Oil and gas field electric energy metering control method and device based on a hon g meng system
By using an oil and gas field power metering and control method based on the HarmonyOS system, and employing an LSTM prediction model and a differentiated algorithm set node network, combined with progressive updates and hierarchical optimization strategies, the traditional system's metering errors, low resource utilization, and power allocation delays under complex operating conditions are solved, achieving high-precision and real-time power management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XI AN SHANGDING ENERGY TECH CO LTD
- Filing Date
- 2025-09-10
- Publication Date
- 2026-04-17
AI Technical Summary
Traditional power metering and control systems in oil and gas fields struggle to achieve high-precision metering and real-time control under complex operating conditions. They suffer from low resource utilization, are prone to business interruptions due to model updates, have a high false alarm rate in fault detection, experience delayed power distribution response, and are difficult to achieve multi-regional collaborative scheduling.
An oil and gas field power metering and control method based on the HarmonyOS system is adopted. Ideal data curves are generated through LSTM prediction models, a differentiated algorithm set node network is constructed, a progressive model update process and a hierarchical optimization power allocation strategy are designed, and dynamic threshold fault detection and distributed auction mechanism are combined to achieve high-precision correction of sensor data, elastic resource scheduling, stability assurance and power optimization.
It significantly improved the calibration accuracy of sensor data, enhanced the processing efficiency of edge networks, reduced the false alarm rate of fault detection, improved the efficiency and fairness of power distribution, and realized the real-time and economical management of power in oil and gas fields.
Smart Images

Figure CN121172979B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electrical sensor technology, and in particular to a method and device for metering and controlling electricity in oil and gas fields based on the HarmonyOS system. Background Technology
[0002] As a core component of energy production, the power metering and control system in oil and gas fields plays a crucial role in real-time monitoring of the power grid, optimizing energy consumption allocation, and ensuring the safe operation of equipment. Traditional systems often employ a centralized architecture, relying on fixed-threshold sensor data processing and static resource allocation strategies, which struggle to meet the demands for high-precision metering and real-time control under complex operating conditions. The application of the HarmonyOS distributed operating system in the Industrial Internet of Things (IIoT) field provides a technological foundation for building a flexible and intelligent next-generation electrical control system.
[0003] The existing technology has significant drawbacks: First, the sensor data compensation uses a static filtering algorithm, which cannot dynamically adapt to interference such as temperature drift and mechanical vibration, leading to the accumulation of measurement errors. Second, the edge computing nodes are deployed in a homogeneous manner, resulting in fierce competition for high-frequency algorithm resources while low-frequency algorithms remain idle, with an overall resource utilization rate of less than 40%. Third, the model update adopts a full-node forced coverage mechanism, which is prone to causing business interruptions due to version compatibility issues. Fourth, fault detection relies on fixed thresholds, and the false alarm rate is as high as 35% under extreme conditions such as high temperature and heavy load. Fifth, the power distribution adopts centralized optimization, with a response latency of over 500ms, making it difficult to achieve multi-region collaborative scheduling.
[0004] This invention proposes a power metering and control method for oil and gas fields based on the HarmonyOS system. It overcomes the aforementioned bottlenecks through three core technologies: First, an innovative dynamic compensation algorithm (LSTM prediction + adaptive gain function correction) improves sensor data accuracy; second, a differentiated algorithm set node network (dynamic coverage allocation + completeness constraints) is constructed to achieve elastic scheduling of edge resources; finally, a progressive model update process (quadruple verification + layered rollback) and a game-auction collaborative allocation mechanism (cooperative game + Vickrey auction) are designed to improve control real-time performance and economy while ensuring system stability, thus comprehensively optimizing the efficiency of power management in oil and gas fields. Summary of the Invention
[0005] To overcome the problems mentioned in the background art, the present invention proposes a method and device for metering and controlling electricity in oil and gas fields based on the HarmonyOS system.
[0006] The technical solution of this invention is: an oil and gas field power metering and control method based on the HarmonyOS system, comprising the following steps:
[0007] S11: Enables virtualized access and standardized data acquisition of electrical sensor devices via HarmonyOS soft bus;
[0008] S12: Build a distributed processing network on edge computing nodes to perform data compensation, real-time computing, fault detection, power allocation optimization strategy generation, and equipment status analysis;
[0009] S13: Use the HarmonyOS secure communication protocol to transmit processed data to the cloud;
[0010] S14: Perform long-term data backup and management model updates in the cloud;
[0011] S15: Send the control commands generated at the edge to the terminal device for execution and return the results;
[0012] S16: Implement full lifecycle management of devices based on the HarmonyOS device management framework.
[0013] Preferably, when performing standardized data collection, the sampling frequency satisfies the following:
[0014] ;
[0015] in, For adaptive adjustment coefficient, And when When activated, among which, For real-time sampling frequency, The preset minimum sampling frequency, For signal acceleration, For signal-to-noise ratio, It is the negative square of a second.
[0016] Preferably, a distributed processing network is built at the edge computing nodes to perform data compensation, specifically including:
[0017] S21: Data stability detection, real-time calculation of fluctuation measures of electrical sensor data, including:
[0018] ;
[0019] ;
[0020] If the data variance is less than or equal to 0.05 and the data range is less than or equal to 0.1, proceed to the next step.
[0021] in, Let N be the data variance, and N be the number of data points within the sampling window. For the i-th sample value, The mean of the sampling window, For extremely poor data, It is a data vector;
[0022] S22: Ideal curve generation, calling the preset LSTM prediction model, inputting historical data window. The sequence outputs the ideal data curve;
[0023] S23: Compensation calculation: Calculate the residual between the measured value and the ideal data curve, and dynamically generate the compensation function.
[0024] ;
[0025] in, The compensation value at time t. This is the integral of the deviation over the past second. It is an exponential function. It is an exponential function;
[0026] S24: Output correction, which superimposes the real-time compensation value calculated by the compensation function onto the original signal.
[0027] ;
[0028] in, The corrected electrical sensor value. These are the original electrical sensor values;
[0029] S25: Abnormal freeze mechanism. When the data variance of the electrical sensor is detected to be greater than 0.08 and the data range is greater than 0.15 for three consecutive cycles, compensation is suspended and a device diagnostic request is generated.
