Wind power plant cloud platform management method based on multi-tenant architecture

Abstract

The invention relates to the technical field of new energy wind power operation and maintenance, in particular to a wind power plant cloud platform management method based on a multi-tenant architecture, and the method comprises the steps: collecting the multi-dimensional operation data of a fan group through an edge computing node, completing the data cleaning, protocol conversion, signature encryption and structured packaging, and generating a standardized data package. And constructing a three-dimensional decision model based on the fan power predicted value, the network delay coefficient and the tenant priority weight, and dividing resource partitions by using an improved K-means algorithm to generate a resource allocation topological graph. And constructing a fault processing sandbox in the trusted execution environment, performing fault propagation analysis by using the graph neural network model, and outputting equipment control parameter correction and a scheduling path. And in combination with the dynamic reconstruction logic mapping relation and the bandwidth weight, generating a resource scheduling instruction, and completing a closed-loop control link. According to the invention, intelligent resource scheduling and fault adaptive processing of the multi-tenant wind power cluster can be realized, and the real-time performance, the credibility and the fine management level of system operation are improved.

Description

Technical Field

[0001] The present invention relates to the field of new energy wind power operation and maintenance technology, and in particular to a wind farm cloud platform management method based on a multi-tenant architecture. Background Art

[0002] With the continuous advancement of the new energy development strategy, the clustered deployment of large-scale wind farms has become an important part of the current power system. To improve the intelligent management level of wind farms, cloud platform architectures have been widely introduced to uniformly dispatch wind turbine operating status, resource allocation, and fault handling. In actual applications, to meet the needs of multi-site and multi-tenant parallel operation, wind farm management systems must have capabilities such as data collection standardization, multi-dimensional optimization of resource scheduling, fault-aware closed-loop control, and trusted verification of instruction execution. At the same time, the linkage between edge computing nodes and cloud resources has become the key foundation for supporting scheduling strategies that coexist in real time and precision. How to achieve cross-platform, highly reliable, and highly dynamic collaborative management of wind power resources while ensuring safety and controllability has become a core focus of the industry.

[0003] However, existing technologies still face the following key technical difficulties in the management of multi-tenant wind power cloud platforms: First, traditional data collection lacks a trusted fingerprint mechanism and time synchronization authentication, which easily leads to the risk of data forgery or playback, affecting the accuracy of system decision-making; Second, the resource allocation process often adopts static mapping and single-factor scheduling strategies, which cannot fully consider the dynamic changes of power prediction, network latency and tenant priority, resulting in low allocation accuracy and high resource redundancy; Third, fault identification mostly relies on static rules and lacks graph structure modeling of the coupling relationship between wind turbines, which limits the collaborative diagnosis capability of multiple devices; Fourth, the command issuance process lacks the support of a trusted execution environment, making it difficult to achieve full-link closed-loop feedback and failure rollback, affecting the security and robustness of resource reconstruction. Summary of the Invention

[0004] The present invention provides a wind farm cloud platform management method based on a multi-tenant architecture, which has multi-dimensional modeling capabilities, graph reasoning fault diagnosis, dynamic resource reconstruction and closed-loop control, so as to realize efficient, reliable and intelligent operation of wind power clusters.

[0005] A wind farm cloud platform management method based on a multi-tenant architecture includes the following steps:

[0006] S1: Collect multi-dimensional operation data of the wind turbine group through the edge computing node, and generate a standardized data packet through the protocol adaptation layer. The standardized data packet includes the device fingerprint hash value, the timestamp encryption block and the data check code;

[0007] S2: Input the standardized data packet into the multi-tenant resource allocation engine, and generate a resource allocation topology map based on a dynamic load balancing algorithm. The resource allocation topology map includes the logical mapping relationship of the virtual computing units and the bandwidth allocation weights;

[0008] S3: Build a fault event processing sandbox based on the resource allocation topology, perform correlation fault diagnosis in an isolated environment, and output a fault handling instruction set, which includes equipment control parameter corrections and spare parts scheduling paths;

[0009] S4: Based on the fault handling instruction set, the resource reallocation module is triggered to generate dynamic resource scheduling instructions and feed them back to the edge computing node to form a closed-loop control link.

[0010] Optionally, the S1 includes:

[0011] S11: Collecting multi-dimensional operation data of the wind turbine group through the edge computing node, wherein the multi-dimensional operation data of the wind turbine group includes a vibration spectrum, a power curve, and a yaw angle;

[0012] S12: Inputting the multi-dimensional operation data of the wind turbine group into a data cleaning submodule, removing abnormal jump data based on a sliding window outlier detection algorithm, and generating purified operation data;

[0013] S13: Inputting the purification operation data into the protocol conversion submodule, dynamically selecting Modbus, IEC104 or OPC-UA conversion protocol according to the device communication protocol type, and generating a unified format data frame;

[0014] S14: Input the unified format data frame into the data signature submodule, use the national secret SM9 algorithm combined with the device private key to generate a device fingerprint hash value, and embed the Beidou timing timestamp encryption block.

