of claim 1, wherein, Further comprising: Step S5, dynamically visualizing and displaying the core computing power pool, the unnecessary computing power pool, the converter station node and the data flow channel.

Images