Task allocation method of intelligent terminal based on AI edge decision and related device

Abstract

The invention provides an intelligent terminal task allocation method based on AI edge decision and a related device, and the method comprises the steps: obtaining environment data and task attribute data in a target power operation region, obtaining target environment data and m pieces of task attribute data, transmitting the target environment data to n intelligent terminals, calculating according to the target environment data through n intelligent terminals to obtain n pieces of first calculation performance data, returning the n pieces of first calculation performance data to the electronic equipment, and determining a first adaptation degree matrix of tasks and intelligent terminal performance according to the n pieces of first calculation performance data and the m pieces of task attribute data, determining a target task allocation strategy according to a preset constraint condition and the first adaptation degree matrix; and controlling the n intelligent terminals to execute the edge decision according to the target task allocation strategy. In an electric power operation scene, the accuracy and robustness of task allocation for distributed terminal decision making can be improved.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a task allocation method and related devices for intelligent terminals based on AI edge decision-making. Background Art

[0002] In power operations, distributed intelligent terminals must leverage AI edge technology to optimally allocate computing tasks across different systems based on environmental factors. However, the diversity and complexity of the environment impacts this optimal allocation. Power operation environments are subject to numerous interference factors, such as electromagnetic interference and temperature and humidity fluctuations. These factors can reduce the accuracy of intelligent terminal allocation decisions. Furthermore, constraints on the intelligent terminals' computing resources, such as computing power and storage capacity, can lead to instability in their decision-making.

[0003] Therefore, how to improve the accuracy and robustness of task allocation in distributed terminal decision-making in power operations needs to be solved urgently. Summary of the Invention

[0004] The embodiments of the present application provide a task allocation method and related devices for intelligent terminals based on AI edge decision-making, which improve the accuracy and robustness of task allocation of distributed terminal decisions in the field of power operations.

[0005] In a first aspect, an embodiment of the present application provides a task allocation method for an intelligent terminal based on AI edge decision-making, which is applied to an electronic device connected to n intelligent terminals in a target power operation area, where n is an integer greater than 1. The method includes:

[0006] Acquire environmental data in the target power operation area and task attribute data of the power operation task to obtain target environmental data and m task attribute data; the power operation task includes m operation tasks, where m is an integer greater than 1;

[0007] Sending the target environment data to the n smart terminals respectively, and having the n smart terminals perform calculations based on the target environment data respectively to obtain n first computing performance data, and returning the n first computing performance data to the electronic device;

[0008] Determining a first fitness matrix between the task and the smart terminal performance according to the n first computing performance data and the m task attribute data;

[0009] Determining a target task allocation strategy based on preset constraints and the first fitness matrix;

[0010] Control the n intelligent terminals to execute edge decision-making according to the target task allocation strategy.

[0011] In a second aspect, an embodiment of the present application provides a task allocation device for an intelligent terminal based on AI edge decision-making, which is applied to an electronic device connected to n intelligent terminals in a target power operation area, where n is an integer greater than 1. The device includes:

[0012] an acquisition module, configured to acquire environmental data in the target power operation area and task attribute data of the power operation task, to obtain target environmental data and m task attribute data; the power operation task includes m operation tasks, where m is an integer greater than 1;

[0013] a calculation module, configured to send the target environment data to the n smart terminals respectively, and have the n smart terminals perform calculations based on the target environment data respectively to obtain n first calculation performance data, and return the n first calculation performance data to the electronic device;

[0014] a determination module, configured to determine a first fitness matrix between tasks and intelligent terminal performance based on the n first computing performance data and the m task attribute data; and determine a target task allocation strategy based on preset constraints and the first fitness matrix;

[0015] A control module is used to control the n intelligent terminals to execute edge decision-making according to the target task allocation strategy.

[0016] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program comprises instructions for executing the steps of any method of the first aspect of the embodiment of the present application.

[0017] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the above-mentioned computer-readable storage medium stores a computer program for electronic data exchange, wherein the above-mentioned computer program enables a computer to execute part or all of the steps described in any method of the first aspect of the embodiment of the present application.

[0018] In a fifth aspect, embodiments of the present application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute some or all of the steps described in any method of the first aspect of the embodiments of the present application. The computer program product may be a software installation package.