[0015] Optionally, the S1 further includes:

[0016] S15: extracting a power curve from the purification operation data, and using a long short-term memory network to predict the wind turbine power forecast value for the next 6 hours;

[0017] S16: Monitor network transmission delay through edge computing nodes and generate a network delay coefficient;

[0018] S17: Read the service level agreement of the current tenant from the tenant management database and parse the tenant priority weight;

[0019] S18: injecting the wind turbine power prediction value, network delay coefficient, and tenant priority weight into the unified format data frame to generate an enhanced unified format data frame, and performing a cyclic redundancy check calculation on the enhanced unified format data frame to generate a data check code;

[0020] S19: Encapsulate the device fingerprint hash value, timestamp encryption block, data check code, wind turbine power prediction value, network delay coefficient and tenant priority weight to generate a standardized data packet.

[0021] Optionally, the S2 includes:

[0022] S21: Parsing the device fingerprint hash value, the timestamp encryption block, the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight in the standardized data packet to extract the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight;

[0023] S22: constructing a three-dimensional decision model based on the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight, wherein the input parameter set of the three-dimensional decision model includes the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight;

[0024] S23: Input the input parameter set into the K-means clustering algorithm, modify the initial cluster center according to the weather forecast compensation factor, perform resource partitioning in combination with the tenant service level agreement constraints, and generate a resource partitioning result.

[0025] Optionally, the S2 further includes:

[0026] S24: establishing a logical mapping relationship of virtual computing units according to the resource partitioning result, and calculating a bandwidth allocation weight of each resource partition based on the network delay coefficient;

[0027] S25: Dynamically update the logical mapping relationship and bandwidth allocation weight of the virtual computing unit through a sliding time window mechanism to generate a resource allocation topology map.

[0028] Optionally, the S3 includes:

[0029] S31: Building an isolated environment for a fault event processing sandbox according to the logical mapping relationship of the virtual computing units and the bandwidth allocation weights in the resource allocation topology diagram;

[0030] S32: Constructing a wind turbine cluster operating status association graph in the fault event processing sandbox, where a node represents a single wind turbine device and an edge weight represents the electrical coupling degree between devices;

[0031] S33: Inputting the wind turbine group operation status association map into a dual-channel graph neural network model to calculate the fault propagation probability matrix.

[0032] Optionally, the S3 further includes:

[0033] S34: generating equipment control parameter corrections and spare parts scheduling paths based on the fault propagation probability matrix;

[0034] S35: Encapsulate the equipment control parameter correction value and spare parts scheduling path, and output a fault handling instruction set.

[0035] Optionally, the S4 includes:

[0036] S41: parsing the equipment control parameter correction value and spare parts scheduling path in the fault handling instruction set, and triggering a resource reallocation module;

[0037] S42: adjusting the logical mapping relationship of the virtual computing unit according to the device control parameter correction amount, and updating the bandwidth allocation weight based on the spare parts scheduling path;

[0038] S43: The adjusted logical mapping relationship and the updated bandwidth allocation weight are integrated to generate an updated resource allocation topology map.

[0039] Optionally, the S4 further includes:

[0040] S44: Generate a dynamic resource scheduling instruction according to the updated resource allocation topology map and encapsulate the instruction data packet;

[0041] S45: The dynamic resource scheduling instruction is signed and verified through the trusted execution environment and fed back to the edge computing node for execution, forming a closed-loop control link.

[0042] Beneficial effects of the present invention:

[0043] The present invention collects multi-dimensional operating data of wind turbine groups at edge computing nodes, and sequentially completes data cleaning, protocol unification, timestamp encryption, fingerprint hash signature and structured encapsulation to generate standardized data packets containing power prediction values, network delay coefficients and tenant priority weights, ensuring that each uploaded data has verifiable timeliness and source credibility. Combined with the three-dimensional decision model and the improved K-means clustering algorithm, the system automatically divides resource partitions based on real-time power requirements, network load status and tenant levels, constructs refined virtual computing unit mapping relationships and bandwidth weight allocation strategies, and significantly improves the matching degree between resource allocation and load balancing. In fault handling scenarios, diagnosis and scheduling decisions are executed through the sandbox mechanism within the TEE trusted execution environment, and an updated resource allocation topology map and resource scheduling instructions are formed to achieve a closed-loop control link from data perception, fault identification, parameter correction to resource reconstruction, greatly enhancing the system's adaptive scheduling capabilities.