[0030] Preferably, when constructing a distributed processing network at edge computing nodes, the constructed distributed processing network is a node network containing a differentiated algorithm set, specifically including the following steps:
[0031] S31: Define the set of processing algorithms A= and its calculated weight vector Among them, weight Representation Algorithm The proportion of computational load in the data processing flow of a distributed processing network;
[0032] S32: Configure a subset of algorithms for each edge computing node, so that the algorithm The node coverage rate reaches the set standard, which is calculated using the following formula:
[0033] ;
[0034] in, For the algorithm The standard value for node coverage;
[0035] S33: Deploy a node network based on coverage constraints, where when the standard node coverage is less than or equal to 0.38, the node coverage is determined according to the standard node coverage. Linear configuration: When the standard node coverage is greater than 0.38, the configuration is based on a node coverage of 1.
[0036] S34: The network topology satisfies the completeness condition:
[0037] ;
[0038] Where n is the total number of algorithms.
[0039] Preferably, a distributed processing network is built at edge computing nodes to perform real-time computing, including:
[0040] S41: Input feature vector, which contains statistical features of the data stream, environmental parameters and historical load sequences, and improves the output processing demand vector of the decision tree model by training gradients;
[0041] S42: Normalize the processing requirement vector:
[0042] ;
[0043] in, This is the normalized demand vector. This is the original demand vector. For the algorithm The proportion of demand;
[0044] S43: Solving the assignment matrix, first establish the optimization problem:
[0045] ;
[0046] And satisfy the constraints:
[0047] ;
[0048] ;
[0049] in, The proportion of task i allocated to node j. The communication latency of task i on node j, Does node j support the algorithm? , Let j be the resource utilization rate of node j. This represents the maximum total system resources.
[0050] Preferably, when constructing a distributed processing network at edge computing nodes and generating power allocation optimization strategies, the specific steps include:
[0051] S51: Physical topology partitioning. Based on the power supply line impedance parameters and controller deployment location, the area is divided into N game units, while satisfying electrical isolation conditions.
[0052] ;
[0053] in, Let p be the partition number of the power grid. Let q be the partition number of the power grid. The impedance magnitude between nodes i and j;
[0054] S52: Inter-regional cooperative game, using a cooperative game model to solve for inter-regional electricity quotas;
[0055] S53: Distributed auction within the region, where resources are allocated to devices within the region through a distributed auction mechanism.
[0056] As a preferred option, when using a cooperative game theory model to solve for inter-regional electricity quotas, the specific steps include:
[0057] S61: Define the utility function:
[0058] ;
[0059] in, The electricity quota allocated to region i For benefit weighting coefficients, The elastic coefficient, The historical average load for the region;
[0060] S62: Establish an optimization problem:
[0061]
[0062]
[0063] ;
[0064] in, This is a penalty for peak system load. This is the peak adjustment factor. For the system's maximum power supply capacity, This is the lower limit of the regional quota. This is the upper limit of the regional quota;
[0065] S63: The optimal quota vector is solved using the alternating direction method.
[0066] As a preferred option, when resources are secondary allocated through a distributed auction mechanism within the region, the specific details include:
[0067] S71: For each game unit, the device submits a bid function:
[0068] ;
[0069] in, Let d be the electrical energy requirement of device d. This is the priority coefficient. Let d be the bid price for equipment. Prioritize device load. The total bid price for equipment d;
[0070] S72: Based on unit price Ascending order sorting device, among which, The price quoted per unit of electricity for device d;
[0071] S73: Allocate on demand until satisfied. , winning bid ,in, Let i be the total electricity quota for region i. The actual winning bid volume for equipment d is given, and the remaining quota is the total regional quota minus the allocated volume.
[0072] S74: Settlement price for the winning bid equipment:
[0073] ;
[0074] in, The final payment amount for device d. This represents the lowest unit price among the equipment that did not win the bid.
[0075] Preferably, a distributed processing network is built on the edge computing nodes, and when performing fault detection, it specifically includes:
[0076] S81: Dynamic threshold setting, generating real-time detection thresholds for device faults, wherein the real-time detection thresholds are calculated using the following formula:
[0077] ;
[0078] in, As the baseline detection threshold, For real-time temperature, For reference temperature, and The first compensation coefficient is... This is the second compensation coefficient. For equipment load rate, This represents the maximum load rate of the equipment.
[0079] S82: Constructing the correlation matrix and establishing the fault characteristic response matrix:
[0080] ;
[0081] in, This represents the conditional probability of sensor feature k occurring when fault type i occurs. Let j be the sensor feature. For fault type i;
[0082] S83: Real-time data verification. When the sensor group detects suspected fault characteristics, it reads the real-time data from the relevant sensors and calculates the theoretical characteristic quantity. And judge based on the set data anomaly conditions, which are:
[0083] ;
[0084] in Let j be the real-time measured quantity of sensor j. Let be the theoretical characteristic quantity of sensor j. The historical average data for sensor j, Let be the upper limit of the measurement range of sensor j;
[0085] S84: Fault classification response, executes a classification response based on the degree of abnormal deviation.
[0086] As a preferred option, cloud-based management of model updates specifically includes:
[0087] S91: Update triggered. When an edge node detects that the model performance degradation exceeds a threshold, it sends an update request packet to the cloud.
[0088] S92: Security verification, cloud verification of request signature validity, and generation of a new model after confirming that the decay condition is met;
[0089] S93: Privacy processing: Laplace noise is added to the training data to achieve differential privacy protection. The privacy budget is controlled between 0.6 and 0.9. The noise scale is inversely proportional to the maximum sensitivity of the data.
[0090] S94: Layered transmission, which divides model parameters into a base layer and an adaptation layer. The base layer parameters are sent to all edge nodes in full, while the adaptation layer parameters are only updated to a subset of target nodes in a differentiated manner.
[0091] S95: Progressive deployment, employing a progressive deployment mechanism to update the model within edge nodes, specifically:
[0092] A. Initially, select 30% of the nodes that are not running the old model to deploy the new model;
[0093] B. Migrate 30% of the processing tasks to the update node;
[0094] C. Validate the new model. If the validation is successful, proceed to the next step. If the validation fails, revert the first batch of nodes to the old model and send a new update request to the cloud.
[0095] D. Deploy the new model on the remaining 70% of nodes that are not running the old model, and then migrate the remaining 70% of processing tasks that have not been migrated to the nodes where the new model is deployed;
[0096] E. After the task migration is completed, deploy the new model on the node after the migration task.
[0097] As a preferred option, when implementing full lifecycle management of devices based on the HarmonyOS device management framework, the specific aspects include:
[0098] S101: Create a unique digital passport when the device is first connected. The unique digital passport contains a unique device identifier, a device type code, an X.509 format security certificate, a set of technical specifications parameters, and a geospatial location code.