[0019] By implementing the embodiments of the present application, the following beneficial effects are achieved:

[0020] The present application describes a task allocation method and related device for smart terminals based on AI edge decision-making, which are applied to electronic devices, wherein the electronic devices are connected to n smart terminals in a target power operation area, where n is an integer greater than 1. The method first obtains environmental data in the target power operation area and task attribute data of the power operation task, and obtains target environmental data and m task attribute data, wherein the power operation task includes m operation tasks, and m is an integer greater than 1; then, the target environmental data are sent to the n smart terminals respectively, and the n smart terminals perform calculations based on the target environmental data respectively to obtain n first computing performance data, and return the n first computing performance data to the electronic device; then, the first fitness matrix of the task and the smart terminal performance is determined based on the n first computing performance data and the m task attribute data, and the target task allocation strategy is determined based on the preset constraints and the first fitness matrix; finally, the n smart terminals are controlled to execute edge decision-making according to the target task allocation strategy. In this way, on the one hand, by incorporating the environmental factors of power operations into the calculation of the model, the perception ability and computing performance of the intelligent terminal are improved, thereby improving the accuracy of the distributed terminal decision-making tasks; on the other hand, by adaptively adjusting the fitness matrix in real time according to the impact of environmental factors on the computing performance of the intelligent terminal, the task allocation strategy is dynamically adjusted, thereby improving the robustness of the distributed terminal decision-making tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 This is an architectural diagram of a task allocation system for intelligent terminals based on AI edge decision-making provided by an embodiment of the present application;

[0023] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0024] Figure 3 This is a flowchart of a task allocation method for an intelligent terminal based on AI edge decision-making provided by an embodiment of the present application;

[0025] Figure 4 This is a flowchart of another method for allocating tasks to smart terminals based on AI edge decision-making provided by an embodiment of the present application;

[0026] Figure 5This is an architectural diagram of a task allocation method for an intelligent terminal based on AI edge decision-making provided in an embodiment of the present application;

[0027] Figure 6 This is an architectural diagram of a task allocation model for an intelligent terminal based on AI edge decision-making provided in an embodiment of the present application;

[0028] Figure 7 This is a schematic diagram of the structure of a smart terminal provided in an embodiment of the present application;

[0029] Figure 8 This is a block diagram of the functional modules of a task allocation device for an intelligent terminal based on AI edge decision-making provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] The terms "first," "second," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0032] It should be understood that the term "and / or" in this document is merely a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document indicates that the associated objects are in an "or" relationship. The "plurality" appearing in the embodiments of this application refers to two or more.

[0033] In the embodiments of the present application, "at least one item" or similar expressions refers to any combination of these items, including any combination of single items or plural items, and refers to one or more, and multiple refers to two or more. For example, at least one item (item) of a, b, or c can represent the following seven situations: a, b, c, a and b, a and c, b and c, a, b, and c. Among them, each of a, b, and c can be an element or a set containing one or more elements.

[0034] The "connection" appearing in the embodiments of the present application refers to various connection methods such as direct connection or indirect connection to achieve communication between devices, and the embodiments of the present application do not impose any limitations on this.

[0035] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0036] During power operations, distributed intelligent terminals need to utilize AI edge technology to optimally allocate computing tasks to different systems based on environmental factors. However, the diversity and complexity of these environmental factors present technical challenges to this process. First, various interference factors exist in the power operation environment, such as electromagnetic interference, temperature and humidity fluctuations, and dust pollution. These factors can affect the perception and computing performance of intelligent terminals, resulting in reduced accuracy in decision allocation. Second, the dynamic changes in the power operation environment also pose challenges to decision allocation. How to adapt to environmental changes in real time, dynamically adjust task allocation strategies, and ensure system robustness is a technical problem that needs to be solved urgently.

[0037] To address the above-mentioned issues, an embodiment of the present application provides a task allocation method and related device for intelligent terminals based on AI edge decision-making, which are applied to electronic devices. The electronic device connects n intelligent terminals in a target power operation area, where n is an integer greater than 1. By acquiring environmental data in the target power operation area and task attribute data of the power operation task, target environmental data and m task attribute data are obtained, where the power operation task includes m operation tasks, where m is an integer greater than 1. The target environmental data is sent to the n intelligent terminals respectively, and the n intelligent terminals perform calculations based on the target environmental data to obtain n first computing performance data, which are then returned to the electronic device. A first fitness matrix between the task and the intelligent terminal performance is determined based on the n first computing performance data and the m task attribute data. Then, a target task allocation strategy is determined based on preset constraints and the first fitness matrix. Finally, the n intelligent terminals are controlled to execute edge decisions based on the target task allocation strategy. In power operations, the accuracy and robustness of task allocation in distributed terminal decisions are improved.

[0038] The following combination Figure 1 The architecture of a task allocation system for intelligent terminals based on AI edge decision-making in an embodiment of the present application is described. Figure 1 This is an architectural diagram of a task allocation system for smart terminals based on AI edge decision-making provided in an embodiment of the present application. The task allocation system 100 for smart terminals based on AI edge decision-making includes: an electronic device 110 and an electric power operation area 120.