[0044] The present invention embeds the wind turbine group operation status correlation map into a dual-channel graph neural network model, extracts power change trend features (LSTM) through the time series channel, and captures the electrical coupling relationship (GAT) between wind turbines through the spatial channel to achieve multi-dimensional data collaborative reasoning. Combining the health status of the equipment with the coupling edge weight, a fault propagation probability matrix is ​​constructed to dynamically identify the primary fault source and the affected nodes. On this basis, the high-confidence fault nodes output by the prediction model are used to automatically generate equipment control parameter corrections and spare parts scheduling paths. Bandwidth availability, physical distance and virtual computing unit location are comprehensively considered in path planning to achieve rapid recovery path planning with minimum bandwidth occupancy. This mechanism significantly improves the fault propagation chain analysis capability and intelligent operation and maintenance efficiency, and still has high reliability and response speed in a multi-fault interference environment.

[0045] The present invention introduces the national secret SM2 / SM4 encryption system, Beidou timing signature verification, CRC redundancy check and multi-level rollback mechanism in the process of scheduling instruction encapsulation and issuance. All scheduling instructions are signed and encrypted in TEE, and after being received by the edge node, multiple security checks are performed, including the legitimacy of the device fingerprint hash value, the validity of the signature, the accuracy of the timestamp and the data integrity verification. If any link fails, the execution of the instruction is terminated and the cloud platform resource status is triggered to roll back, restoring to the previous valid version of the resource topology to avoid resource configuration anomalies caused by instruction forgery, loss or delay. This mechanism effectively ensures the trustworthiness and consistency control of the entire process of resource instructions from generation to execution, significantly improving the system's anti-tampering ability and scheduling security level. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention;

[0048] Figure 2 Schematic diagram of the S2 process of an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0050] It should be noted that references in the specification to "one embodiment," "an embodiment," "an exemplary embodiment," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. In addition, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0051] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0052] like Figure 1-2 As shown, a wind farm cloud platform management method based on a multi-tenant architecture includes the following steps:

[0053] S1: The multi-dimensional operating data of the wind turbine group is collected through the edge computing node and a standardized data packet is generated through the protocol adaptation layer. The standardized data packet includes the device fingerprint hash value, the timestamp encryption block and the data check code;

[0054] S2: Input the standardized data packet into the multi-tenant resource allocation engine and generate a resource allocation topology based on the dynamic load balancing algorithm. The resource allocation topology includes the logical mapping relationship of the virtual computing units and the bandwidth allocation weights.

[0055] S3: Build a fault event processing sandbox based on the resource allocation topology, perform correlation fault diagnosis in an isolated environment, and output a fault handling instruction set, which includes equipment control parameter corrections and spare parts scheduling paths.

[0056] S4: Based on the fault handling instruction set, the resource reallocation module is triggered to generate dynamic resource scheduling instructions and feed them back to the edge computing node to form a closed-loop control link.

[0057] S1 includes:

[0058] S11: Collect multi-dimensional operation data of the wind turbine group through the edge computing nodes deployed in the wind farm. The multi-dimensional operation data of the wind turbine group includes the following:

[0059] Vibration spectrum: The accelerometer installed on the fan base collects the vibration signal of the fan during operation and performs frequency domain analysis.

[0060] Power curve: A wind turbine power sensor (such as a current and voltage sensor) is installed in the wind turbine's electrical system to collect the wind turbine's output power at different wind speeds.

[0061] Yaw angle: The yaw sensor installed in the wind turbine nacelle collects the yaw angle of the wind turbine blades to determine the relative angle between the wind turbine and the wind direction.

[0062] S12: The collected multi-dimensional operation data will be passed as input to the data cleaning submodule. In the data cleaning submodule, the sliding window outlier detection algorithm is used to remove abnormal jump data to ensure data quality. The specific steps are as follows:

[0063] Sliding Window: Use a fixed-length sliding window W to segment the data. Each time the window slides one data point, all data points within the window are checked for outliers.

[0064] Outlier detection: By calculating the mean μ and standard deviation σ of the data in the window, it is determined whether the data point exceeds the set threshold k:

[0065]

[0066] Where x is the current data point, μ is the mean of the data in the window, σ is the standard deviation of the data in the window, and k is the set outlier detection threshold coefficient;

[0067] After cleaning, the purified running data is output to remove all abnormal jump data;

[0068] S13: The cleaned operating data is input into the protocol conversion submodule. In this submodule, the appropriate protocol is dynamically selected for data conversion based on the device's communication protocol type (such as Modbus, IEC104, or OPC-UA). The converted data generates a unified data frame with the following structure:

[0069] D frame =[Header,Payload,CRC];

[0070] Among them, D frame It is a data frame with a unified format. The Header is the protocol header, which contains information such as the protocol identifier. The Payload is the payload, which contains the actual transmitted data. The CRC is the data check code used for data verification.