[0099] S102: The health index is calculated using reliability and life indicators. The reliability index is the ratio of mean time between failures to mean time to repair, and the reliability index accounts for 60% of the weight. The life indicator is the ratio of predicted remaining life to design life, and the life indicator accounts for 40% of the weight. The predicted remaining life is calculated using a long short-term memory neural network model.
[0100] S103: Tiered maintenance trigger, specifically:
[0101] A. When the health index is less than or equal to 0.6, an emergency shutdown maintenance will be triggered within four hours.
[0102] B. When the health index is greater than 0.6 and less than or equal to 0.8, preventative maintenance will be triggered within 72 hours.
[0103] C. When the health index is greater than 0.8, a 48-hour cycle health report is generated.
[0104] Oil and gas field power metering and control devices based on the HarmonyOS system include:
[0105] The distributed sensing layer module is used to enable plug-and-play access and standardized data acquisition of electrical sensors via the HarmonyOS soft bus.
[0106] The edge collaborative computing module is used to build a distributed processing network at edge nodes, perform dynamic data compensation, real-time computing, fault detection and power allocation optimization. Specifically, it dynamically sets the node coverage standard for each processing algorithm according to its computing weight, and configures a subset of algorithms for edge nodes based on this standard. Finally, it ensures that the product of the coverage of all algorithms meets the network completeness requirements, thereby building a distributed processing network.
[0107] The secure communication transmission module is used to achieve secure data transmission between the edge and the cloud using the HarmonyOS TEE encrypted channel, ensuring end-to-end communication privacy.
[0108] The cloud-based data management module is used to perform long-term tiered storage of data (hot and cold) and differential privacy updates of models.
[0109] The closed-loop control execution module is used to send control commands generated at the edge to the terminal device and provide real-time feedback on the execution status to form a control closed loop.
[0110] The full lifecycle management module is used to generate device digital passports, calculate health indexes, and trigger graded maintenance strategies based on the HarmonyOS framework.
[0111] The beneficial effects of this invention are:
[0112] 1. Compared with existing sensor calibration schemes that use fixed thresholds or simple filtering, which have drawbacks such as inability to adapt to complex operating conditions and compensation lag, this scheme innovatively introduces an LSTM time-series prediction model to generate ideal data curves, combines a dynamic gain function to calculate the compensation amount in real time, controls the compensation intensity and smooth transition through the coordinated control of integral and exponential functions, and sets up dual stability judgment (variance and range) and anomaly freezing mechanism to achieve high-precision adaptive calibration of sensor data in the harsh environment of oil and gas fields, significantly improving the reliability of power metering.
[0113] 2. Compared with the existing edge computing networks that use homogeneous algorithm deployment, which has disadvantages such as low resource utilization and rigid task allocation, this solution constructs a differentiated algorithm set node network, dynamically allocates node coverage based on algorithm calculation weight (high-frequency algorithms have wide coverage, and low-frequency algorithms are distributed on demand), and ensures the collaborative capability of multiple algorithms through the completeness constraint of the product, so as to realize the elastic scheduling of computing resources and optimal load allocation, and significantly improve the processing efficiency of the edge network.
[0114] 3. Compared to the business interruption risk caused by the forced full-node model update of existing technologies, this solution designs a gradual update mechanism: it prioritizes the deployment of the new model and migration of some tasks on idle nodes, and after confirming stability through four verifications (data distribution, decision boundary, business indicators, and security boundary), it completes the full network update in batches; if the verification fails, it automatically rolls back and triggers the re-update process, achieving a smooth upgrade with zero business interruption while ensuring data privacy (differential privacy protection);
[0115] 4. Compared with the traditional fault detection scheme that uses a fixed threshold, which has the disadvantages of high false alarm rate and poor adaptability to operating conditions, this scheme innovatively integrates the dual parameters of ambient temperature and equipment load rate, and dynamically adjusts the detection threshold through compensation coefficient (relaxing the threshold in high temperature environment and tightening the threshold in heavy load condition), responds to changes in equipment operating status in real time, effectively distinguishes between real faults and operating condition fluctuations, and greatly reduces the false alarm and missed alarm rates.
[0116] 5. Compared to centralized power allocation strategies, which are slow to respond and struggle to balance global optimization with local fairness, this solution proposes a hierarchical optimization architecture: a cooperative game model (maximizing utility function while considering peak penalty) is used between regions to solve for the optimal quota; a Vickrey auction mechanism is introduced within each region (equipment bids based on priority coefficients, with settlement at the second-lowest price), with game theory and auction executed in parallel; field tests show that this mechanism improves inter-regional load allocation efficiency by 25% and equipment-level resource allocation fairness index by 38%.
[0117] 6. Compared to independent sensor detection, which is susceptible to single-point distortion and makes it difficult to distinguish between equipment failure and sensor malfunction, this solution constructs a feature correlation matrix based on historical fault samples and quantifies the correlation between sensor responses through conditional probability (e.g., motor overheating requires simultaneous triggering of temperature surge and current harmonic characteristics). When the feature response probability deviation exceeds 20% of the range (temperature exceeds the limit but current has no harmonics), the sensor is determined to be faulty and compensation is frozen, effectively isolating the actual equipment fault from sensor distortion, and improving fault location accuracy by 89%. Attached Figure Description
[0118] Figure 1 The diagram shown is a flowchart of the oil and gas field power metering and control method based on the HarmonyOS system of the present invention.
[0119] Figure 2 The diagram shown is a structural schematic of the oil and gas field power metering and control device based on the HarmonyOS system of the present invention. Detailed Implementation
[0120] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0121] Please see Figure 1 - Figure 2 This invention provides an embodiment of an oil and gas field power metering and control method based on the HarmonyOS system, comprising the following steps:
[0122] Step 1: Implement virtualized access and standardized data acquisition of electrical sensor devices via HarmonyOS soft bus, while ensuring the sampling frequency meets the following requirements: ;in, For adaptive adjustment coefficient, And when When activated, among which, For real-time sampling frequency, The preset minimum sampling frequency, For signal acceleration, For signal-to-noise ratio, It is the negative square of a second.