[0039] Among them, the electronic device 110 includes one or more smart terminals 111. The electronic device 110 is used as a carrier and management carrier of the smart terminal to build a decision-making and control center for task allocation. It coordinates and dispatches power operation tasks and smart terminals based on the built-in edge computing framework, AI algorithm model and communication module; one or more smart terminals 111 serve as task execution entities, equipped with a hardware platform with edge computing capabilities and special sensors for power operations (such as temperature and humidity sensors, electrical parameter acquisition modules), receive and execute the operation tasks issued by the electronic device 110, collect power equipment operation data and execution status feedback in real time, and have local lightweight data processing and preliminary decision-making capabilities to achieve efficient response to edge-side tasks.

[0040] Among them, the power operation area 120 includes one or more operation tasks 121. The power operation area 120 is used to define the physical space scope of the power operation and maintenance business, including scenes such as transmission line corridors, substation areas, and distribution areas. Its environmental factors and other characteristics constitute the environmental constraints for task execution, and provide scenario-based input for the electronic device 110 to formulate task allocation strategies; one or more operation tasks 121 are used to represent the specific business needs of power operation and maintenance, including equipment status inspections, fault emergency response, routine data collection and other types. Each task carries a real-time threshold (for example, attribute tags such as task success rate and resource requirements).

[0041] In one possible embodiment, the power operation area 120 maps the state changes of the power operation area 120 (such as equipment load fluctuations, abnormal ambient temperature and humidity) in real time based on the power equipment data and environmental data. The generation and triggering of one or more operation tasks 121 supports both manual work order entry and automatic identification of task requirements through the intelligent perception system. The task attribute labels are dynamically assigned by the power operation and maintenance knowledge base, for example, based on equipment type and historical failure mode. During the task execution process, the environmental data of the power operation area 120 is synchronized to the electronic device 110 in real time, and is used to dynamically adjust the task execution parameters of the smart terminal 111 (such as communication frequency band switching, sampling frequency optimization, etc.). Furthermore, under the scheduling of the electronic device 110, the smart terminal 111 can flexibly adapt the execution mode according to the task attributes: in the face of high-real-time relay protection tasks, the hard real-time scheduling algorithm is enabled to seize the terminal computing power resources to ensure the task deadline; for routine inspection tasks, a dynamic resource allocation strategy is adopted to time-share the terminal hardware with other tasks to improve resource utilization. The scenario model of the power operation area 120 works in deep coordination with the AI ​​decision-making model of the electronic device 110. When electromagnetic interference in the area intensifies, the electronic device 110 automatically increases the task allocation weight of the anti-interference communication protocol and prioritizes assigning tasks to smart terminals 111 that support low-power wide-area communications such as NB-IoT to ensure the reliability of task data transmission.

[0042] It can be seen that the electronic device 110 integrates terminal capabilities and task requirements to drive the intelligent terminal 111 to accurately perform operations; the power operation area 120, as the execution entity, carries task requirements and feedbacks environmental constraints, and through the interaction of task allocation strategies, improves the robustness and accuracy of distributed terminal task allocation in the power operation scenario.

[0043] The following combination Figure 2 The electronic device in the embodiment of the present application is described. Figure 2 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown in FIG. Figure 2As shown, the electronic device 200 includes one or more processors 210, a memory 220, a communication interface 230 and one or more programs 221. The processor 210 is communicatively connected to the memory 220 and the communication interface 230 via an internal communication bus.

[0044] Among them, the one or more programs 221 are stored in the above-mentioned memory 220 and are configured to be executed by the above-mentioned processor 210. The one or more programs include instructions for executing any step in the embodiment of the following task allocation method for intelligent terminals based on AI edge decision-making.

[0045] Among them, the processor 210 can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute the various exemplary logic blocks, units and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The communication unit can be a communication interface, a transceiver, a transceiver circuit, etc., and the storage unit can be a memory.

[0046] The memory 220 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM).

[0047] It is understandable that the electronic device 200 may include more or fewer structural elements than those in the above structural block diagram, for example, a power module, physical buttons, a Wi-Fi module, a speaker, a Bluetooth module, a sensor, a display module, etc., which are not limited here.

[0048] After understanding the software and hardware architecture of this application, Figure 3 A task allocation method for an intelligent terminal based on AI edge decision-making in an embodiment of the present application is described. Figure 3 This is a flow chart of a method for allocating tasks to smart terminals based on AI edge decision-making provided by an embodiment of the present application. The method is applied to an electronic device connected to n smart terminals in a target power operation area, where n is an integer greater than 1. The method for allocating tasks to smart terminals based on AI edge decision-making specifically includes the following steps:

[0049] Step S310 , acquiring environmental data in the target power operation area and task attribute data of the power operation task, to obtain target environmental data and m task attribute data; the power operation task includes m operation tasks, where m is an integer greater than 1.