[0071] S14: Input the unified format data frame into the data signature submodule. This module uses the national encryption 5M9 algorithm combined with the device private key to generate the device fingerprint hash value and embed it into the timestamp encryption block generated by Beidou timing. The specific steps are as follows:

[0072] Device fingerprint hash value: Use the national secret SM9 algorithm to sign the data frame and generate the device fingerprint hash value: Device Fingerprint Hash = H SM9 (D frame ,Private Key);

[0073] Among them, H SM9 Is the SM19 signature algorithm; D frame It is a data frame in a unified format. PrivateKey is the device private key.

[0074] Timestamp encryption block: Timestamp T generated by BeiDou timing system timestamp Perform encryption processing and embed it into the data frame.

[0075] S15: Extract the power curve from the purification operation data and use the **long short-term memory network to predict the wind turbine power forecast value for the next 6 hours. The specific steps are as follows:

[0076] Extract power curve: Extract power curve data from the purified multi-dimensional operation data of the fan group. These data reflect the power output of the fan under different wind speed conditions.

[0077] Data preprocessing: The extracted power curve data is normalized so that the data values ​​are in a uniform scale (normalized to the [0,1] interval).

[0078] Long Short-Term Memory (LSTM) network model training: Based on historical power curve data, a LSTM network model is used for training. The LSTM model effectively captures long-term dependencies in time series data. This model can be used to predict wind turbine power trends within the next six hours.

[0079] Power forecast for the next 6 hours: Through the trained long short-term memory network model, the power data of the current time period is input and the wind turbine power forecast value for the next 6 hours is output. The forecast value is based on past power data and wind speed conditions to estimate future power output.

[0080] S16: Monitor network transmission delay through edge computing nodes and generate network delay coefficient. The specific steps are as follows:

[0081] Latency monitoring: The edge computing node continuously monitors the network connection latency between the edge computing node and the cloud platform or central server. This can be achieved by sending timestamped data packets and calculating their round-trip time (RTT).

[0082] Transmission delay calculation: Each time a data packet is sent, the sending timestamp and receiving timestamp are recorded to calculate the network transmission delay T delay :

[0083] T d e l ay =T receive -T send ;

[0084] Among them, T send The timestamp when the data packet is sent, T receive The timestamp when the data packet arrives at the backend server.

[0085] Delay coefficient generation: Based on multiple monitored delay values, the network delay coefficient α is calculated as the influencing factor of network transmission delay. The delay coefficient is obtained by calculating the average and maximum delay values:

[0086]

[0087] Where n is the number of sampling times of the delay value, T delay [i] is the transmission delay value measured for the i-th time, and the output result is the network delay coefficient.

[0088] S17: Read the current tenant's service level agreement (SLA) from the tenant management database and parse the tenant priority weight. The specific steps are as follows:

[0089] Tenant Service Level Agreement Query: Query the tenant management database to obtain the current tenant's Service Level Agreement (SLA). This agreement defines the tenant's priority for resource requirements, such as bandwidth and computing power requirements.

[0090] Analyze tenant priority weights: Analyze tenant priority weights based on their service level agreements. Priority weights represent the importance of a tenant in multi-tenant resource allocation. Generally, a higher weight indicates a tenant with more urgent resource needs and warrants priority.

[0091] Weight extraction: Parse the priority information in the tenant agreement, such as quantifying the tenant's resource requirements through weight coefficients. Express the priority weight as a numerical value to indicate the relative size of the tenant's priority.

[0092] S18: Inject the wind turbine power prediction value, network delay coefficient and tenant priority weight into the unified format data frame to generate an enhanced unified format data frame, and perform cyclic redundancy check (CRC) calculation on the enhanced unified format data frame to generate a data check code. The specific steps are as follows:

[0093] Injecting wind turbine power prediction value, network delay coefficient and tenant priority weight: inserting the previously generated wind turbine power prediction value, network delay coefficient and tenant priority weight into the payload part of the unified format data frame;

[0094] Calculate CRC check: Perform cyclic redundancy check (CRC) calculation on the enhanced unified format data frame to ensure data integrity. The CRC calculation process is as follows:

[0095]

[0096] Among them, D frame [i] is the i-th bit of the data frame, and P is the selected generator polynomial;

[0097] S19: Encapsulate the device fingerprint hash value, timestamp encryption block, data checksum, wind turbine power prediction value, network delay coefficient and tenant priority weight to generate a standardized data packet.

[0098] S2 includes:

[0099] S21: Parse the device fingerprint hash value, timestamp encryption block, fan power prediction value, network delay coefficient, and tenant priority weight in the standardized data packet to extract the fan power prediction value, network delay coefficient, and tenant priority weight. The specific steps are as follows:

[0100] The Beidou timing information in the timestamp encrypted block is decrypted using the decryption key to obtain the precise time reference (UTC+8 time zone) for data collection.

[0101] The device fingerprint hash value is compared with the hash value in the pre-stored device identity library. If the match degree exceeds 99%, the data source is confirmed to be legal.