[0123] In this embodiment, the physical electrical sensors are virtualized into a unified software service object through a soft bus protocol, shielding the differences in underlying hardware and enabling plug-and-play access and a standardized real-time data acquisition process. The core innovation lies in the dynamic sampling frequency control algorithm: the system continuously monitors signal characteristics and adjusts the sampling frequency according to a preset minimum sampling frequency (…). ) and real-time calculated signal acceleration ( ) and signal-to-noise ratio ( ), combined with adaptive adjustment coefficient ( ), automatically adjust sampling frequency ( It follows the formula. When α ≥ 1, adaptive adjustment is triggered—that is, when the signal changes drastically (acceleration a increases) or the signal-to-noise ratio (SNR) increases, the system proportionally increases the sampling rate (ensuring it is not lower than 1). Conversely, the sampling rate is reduced to save resources. This mechanism significantly optimizes the computing load and communication bandwidth consumption of edge devices while ensuring the accuracy of key power data capture, making it particularly suitable for the high-efficiency power metering needs under complex operating conditions in oil and gas fields.
[0124] Step 2: Build a distributed processing network on edge computing nodes to perform data compensation, real-time computing, fault detection, power distribution optimization strategy generation, and equipment status analysis;
[0125] The constructed distributed processing network is a node network containing a differentiated algorithm set, and the construction steps are as follows:
[0126] Define a set of processing algorithms A = and its calculated weight vector Among them, weight Representation Algorithm The proportion of computational load in the data processing flow of a distributed processing network;
[0127] Configure a subset of algorithms for each edge computing node, so that the algorithms The node coverage rate reaches the set standard, which is calculated using the following formula: ;in, For the algorithm The standard value for node coverage;
[0128] The node network is deployed based on coverage constraints. Specifically, when the standard node coverage is less than or equal to 0.38, the node coverage is determined according to the specified value. Linear configuration: When the standard node coverage is greater than 0.38, the configuration is based on a node coverage of 1.
[0129] Verify that the network topology meets the completeness requirements: Where n is the total number of algorithms.
[0130] In this embodiment, a set A containing various data processing algorithms (such as filtering, feature extraction, encryption, etc.) and corresponding computational weight vectors are first defined. (weight) Representation Algorithm The proportion of computational workload in the data processing flow of a distributed processing network); secondly, according to the formula Calculate the standard node coverage value for each algorithm (ensuring wider coverage for high-frequency algorithms and distribution of low-frequency algorithms as needed); then deploy nodes according to the coverage standard—when When ≤0.38, a linear allocation strategy is adopted (e.g.) =0.3 then 30% of the nodes deploy this algorithm), when When the value is greater than 0.38, full node coverage is enforced; finally, network completeness is verified, requiring the product of the coverage rates of all algorithms to satisfy... (Ensuring multi-algorithm collaboration capabilities) thereby achieving elastic resource scheduling and optimal allocation of computing load.
[0131] When performing real-time calculations, this includes:
[0132] The input feature vector contains statistical features of the data stream, environmental parameters, and historical load sequences, and the output processing requirement vector is improved by training gradients to enhance the decision tree model.
[0133] Normalize the processing requirement vector: ;in, This is the normalized demand vector. This is the original demand vector. For the algorithm The proportion of demand;
[0134] To solve the assignment matrix, we first need to establish the optimization problem: And satisfy the following constraints: ; ;in, The proportion of task i allocated to node j. The communication latency of task i on node j, Does node j support the algorithm? , Let j be the resource utilization rate of node j. This represents the maximum total system resources.
[0135] In this embodiment, a multi-dimensional feature vector (including data stream statistical features, environmental parameters, and historical load sequences) is first input. A pre-trained Gradient Boosting Decision Tree (GBDT) model is then used to predict the processing demand ratio of each algorithm, generating an original demand vector. Subsequently, the demand vector is normalized. Forced satisfaction This ensures that each algorithm receives the minimum computational resource guarantee; finally, an allocation matrix optimization model is established to minimize the total communication latency. The objective is to simultaneously satisfy two constraints: 1. The requirement coverage constraint of each algorithm ( 1. Ensure algorithm execution capability; 2. System resource upper limit constraints ( This prevents node overload, thereby achieving adaptive matching between computing resources and task requirements.
[0136] Furthermore, when performing data compensation, the specific steps include:
[0137] Data stability detection involves real-time calculation of fluctuations in electrical sensor data, including: ; If the data variance is less than or equal to 0.05 and the data range is less than or equal to 0.1, proceed to the next step. Let N be the data variance, and N be the number of data points within the sampling window. For the i-th sample value, The mean of the sampling window, For extremely poor data, It is a data vector;
[0138] Ideal curve generation, calling a preset LSTM prediction model, inputting historical data window The sequence outputs the ideal data curve;
[0139] The compensation amount is calculated by determining the residual between the measured value and the ideal data curve, and a compensation function is dynamically generated. ;in, The compensation value at time t. This is the integral of the deviation over the past second. It is an exponential function. It is an exponential function;
[0140] Output correction involves superimposing the real-time compensation value calculated by the compensation function onto the original signal. ;in, The corrected electrical sensor value. These are the original electrical sensor values;
[0141] An abnormal freeze mechanism is implemented when the data variance of an electrical sensor is greater than 0.08 and the data range is greater than 0.15 for three consecutive cycles. In this case, compensation is paused and a device diagnostic request is generated.
[0142] In this embodiment, the data variance is first calculated in real time. ≤0.05) and range ( ≤0.1) Data stability is determined by dual indicators, and the compensation process is triggered only when both are met simultaneously; then, the LSTM time series prediction model is called to generate an ideal curve based on historical data windows; then, the measured values are calculated. Residuals from the ideal value And through dynamic gain function Generate real-time compensation values (integral term controls compensation intensity, exponential function smooths transition); superimpose the compensation values onto the original signal output correction values. Finally, abnormal freeze protection is set—when abnormal freezes are detected for 3 consecutive cycles... >0.08 or When the value is greater than 0.15, compensation is paused and diagnostics are triggered to improve accuracy while ensuring system security.
[0143] Specifically, when building a distributed processing network at edge computing nodes and generating power allocation optimization strategies, the process includes:
[0144] Physical topology partitioning: Based on the power supply line impedance parameters and controller deployment locations, the area is divided into N game units, while satisfying electrical isolation conditions. ;in, Let p be the partition number of the power grid. Let q be the partition number of the power grid. The impedance magnitude between nodes i and j;
[0145] Inter-regional cooperative game theory: using a cooperative game model to solve for inter-regional electricity quotas;
[0146] Distributed auctions are conducted within the region, and resources are allocated secondaryly to devices within the region through a distributed auction mechanism.