[0102] The wind turbine power prediction value, network delay coefficient and tenant priority weight are separated from the standardized data packet, retaining the original data format.

[0103] A cyclic redundancy check (CRC) is performed on the separated wind turbine power prediction value to verify its consistency with the data check code.

[0104] S22: Construct a three-dimensional decision model based on the predicted wind turbine power value, the network delay coefficient, and the tenant priority weight. The input parameter set of the three-dimensional decision model includes the predicted wind turbine power value, the network delay coefficient, and the tenant priority weight. The specific steps are as follows:

[0105] The wind turbine power prediction value is normalized, and the formula is:

[0106]

[0107] (output: power impact factor, range [0,1]);

[0108] The network delay coefficient is converted into a logarithmic scale delay level parameter using the formula:

[0109]

[0110] A resource allocation priority matrix is ​​generated according to the tenant priority weights, where the matrix element value is equal to the inverse of the tenant priority weight.

[0111] Establish a three-dimensional coordinate system and define:

[0112] X-axis: power impact factor;

[0113] Y-axis: delay level parameter;

[0114] Z axis: priority matrix value;

[0115] Calculate the resource demand vector for each tenant in a three-dimensional coordinate system:

[0116]

[0117] S23: Input the input parameter set into the K-means clustering algorithm, modify the initial cluster center according to the weather forecast compensation factor, and perform resource partitioning in combination with the tenant service level agreement constraints to generate a resource partitioning result. The specific steps are as follows:

[0118] Get the wind speed forecast data for the next 6 hours and calculate the weather forecast compensation factor:

[0119]

[0120] Correct the initial cluster center coordinates: C new =C original ×(1+α)

[0121] Apply tenant service level agreement (SLA) constraints and set a protection radius r = 0.2 × priority matrix value for the cluster center where high-priority tenants are located;

[0122] Perform weighted Euclidean distance calculation to divide resource partitions:

[0123] Distance weight = (w x ,w y ,w z )=(0.5,0.3,0.2);

[0124]

[0125] Output resource partitioning results, including partition number, list of included virtual computing units, and constraints (SLA protection radius, resource quota upper limit);

[0126] S24: Establish a logical mapping relationship of virtual computing units based on the resource partitioning result and calculate the bandwidth allocation weight. The specific steps are as follows:

[0127] Virtual computing units (VMs) are allocated to each resource partition, and the number is determined by the number of tenants in the partition and the weighted value of the priority matrix.

[0128] Create a logical mapping table to record: virtual computing unit ID, resource partition number, and bound physical server IP address;

[0129] The bandwidth allocation weight of each partition is calculated based on the network delay coefficient, which is expressed as:

[0130]

[0131] Normalize the bandwidth allocation weights so that the sum is 100%.

[0132] S25: Dynamically update the logical mapping relationship and bandwidth allocation weights of the virtual computing units through the sliding time window mechanism to generate a resource allocation topology. The specific steps are as follows:

[0133] Sliding time window mechanism: Define a 5-minute time window that slides every 5 minutes to recollect and update data. Within each time window, monitor changes in wind turbine power forecasts, network latency coefficients, and tenant priority weights.

[0134] Trigger condition: If the network delay coefficient changes by more than 10% at the end of the current time window, or the time window period ends, a data update is triggered. The rate of change exceeding 10% can be calculated using the following formula:

[0135]

[0136] Among them D current and D previous Represent the network delay coefficients in the current and previous time windows respectively;

[0137] Re-collect the latest wind turbine power forecast value, network delay coefficient, and tenant priority weight.

[0138] Recollect data: At the beginning of each new time window, recollect the following data:

[0139] Wind turbine power forecast: Get the latest wind turbine power forecast from the data source to update resource requirements.

[0140] Network delay coefficient: Remeasure network delay and update related data.

[0141] Tenant priority weight: Recalculate tenant priority based on changes in tenant service level agreements (SLAs).

[0142] Repeat steps S22-S24 according to the latest data to generate updated logical mapping relationships and bandwidth allocation weights.

[0143] Repeat steps S22-S24: Based on the latest data collected, re-execute the operations of steps S22 to S24, including:

[0144] S22: Reconstruct the three-dimensional decision model based on the new wind turbine power prediction value, network delay coefficient and tenant priority weight.

[0145] S23: Use the K-means clustering algorithm to process the data and generate new resource partitioning results.

[0146] S24: Recalculate the logical mapping relationship of the virtual computing units according to the new resource partitioning result, and calculate the bandwidth allocation weight based on the network delay coefficient.

[0147] Compare the old and new resource allocation topologies and only update the differences, reducing system overhead.

[0148] Topology comparison: Compare the resource allocation topology before and after an update to identify any differences. Update only the differences, avoiding recalculation of the entire system and reducing computational overhead. By comparing the new resource allocation topology with the old one, you can identify changes in resource allocation or bandwidth weights.

[0149] The specific operations are:

[0150] Computing difference part: By comparing the mapping relationship and bandwidth allocation of virtual computing units (VCUs) in the new and old topologies, the changed resource areas are identified.

[0151] Update only the differences: By marking the difference areas, unnecessary calculations are reduced and only the changed parts are updated.

[0152] Push the updated resource allocation topology map to the edge computing node for execution.

[0153] Push the updated topology map: Push the latest resource allocation topology map (including the updated virtual computing unit mapping relationship and bandwidth allocation weight) to the edge computing node for actual resource allocation operations.

[0154] Edge computing node execution: Edge computing nodes schedule and allocate resources based on the latest topology map to ensure optimal utilization of resources within each time window.

[0155] S3 includes:

[0156] S31: Based on the logical mapping relationship of the virtual computing units and the bandwidth allocation weights in the resource allocation topology, an isolated environment for the fault event processing sandbox is constructed. The specific steps are as follows:

[0157] The logical mapping relationship table of the virtual computing units in the resource allocation topology diagram is read to create virtualized sandbox containers decoupled from the physical server, and each container is bound to a specified virtual computing unit ID.

[0158] An independent network channel is allocated to each virtualized sandbox container based on the bandwidth allocation weight. The channel bandwidth calculation formula is:

[0159]

[0160] Load the trusted execution environment in the virtualized sandbox container, use the national secret SM4 algorithm to initialize the encrypted memory area, and store the preset key pair;

[0161] S32: Construct a wind turbine cluster operating status association graph in the fault event processing sandbox. The nodes represent individual wind turbine devices, and the edge weights represent the electrical coupling between devices. The specific steps are as follows:

[0162] The device fingerprint hash value is extracted from the standardized data packet and associated with the real-time operating data (vibration spectrum, power curve) of the corresponding wind turbine equipment.

[0163] Define graph node attributes, including node ID and node attribute set, where;

[0164] Node ID: device fingerprint hash value;

[0165] Node attribute set: real-time power value (kW), health status code (0-100), temperature sensor reading;

[0166] Calculate the edge weight of the electrical coupling between devices using the following formula:

[0167] The electrical distance is obtained based on the wind farm electrical wiring diagram;

[0168] Store the associated graph of wind turbine cluster operation status in the encrypted memory area of ​​the Trusted Execution Environment (TEE) and generate a graph access token;

[0169] S33: Input the wind turbine cluster operating status association map into the dual-channel graph neural network model to calculate the fault propagation probability matrix. The specific steps are as follows:

[0170] Time series data processing channel:

[0171] Input: real-time power value time series in node attributes (length 24 hours);

[0172] Processing: Use bidirectional LSTM network to extract time dimension features and output time series feature vector

[0173] Spatial topology processing channel:

[0174] Input: edge weight (electrical coupling) and node health status code;

[0175] Processing: Use the Graph Attention Network (GAT) to calculate spatial correlation features and output topological feature vectors

[0176] Feature fusion and probability calculation:

[0177] Fusion formula:

[0178] Calculate the fault propagation probability through the fully connected layer:

[0179] Where σ is the Sigmoid function, which outputs the fault propagation probability matrix P.

[0180] S34: Generate equipment control parameter corrections and spare parts scheduling paths based on the fault propagation probability matrix. The specific steps are as follows:

[0181] Dynamic threshold judgment:

[0182] when When , device i is marked as a primary fault source. is an existential quantifier, when there exists a j such that P ij >0.7, that is, the probability of fault propagation between device i and at least one device j is greater than 0.7;

[0183] When 0.4 <P ij When ≤0.7, the marked device j is a secondary affected device;

[0184] Calculate the equipment control parameter correction:

[0185] Δθ i =α·(1-health status code / 100)·Σ j P ij , (α is the preset learning rate, and the health status code comes from the node attributes);

[0186] Generate spare parts scheduling path:

[0187] Input: virtual computing unit location information (physical server IP address) in the resource allocation topology diagram;

[0188] Solution: Calculate the shortest physical path from the warehouse to the faulty device using the Dijkstra algorithm. The path weight is: Weight = Distance × (1 - Available bandwidth of the path / Total bandwidth);

[0189] S35: Encapsulate the equipment control parameter correction value and spare parts scheduling path, and output the fault handling instruction set. The specific steps are as follows:

[0190] Classify the device control parameter corrections according to the device fingerprint hash value and generate an encrypted instruction block:

[0191] Instruction block structure: {device fingerprint hash value, parameter correction amount, effective timestamp};

[0192] Encode the spare parts dispatch path into JSON format and add Beidou geographic coordinate encryption tags;

[0193] Digitally sign instruction blocks and path data using the device private key within the Trusted Execution Environment (TEE);

[0194] Encapsulate the signed data packet, generate a fault handling instruction set, and push it to the edge computing node for execution.