[0147] The specific process of inter-regional cooperative game is as follows:
[0148] S61: Define the utility function: ;in, The electricity quota allocated to region i For benefit weighting coefficients, The elastic coefficient, The historical average load for the region;
[0149] Establish an optimization problem: ; ; ;in, This is a penalty for peak system load. This is the peak adjustment factor. For the system's maximum power supply capacity, This is the lower limit of the regional quota. This is the upper limit of the regional quota;
[0150] The optimal quota vector is solved using the alternating direction method.
[0151] As a preferred option, when resources are secondary allocated through a distributed auction mechanism within the region, the specific details include:
[0152] For each game unit, the device submits a bid function: ;in, Let d be the electrical energy requirement of device d. This is the priority coefficient. Let d be the bid price for equipment. Prioritize device load. The total bid price for equipment d;
[0153] According to unit price Ascending order sorting device, among which, The price quoted per unit of electricity for device d;
[0154] Allocate as needed until satisfied. , winning bid ,in, Let i be the total electricity quota for region i. The actual winning bid volume for equipment d is given, and the remaining quota is the total regional quota minus the allocated volume.
[0155] Payment settlement price for the winning bid equipment: ;in, The final payment amount for device d. This represents the lowest unit price among the equipment that did not win the bid.
[0156] In this embodiment, the technical solution achieves intelligent power allocation through hierarchical optimization: firstly, it divides the power grid into game units based on the physical topology of the power grid, requiring the impedance ratio between regions to meet certain conditions. Ensure electrical isolation; secondly, adopt a cooperative game model between regions and define utility functions. ,by With the goal of, in and Under constraints, the optimal quota is solved using the ADMM algorithm. Finally, the Vickrey auction mechanism was adopted within the region, and the equipment was auctioned off according to... Bidding, based on unit price Assigning prices in ascending order and settling at the second-lowest price achieves a balance between global optimization and local fairness.
[0157] Specifically, fault detection includes:
[0158] Dynamic threshold setting generates real-time detection thresholds for equipment faults. These real-time detection thresholds are calculated using the following formula: ;in, As the baseline detection threshold, For real-time temperature, For reference temperature, and The first compensation coefficient is... This is the second compensation coefficient. For equipment load rate, This represents the maximum load rate of the equipment.
[0159] Correlation matrix construction, establishing a fault characteristic response matrix: ;in, This represents the conditional probability of sensor feature k occurring when fault type i occurs. Let j be the sensor feature. For fault type i;
[0160] Real-time data verification: When the sensor array detects suspected fault characteristics, it reads real-time data from the relevant sensors and calculates the theoretical characteristic quantities. And judge based on the set data anomaly conditions, which are: ;in Let j be the real-time measured quantity of sensor j. Let be the theoretical characteristic quantity of sensor j. The historical average data for sensor j, Let be the upper limit of the measurement range of sensor j;
[0161] Fault classification response: A classification response is executed based on the degree of abnormal deviation.
[0162] Specifically, when implementing a graded response based on the degree of abnormal deviation, the graded response used is as follows:
[0163] when A level 3 alarm is generated at any time;
[0164] when Reduce equipment load and generate a level 2 alarm;
[0165] when Immediately stop the machine and generate a Level 1 alarm;
[0166] in, The degree of abnormal deviation.
[0167] In this embodiment, the detection threshold is first dynamically adjusted based on the ambient temperature and equipment load rate. Real-time compensation for the impact of operating conditions; secondly, construction of a fault feature correlation matrix ( ), statistics on fault types With sensor characteristics The conditional probability relationship; when a suspected fault is detected, the theoretical value of the sensor is calculated. And through deviation judgment conditions Identify data anomalies; and finally respond in stages according to the degree of deviation—an alarm is generated when the deviation is ≤30% of the range, the load is reduced when the deviation is 30%-50%, and an emergency shutdown is performed when the deviation is >50%, thus achieving closed-loop management from anomaly detection to precise handling.
[0168] In summary, when building a distributed processing network at edge computing nodes to perform data compensation, real-time computing, fault detection, power allocation optimization strategy generation, and equipment status analysis, the overall process is as follows:
[0169] Data compensation: An ideal curve is generated through dynamic stability detection and LSTM prediction, and the real-time compensation value is calculated to correct the sensor output;
[0170] Real-time computing: Based on the GBDT model, predict the processing demand vector and solve the task allocation matrix to optimize resource scheduling;
[0171] Fault detection: Verify data anomalies by combining dynamic thresholds and feature correlation matrices, and trigger a tiered response;
[0172] Power allocation optimization: A cooperative game model is used to solve the inter-regional quotas, and a secondary allocation at the equipment level is achieved through an auction mechanism;
[0173] Equipment status analysis: Calculate health indices and generate maintenance decisions;
[0174] In the above process, fault detection, power optimization and allocation, and equipment status analysis are carried out simultaneously.
[0175] In this embodiment, the stability of sensor data is first dynamically monitored by the data compensation module, which then calls the LSTM model to generate an ideal curve and calculates real-time compensation values to correct the original signal. Next, a multi-dimensional feature vector (statistical features / environmental parameters / historical load) is input into the real-time calculation module, which predicts the processing demand vector based on the GBDT model and minimizes communication latency by optimizing the matrix allocation task. The fault detection module combines dynamic thresholds (temperature / load compensation) with the feature correlation matrix M to verify data anomalies (alarm when deviation > 0.2 times the range) and executes a three-level response (alarm / load reduction / shutdown). The power allocation module uses a cooperative game model to solve for regional quotas and then allocates equipment-level resources a second time through the Vickrey auction mechanism. Finally, the equipment status analysis module calculates the health index and triggers a graded maintenance strategy, forming an edge intelligent closed loop from data acquisition to decision execution.
[0176] Step 3: Use the HarmonyOS secure communication protocol to transmit processed data to the cloud;
[0177] Among them, quantum key distribution is integrated into the critical transmission path, and the key update cycle satisfies:
[0178] ;
[0179] in, For the key update cycle, This refers to the transmission distance.
[0180] In this embodiment, unconditionally secure key distribution is achieved through quantum state characteristics. The key update cycle is inversely proportional to the transmission distance (T = k / L, where k is a system constant). This ensures that potential eavesdropping risks are mitigated by more frequent key updates (e.g., reducing the update cycle by 50% every 10 kilometers) during long-distance transmission. At the same time, decoy state protocol and adaptive polarization compensation technology are used to eliminate channel noise interference. In actual tests, a key update every 30 seconds can be achieved at a fiber optic distance of 50 kilometers, meeting the millisecond-level encryption requirements of oil and gas field industrial control data (compliant with ISO / IEC 18045-5 standard).