[0195] S4 includes:

[0196] S41: Analyze the equipment control parameter correction value and spare parts scheduling path in the fault handling instruction set, and trigger the resource reallocation module. The specific steps are as follows:

[0197] Decrypt the encrypted instruction block in the fault handling instruction set within the Trusted Execution Environment (TEE), extract the device control parameter correction value, and verify the consistency of the device fingerprint hash value with the pre-stored device identity library. If the hash value match is less than 99%, the instruction is discarded.

[0198] Parse the JSON structure data of the spare parts scheduling path, decrypt the Beidou geographic coordinate encryption tag, and extract the latitude and longitude information and path weight parameters of the path nodes.

[0199] The national secret SM2 algorithm is used to verify the legitimacy of the digital signature and to check whether the signature public key is valid in the certificate bound to the device fingerprint hash value.

[0200] If all checks pass, the initialization process of the resource reallocation module is triggered to load the logical mapping relationship table and bandwidth allocation weight table of the current virtual computing unit.

[0201] S42: Adjust the logical mapping relationship of the virtual computing unit according to the device control parameter correction amount, and update the bandwidth allocation weight based on the spare parts scheduling path. The specific steps are as follows:

[0202] Traverse the device control parameter correction list, match the target device in the logical mapping table according to the device fingerprint hash value, and update the logical mapping relationship. The update calculation formula is:

[0203] Where Δθ i is the parameter correction of device i, θ max The maximum threshold of the device control parameter, read from the device configuration library.

[0204] Recalculate the bandwidth allocation weight based on the physical distance and available bandwidth of the spare parts scheduling path.

[0205] The update formula is:

[0206] The adjusted logical mapping relationship and bandwidth allocation weight are written to the temporary configuration library and marked as "pending effective version".

[0207] S43: The adjusted logical mapping relationship and the updated bandwidth allocation weight are integrated to generate an updated resource allocation topology. The specific steps are as follows:

[0208] The adjusted logical mapping relationship table and bandwidth allocation weight table are read from the temporary configuration library as an input parameter set.

[0209] The three-dimensional decision model is used to reconstruct the resource demand vector. The resource vector of each tenant is redefined as: Where x′ i is the power impact factor, y′ i is the delay level parameter, z′ i is the priority matrix value.

[0210] According to the sliding time window mechanism (window period is 5 minutes), the cluster center coordinates are dynamically updated to reflect the time-varying nature of resource demand;

[0211] The above parameters are integrated to generate an updated resource allocation topology. The topology includes: the updated logical mapping relationship of virtual computing units, the new bandwidth allocation weight matrix, the timestamp encryption block generated by Beidou timing, and the cyclic redundancy check code.

[0212] S44: Generate dynamic resource scheduling instructions based on the updated resource allocation topology map and encapsulate the instruction data packet. The specific steps are as follows:

[0213] The resource allocation topology structure is translated into a set of instruction fields, which include a logic mapping relationship table, a bandwidth allocation weight matrix, a timestamp encryption block, and a data check code.

[0214] The Beidou timing module is introduced to generate the timestamp encryption block of the current time point.

[0215] The CRC32 algorithm is used to calculate the data check code and perform cyclic redundancy check on the scheduling instructions to ensure the integrity of the instructions.

[0216] Encapsulate the above fields to generate a complete dynamic resource scheduling instruction data packet for signature verification by the trusted execution environment.

[0217] S45: The dynamic resource scheduling instruction is signed and verified through the trusted execution environment and fed back to the edge computing node for execution, forming a closed-loop control link. The specific steps are as follows:

[0218] In a trusted execution environment, the device private key is used to sign the encapsulated scheduling instruction data packet using the national secret SM2 algorithm.

[0219] The signed instruction data packet is transmitted to the target edge computing node through a secure transmission tunnel (such as IPSec VPN).

[0220] After receiving the data packet, the edge computing node performs three checks, including: checking whether the CRC checksum matches, decrypting the timestamp encryption block, and verifying whether the difference with the local time is less than ±30 seconds, and verifying whether the digital signature matches the certificate of the device fingerprint hash value.

[0221] If all checks pass, the edge computing node migrates the virtual computing unit according to the updated logical mapping relationship and adjusts the network channel queue according to the new bandwidth allocation weight.

[0222] If no execution feedback is received within 5 seconds, the cloud platform will trigger the resource rollback mechanism and automatically restore to the last valid version of the resource allocation topology.

[0223] Record all instructions, execution results, and topology version numbers of this closed-loop control link and write them into the global resource allocation log as a basis for scheduling audits and subsequent optimization.

[0224] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0225] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A wind farm cloud platform management method based on a multi-tenant architecture, characterized in that: The following steps are involved: S1: Collect multi-dimensional operation data of the wind turbine group through the edge computing node, and generate a standardized data packet through the protocol adaptation layer. The standardized data packet includes the device fingerprint hash value, the timestamp encryption block and the data check code; S2: Input the standardized data packet into the multi-tenant resource allocation engine, and generate a resource allocation topology map based on a dynamic load balancing algorithm. The resource allocation topology map includes the logical mapping relationship of the virtual computing units and the bandwidth allocation weights; S3: Build a fault event processing sandbox based on the resource allocation topology, perform correlation fault diagnosis in an isolated environment, and output a fault handling instruction set, which includes equipment control parameter corrections and spare parts scheduling paths; S4: Based on the fault handling instruction set, the resource reallocation module is triggered to generate dynamic resource scheduling instructions and feed them back to the edge computing node to form a closed-loop control link.

2. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 1, characterized in that: Said S1 comprises: S11: Collecting multi-dimensional operation data of the wind turbine group through the edge computing node, wherein the multi-dimensional operation data of the wind turbine group includes a vibration spectrum, a power curve, and a yaw angle; S12: Inputting the multi-dimensional operation data of the wind turbine group into a data cleaning submodule, removing abnormal jump data based on a sliding window outlier detection algorithm, and generating purified operation data; S13: Inputting the purification operation data into the protocol conversion submodule, dynamically selecting Modbus, IEC104 or OPC-UA conversion protocol according to the device communication protocol type, and generating a unified format data frame; S14: Input the unified format data frame into the data signature submodule, use the national secret SM9 algorithm combined with the device private key to generate a device fingerprint hash value, and embed the Beidou timing timestamp encryption block.

3. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 2, characterized in that: Said S1 further comprises: S15: extracting a power curve from the purification operation data, and using a long short-term memory network to predict the wind turbine power forecast value for the next 6 hours; S16: Monitor network transmission delay through edge computing nodes and generate a network delay coefficient; S17: Read the service level agreement of the current tenant from the tenant management database and parse the tenant priority weight; S18: injecting the wind turbine power prediction value, network delay coefficient, and tenant priority weight into the unified format data frame to generate an enhanced unified format data frame, and performing a cyclic redundancy check calculation on the enhanced unified format data frame to generate a data check code; S19: Encapsulate the device fingerprint hash value, timestamp encryption block, data check code, wind turbine power prediction value, network delay coefficient and tenant priority weight to generate a standardized data packet.

4. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 3, characterized in that: The S2 includes: S21: Parsing the device fingerprint hash value, the timestamp encryption block, the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight in the standardized data packet to extract the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight; S22: constructing a three-dimensional decision model based on the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight, wherein the input parameter set of the three-dimensional decision model includes the wind turbine power prediction value, the network delay coefficient, and the tenant priority weight; S23: Input the input parameter set into the K-means clustering algorithm, modify the initial cluster center according to the weather forecast compensation factor, perform resource partitioning in combination with the tenant service level agreement constraints, and generate a resource partitioning result.

5. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 4, characterized in that: Said S2 further comprises: S24: establishing a logical mapping relationship of virtual computing units according to the resource partitioning result, and calculating a bandwidth allocation weight of each resource partition based on the network delay coefficient; S25: Dynamically update the logical mapping relationship and bandwidth allocation weight of the virtual computing unit through a sliding time window mechanism to generate a resource allocation topology map.

6. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 5, characterized in that: The S3 includes: S31: Building an isolated environment for a fault event processing sandbox according to the logical mapping relationship of the virtual computing units and the bandwidth allocation weights in the resource allocation topology diagram; S32: Constructing a wind turbine cluster operating status association graph in the fault event processing sandbox, where a node represents a single wind turbine device and an edge weight represents the electrical coupling degree between devices; S33: Inputting the wind turbine group operation status association map into a dual-channel graph neural network model to calculate the fault propagation probability matrix.

7. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 6, characterized in that: Said S3 further comprises: S34: generating equipment control parameter corrections and spare parts scheduling paths based on the fault propagation probability matrix; S35: Encapsulate the equipment control parameter correction value and spare parts scheduling path, and output a fault handling instruction set.

8. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 7, characterized in that: The S4 includes: S41: parsing the equipment control parameter correction value and spare parts scheduling path in the fault handling instruction set, and triggering a resource reallocation module; S42: adjusting the logical mapping relationship of the virtual computing unit according to the device control parameter correction amount, and updating the bandwidth allocation weight based on the spare parts scheduling path; S43: The adjusted logical mapping relationship and the updated bandwidth allocation weight are integrated to generate an updated resource allocation topology map.

9. A wind farm cloud platform management method based on a multi-tenant architecture according to claim 8, characterized in that: Said S4 further comprises: S44: Generate a dynamic resource scheduling instruction according to the updated resource allocation topology map and encapsulate the instruction data packet; S45: The dynamic resource scheduling instruction is signed and verified through the trusted execution environment and fed back to the edge computing node for execution, forming a closed-loop control link.

Images