[0181] Step 4: Perform long-term data backup and management model update in the cloud;
[0182] Specifically, updating the management model includes:
[0183] Update trigger: When an edge node detects that the model performance degradation exceeds a threshold, it sends an update request packet to the cloud.
[0184] Security verification: The cloud verifies the validity of the request signature and generates a new model after confirming that the decay condition is met.
[0185] For privacy processing, Laplace noise is added to the training data to achieve differential privacy protection. The privacy budget is controlled between 0.6 and 0.9, and the noise scale is inversely proportional to the maximum sensitivity of the data.
[0186] Layered transmission divides model parameters into a base layer and an adaptation layer. Base layer parameters are sent to all edge nodes in full, while adaptation layer parameters are only updated differentially to a subset of target nodes.
[0187] Progressive deployment employs a progressive deployment mechanism to update the model within edge nodes, specifically as follows:
[0188] A. Initially, select 30% of the nodes that are not running the old model to deploy the new model;
[0189] B. Migrate 30% of the processing tasks to the update node;
[0190] C. Validate the new model. If the validation is successful, proceed to the next step. If the validation fails, revert the first batch of nodes to the old model and send a new update request to the cloud.
[0191] D. Deploy the new model on the remaining 70% of nodes that are not running the old model, and then migrate the remaining 70% of processing tasks that have not been migrated to the nodes where the new model is deployed;
[0192] E. After the task migration is completed, deploy the new model on the node after the migration task.
[0193] In this embodiment, when an edge node detects that the model performance degradation exceeds a threshold (loss change rate ≥ 15% or data drift modulus ≥ 0.25), it sends a signature-verified update request to the cloud. After the cloud passes the security verification, it generates a new model and adds Laplacian noise to the training data to achieve differential privacy protection (privacy budget [0.6, 0.9], noise scale inversely proportional to data sensitivity). A layered transmission strategy is adopted—the basic layer parameters are fully distributed to all nodes, and the adaptation layer parameters are only updated to a subset of target nodes. Through a five-stage progressive deployment: the first batch of 30% of idle nodes are deployed and 30% of tasks are migrated. After verification (KL divergence ≤ 0.5 and PCA error change rate ≤ 20%), the remaining 70% of nodes are updated synchronously and the remaining tasks are migrated. If the verification fails, the version is immediately rolled back and a re-update process is triggered, achieving a smooth upgrade with zero business interruption while ensuring data privacy.
[0194] The validation methods used when validating the new model include:
[0195] Input data distribution verification: the KL divergence value of the input data of the new and old models does not exceed 0.5, and at the same time, the change rate of the principal component analysis reconstruction error in the feature space does not exceed 20%.
[0196] Consistency verification of decision boundaries: On a predefined key sample set, the average absolute deviation of the output values of the old and new models does not exceed 10% of the output standard deviation;
[0197] Business indicator constraints verification: After the update, the model meets the following requirements in the metering control chain: the power metering error does not exceed 0.5%, the fault detection false alarm rate does not exceed 2%, and the optimization strategy generation delay does not exceed 200 milliseconds.
[0198] Safety boundary verification: the difference between the Lyapunov function values of the new and old control commands does not exceed 30% of the maximum allowable value, and the condition number of the control matrix of the new model does not exceed five.
[0199] In this embodiment, the technical solution ensures the safety and reliability of model updates through four verification mechanisms: input data distribution verification requires the KL divergence of the input data of the new and old models to be ≤0.5 (controlling data drift) and the PCA reconstruction error change rate to be ≤20% (ensuring feature space stability); decision boundary consistency verification forces the average absolute deviation of the output of the new and old models to be ≤10% on key sample sets (such as the voltage drop of 90%) (suppressing logic mutations); business indicator verification is bound to industry standards—electricity metering error ≤0.5% (GB / T 17215), false alarm rate ≤2% (SY / T 6327), and policy delay ≤200ms (IEEE 1646); safety boundary verification uses Lyapunov functions to evaluate the energy change of control commands and constrains the number of control matrix conditions to be ≤5 (IEC 61508 SIL2 stability requirement), forming a comprehensive verification system covering data, logic, performance, and security.
[0200] Step 5: Send the control commands generated at the edge to the terminal device for execution and report the results;
[0201] Step Six: Implement full lifecycle management of devices based on the HarmonyOS device management framework.
[0202] When implementing full lifecycle management of devices based on the HarmonyOS device management framework, the specific aspects include:
[0203] A unique digital passport is created when the device is first connected. The unique digital passport contains a unique device identifier, a device type code, an X.509 format security certificate, a set of technical specifications parameters, and a geospatial location code.
[0204] The health index is calculated using reliability and life indicators. The reliability index is the ratio of mean time between failures to mean time to repair, and the reliability index accounts for 60% of the weight. The life indicator is the ratio of predicted remaining life to design life, and the life indicator accounts for 40% of the weight. The predicted remaining life is calculated using a long short-term memory neural network model.
[0205] Hierarchical maintenance is triggered, specifically as follows:
[0206] A. When the health index is less than or equal to 0.6, emergency shutdown maintenance within 4 hours is triggered;
[0207] B. When the health index is greater than 0.6 and less than or equal to 0.8, preventive maintenance within 72 hours is triggered;
[0208] C. When the health index is greater than 0.8, a 48-hour periodic health report is generated.
[0209] In this embodiment, a unique digital passport (including the device unique identifier, X.509 security certificate, technical specification parameters, and a 0.1-meter precision geocoding) is created when the device is first connected; the health index HI is dynamically calculated through a two-dimensional health index assessment - the reliability index (MTBF / MTTR accounts for 60%) and the lifespan index (LSTM predicted remaining lifespan / design lifespan accounts for 40%); hierarchical maintenance is triggered according to the HI value: emergency shutdown maintenance within 4 hours when HI≤0.6, preventive maintenance within 72 hours when 0.6<HI≤0.8, and a 48-hour health report (including vibration spectrum and thermal imaging analysis) is generated when HI>0.8, realizing digital closed-loop management from device network access to retirement.
[0210] The oil and gas field electric energy metering control device based on the HarmonyOS system includes:
[0211] The distributed sensing layer module is used to achieve the plug-and-play access of electrical sensors and standardized data acquisition through the HarmonyOS soft bus;
[0212] The edge collaborative computing module is used to build a distributed processing network at the edge node, perform data dynamic compensation, real-time calculation, fault detection, and power distribution optimization. Specifically, for each processing algorithm, the node coverage standard is dynamically set according to its calculation weight, and an algorithm subset is differentially configured for the edge node based on this standard, ultimately ensuring that the product of all algorithm coverages meets the network completeness requirements, thereby building a distributed processing network;
[0213] The secure communication transmission module is used to achieve secure data transmission between the edge and the cloud through the HarmonyOS TEE encryption channel, ensuring end-to-end communication privacy;
[0214] The cloud data management module is used to perform long-term data cold and hot hierarchical storage and model differential privacy update;
[0215] The closed-loop control execution module is used to send the control instructions generated at the edge to the terminal device and real-time feedback the execution status to form a control closed-loop;
[0216] The full lifecycle management module is used to generate device digital passports, calculate health indexes, and trigger graded maintenance strategies based on the HarmonyOS framework.
[0217] Specifically, the edge collaborative computing module includes:
[0218] The dynamic compensation core module is used to generate ideal data curves through LSTM, calculate real-time compensation amounts, and automatically correct sensor outputs when data is stable.
[0219] The node network scheduling module is used to dynamically allocate node coverage according to algorithm weights and optimize the task allocation matrix to minimize communication latency.
[0220] The fault diagnosis engine module is used to combine dynamic thresholds and feature correlation matrices to achieve multi-level fault determination;
[0221] The game optimization decision module is used to allocate regional power quotas using a cooperative game model and to achieve secondary allocation at the equipment level through Vickrey auctions.
[0222] Example 1: Power Control System for Offshore Drilling Platforms
[0223] Scenario: This solution is deployed on an offshore drilling platform, connecting over 200 electrical sensors (voltage, current, vibration, etc.) via HarmonyOS soft bus. The dynamic sampling frequency automatically increases to 250Hz to capture transient current when the motor starts, and decreases to 10Hz in steady state to save bandwidth. Twelve heterogeneous computing nodes are constructed at the edge, including:
[0224] The data compensation module detected that the variance of the vibration sensor data of a certain pump was 0.03 (≤0.05) and the range was 0.08 (≤0.1), which triggered the LSTM model to generate an ideal curve. The measured value of 9.2 mm / s was compensated by 1.5 mm / s, and the output correction value was 7.7 mm / s (close to the actual vibration value).
[0225] Based on characteristics such as ambient temperature of 45℃ and load rate of 85%, the real-time computing module uses GBDT to predict which filtering algorithm should be executed first (with a weight of 35%), and task scheduling reduces communication latency from 120ms to 75ms.
[0226] The fault detection module dynamically adjusts the vibration threshold to 6.8 mm / s (the original threshold was 8 mm / s, compensated for due to high temperature and heavy load). When a motor bearing failure triggers a vibration value of 9.7 mm / s, the correlation matrix verifies that the temperature sensor is not synchronously abnormal (theoretically it should be 68℃, but the actual measured temperature is 52℃). The sensor is determined to be distorted and is frozen for compensation.
[0227] The power distribution module divides the platform into three game units: a cooperative game to allocate 50MW to the drilling area and 30MW to the living area. Within the area, power supply to the drilling pumps is prioritized through Vickrey auctions (load priority 0.95, winning bid price 1.25 yuan / MW).
[0228] When updating the model, the first four nodes deployed the new fault detection model. After verifying that the KL divergence was 0.42 (<0.5), the model was launched on the entire network, and the false alarm rate dropped from 35% to 2%.
[0229] Full lifecycle management generates digital passports for critical equipment. When the health index of a compressor reaches 0.65, preventive maintenance is triggered within 72 hours to avoid unplanned downtime.
[0230] Example 2: Smart Grid Application in Desert Oilfields
[0231] Scenario: A desert oilfield uses this solution to manage the electricity metering of 50 oil wells.
[0232] Dynamic sampling is based on signal abrupt changes caused by sandstorms (acceleration 0.15). The sampling rate of the RTU cabinet temperature sensor was increased from 20Hz to 100Hz.
[0233] The node network is deployed according to algorithm weights: the high-frequency filtering algorithm covers 80% of the nodes, the low-frequency encryption algorithm covers 30% of the nodes, and the product completeness is 0.53 > 0.32 (n=5).
[0234] In data compensation, LSTM predicts the ideal current curve of a well electric pump, and compensates for a drift value of 105A by 8A to output 97A (rated value 100A). An abnormal freezing mechanism prevents erroneous compensation when the voltage drops sharply.
[0235] The dynamic threshold for fault detection is automatically relaxed by 18% due to a surface temperature difference of 40℃. A real overheating fault (temperature 92℃) of a transformer triggers a first-level alarm, and the correlation matrix synchronously verifies that the current harmonics exceed the standard.
[0236] The power allocation divides the oilfield into 5 game units, with the central processing plant allocated a 60MW quota (utility function weights). =0.9), the well site equipment was allocated a second time through an auction mechanism, with all high-priority electric pumps winning the bid;
[0237] The first batch of 30% of nodes will be gradually updated to deploy the new energy consumption model. After business verification shows that the metering error is 0.3% (<0.5%), the entire network will be updated.
[0238] Quantum encryption enables key updates every 30 seconds on a 50km backbone network, mitigating the risk of eavesdropping in uninhabited desert areas.
[0239] The equipment health index is based on LSTM to predict the remaining lifespan. A certain aging oil pumping unit with a health index of 0.55 triggered 4 hours of emergency maintenance to prevent the failure from escalating.
[0240] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A method for metering and controlling electricity in oil and gas fields based on the HarmonyOS system, characterized in that: Includes the following steps: S11: Virtualize the access of electrical sensor devices and standardize data acquisition through HarmonyOS soft bus; S12: Build a distributed processing network on edge computing nodes to perform data compensation, real-time computing, fault detection, power allocation optimization strategy generation, and equipment status analysis; S13: Use the HarmonyOS secure communication protocol to transmit processed data to the cloud; S14: Perform long-term data backup and management model updates in the cloud; S15: Send the control commands generated at the edge to the terminal device for execution and return the results; S16: Implement full lifecycle management of devices based on the HarmonyOS device management framework; Specifically, when constructing a distributed processing network at edge computing nodes, the constructed distributed processing network is a node network containing a differentiated algorithm set, which includes the following steps: S31: Define the set of processing algorithms A= and its calculated weight vector Among them, weight Representation Algorithm The proportion of computational workload in the data processing flow of a distributed processing network; S32: Configure a subset of algorithms for each edge computing node, so that the algorithm The node coverage rate reaches the set standard, which is calculated using the following formula: ; in, For the algorithm The standard value for node coverage; S33: Deploy node networks based on coverage constraints; S34: The network topology satisfies the completeness condition: ; Where n is the total number of algorithms, For the algorithm The standard value for node coverage.
2. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 1, characterized in that: When performing standardized data collection, the sampling frequency must satisfy: ; in, For adaptive adjustment coefficient, And when When activated, among which, For real-time sampling frequency, The preset minimum sampling frequency, For signal acceleration, For signal-to-noise ratio, It is the negative square of a second.
3. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 2, characterized in that: When building a distributed processing network at edge computing nodes to perform data compensation, the specific steps include: S21: Data stability detection, real-time calculation of fluctuation measures of electrical sensor data, and proceed to the next step when the data variance is less than or equal to 0.05 and the data range is less than or equal to 0.1; S22: Ideal curve generation, calling the preset LSTM prediction model, inputting historical data window. The sequence outputs the ideal data curve; S23: Compensation calculation: Calculate the residual between the measured value and the ideal data curve, and dynamically generate the compensation function. ; in, The compensation value at time t. This is the integral of the deviation over the past second. It is an exponential function. For deviation; S24: Output correction, which superimposes the real-time compensation value calculated by the compensation function onto the original signal; S25: Abnormal freeze mechanism. When the data variance of the electrical sensor is detected to be greater than 0.08 and the data range is greater than 0.15 for three consecutive cycles, compensation is suspended and a device diagnostic request is generated.
4. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 3, characterized in that: When building a distributed processing network at edge computing nodes to perform real-time computing, it includes: S41: Input feature vector, which contains statistical features of the data stream, environmental parameters and historical load sequences, and improves the output processing demand vector of the decision tree model by training gradients; S42: Normalize the processing requirement vector; S43: Solving the assignment matrix, first establish the optimization problem: ; And satisfy the constraints: ; ; in, The proportion of task i allocated to node j. The communication latency of task i on node j, Does node j support the algorithm? , Let j be the resource utilization rate of node j. This represents the maximum total system resources. For the algorithm The proportion of demand.
5. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 4, characterized in that: When building a distributed processing network at edge computing nodes and generating power allocation optimization strategies, the specific steps include: S51: Physical topology partitioning. Based on the power supply line impedance parameters and controller deployment location, the area is divided into N game units, and electrical isolation conditions are met. S52: Inter-regional cooperative game, using a cooperative game model to solve for inter-regional electricity quotas; S53: Distributed auction within the region, where resources are allocated to devices within the region through a distributed auction mechanism.
6. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 5, characterized in that: When using a cooperative game theory model to solve for inter-regional electricity quotas, the specific steps include: S61: Define the utility function: ; in, The electricity quota allocated to region i For benefit weighting coefficients, The elastic coefficient, The historical average load for the region, This is the penalty coefficient; S62: Establish an optimization problem: ; in, This is a penalty for peak system load. This is the peak adjustment factor. For the system's maximum power supply capacity, This is the lower limit of the regional quota. This is the upper limit of the regional quota; S63: The optimal quota vector is solved using the alternating direction method.
7. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 6, characterized in that: When building a distributed processing network at edge computing nodes and performing fault detection, the specific steps include: S81: Dynamic threshold setting, generating real-time detection thresholds for device faults, wherein the real-time detection thresholds are calculated using the following formula: ; in, As the baseline detection threshold, For real-time temperature, For reference temperature, and The first compensation coefficient, This is the second compensation coefficient. For equipment load rate, This represents the maximum load rate of the equipment. S82: Constructing the correlation matrix and establishing the fault characteristic response matrix; S83: Real-time data verification. When the sensor group detects suspected fault characteristics, it reads the real-time data of the relevant sensors, calculates the theoretical characteristic quantity, and makes a judgment based on the set data anomaly conditions. S84: Fault classification response, executes a classification response based on the degree of abnormal deviation.
8. The oil and gas field power metering and control method based on the HarmonyOS system according to claim 7, characterized in that: When managing model updates in the cloud, the specific steps include: S91: Update triggered. When an edge node detects that the model performance degradation exceeds a threshold, it sends an update request packet to the cloud. S92: Security verification, cloud verification of request signature validity, and generation of a new model after confirming that the decay condition is met; S93: Privacy processing: Laplace noise is added to the training data to achieve differential privacy protection. The privacy budget is controlled between 0.6 and 0.
9. The noise scale is inversely proportional to the maximum sensitivity of the data. S94: Layered transmission, which divides model parameters into a base layer and an adaptation layer. The base layer parameters are sent to all edge nodes in full, while the adaptation layer parameters are only updated to a subset of target nodes in a differentiated manner. S95: Progressive deployment, employing a progressive deployment mechanism to update the model within edge nodes, specifically: A. Initially, select 30% of the nodes that are not running the old model to deploy the new model; B. Migrate 30% of the processing tasks to the update node; C. Validate the new model. If the validation is successful, proceed to the next step. If the validation fails, revert the first batch of nodes to the old model and send a new update request to the cloud. D. Deploy the new model on the remaining 70% of nodes that are not running the old model, and then migrate the remaining 70% of processing tasks that have not been migrated to the nodes where the new model is deployed; E. After the task migration is completed, deploy the new model on the node after the migration task.
9. An oil and gas field power metering and control device based on the HarmonyOS system, characterized in that: include: The distributed sensing layer module is used to enable plug-and-play access and standardized data acquisition of electrical sensors via the HarmonyOS soft bus. The edge collaborative computing module is used to build a distributed processing network at edge nodes, perform dynamic data compensation, real-time computing, fault detection and power allocation optimization. Specifically, it dynamically sets the node coverage standard for each processing algorithm according to its computing weight, and configures a subset of algorithms for edge nodes based on this standard. Finally, it ensures that the product of the coverage of all algorithms meets the network completeness requirements, thereby building a distributed processing network. The secure communication transmission module is used to achieve secure data transmission between the edge and the cloud using the HarmonyOS TEE encrypted channel, ensuring end-to-end communication privacy. The cloud-based data management module is used to perform long-term tiered storage of data (hot and cold) and differential privacy updates of models. The closed-loop control execution module is used to send control commands generated at the edge to the terminal device and provide real-time feedback on the execution status to form a control closed loop. The full lifecycle management module is used to generate device digital passports, calculate health indexes, and trigger graded maintenance strategies based on the HarmonyOS framework.
Citation Information
Patent Citations
Electric energy load real-time data acquisition and analysis method based on low-cost scheme
CN120237632A
Real-time metering method and system of intelligent modular electric energy metering box
CN120611337A