[0050] Target environment data is a collection of multidimensional data representing the state of the power operation environment. By deploying multiple sensors in smart terminals, key influencing factors such as electromagnetic interference, temperature and humidity changes, and dust pollution in the power operation environment are collected in real time. These factors are not limited here. Electromagnetic interference data reflects the stability of the electromagnetic environment during power equipment operation. Temperature and humidity data measures the impact of ambient temperature and humidity on smart terminal performance. Dust pollution data reflects the cleanliness level of the operating environment. Task attribute data is a detailed description of the task, including the task category, priority level, and required computing resources. Task attribute data focuses on the core requirements of power operation tasks: real-time, reliability, and security. Real-time requires a time threshold for task completion. Reliability indicators represent the success rate standard for task execution. Security parameters involve security constraints such as data encryption level and access control policy.

[0051] The frequency of environmental data collection is related to the risk level of the operation scenario. If the target power operation area is in the early stage of equipment commissioning, the high-load operation stage, or the period of drastic seasonal climate change (such as hot and humid summer, windy and sandy spring), a high-frequency collection strategy is adopted to capture subtle fluctuations in environmental parameters in real time at intervals of minutes to avoid abnormal task execution caused by sudden environmental changes that lead to a sudden drop in the performance of smart terminals. For routine inspection scenarios or areas with stable historical environmental conditions, the collection interval can be extended to the hourly or daily level. While ensuring the effectiveness of environmental monitoring, it also reduces data processing pressure and communication bandwidth consumption. The extraction of task attribute data needs to be combined with the business characteristics of the power system. For example, relay protection tasks have extremely high real-time requirements, while equipment status analysis tasks focus more on reliability. The attribute dimension weights of different types of tasks need to be dynamically adjusted according to business priorities.

[0052] Specifically, environmental data collection is achieved through the sensor array built into the smart terminal: the electromagnetic sensor uses a Hall-effect sensor, which can detect changes in magnetic field strength within the range of 1mT to 100mT, meeting the needs of electromagnetic interference monitoring around high-voltage equipment in substations. The temperature and humidity sensor uses an SHT31 digital sensor with a temperature measurement range of -40°C to 125°C and a humidity measurement accuracy of ±2% RH, ensuring the accuracy of ambient temperature and humidity data. The dust sensor uses the Sharp GP2Y1010AU0F, which can detect the concentration of dust particles with a size of ≥1μm and is suitable for dust monitoring in indoor and outdoor power operation environments. The raw environmental data collected by these sensors is transmitted to the nearest edge computing node via an RS485 bus at a rate of 9600bps, with a transmission distance of up to 2km. The edge node is equipped with an Intel NUC minicomputer running the Ubuntu 20.04 operating system to preprocess the data. During data preprocessing, a sliding average filter algorithm is first used to remove random noise. Maximum and minimum normalization is then used to map the data to the [0, 1] interval. Finally, feature extraction is performed for different data types. For example, a wavelet transform is performed on electromagnetic data to obtain frequency domain features. The mean, variance, and rate of change within a 5-minute sliding window are calculated for temperature and humidity data. Minute-by-minute peak and cumulative particle concentrations are calculated for dust data. Finally, a Kalman filter fusion algorithm is used to generate a 20-dimensional comprehensive feature vector (with a weight of 0.5 for electromagnetic features, 0.3 for temperature and humidity features, and 0.2 for dust features) to obtain environmental data. Task attribute data extraction relies on the power system task management platform, which automatically identifies key parameters by parsing task description files. Real-time performance measures the maximum allowable delay from task triggering to result output, while reliability measures the expected success rate by calculating historical task execution records.

[0053] Step S320: Send the target environment data to the n smart terminals respectively, and use the n smart terminals to perform calculations based on the target environment data to obtain n first computing performance data, and return the n first computing performance data to the electronic device.

[0054] Among them, the first computing performance data is a set of quantitative indicators that represent the perception and computing capabilities of the intelligent terminal under current environmental conditions, including CPU frequency fluctuation coefficient, memory bandwidth utilization, and storage read and write rate attenuation rate parameters. There is no limitation here. This parameter converts the environmental feature vector into a predicted value of terminal performance through a mapping model, which can intuitively reflect the degree of impact of environmental factors such as electromagnetic interference and temperature and humidity changes on the terminal's computing efficiency. For example, when the ambient temperature and humidity exceed the rated operating range of the terminal, the CPU frequency may automatically reduce to avoid overheating. At this time, the frequency fluctuation coefficient in the first computing performance data will be reflected as a negative value, indicating performance attenuation; and when the degree of dust pollution is high, the read and write error rate of the storage device increases, and the corresponding storage rate attenuation rate indicator increases.

[0055] The intelligent terminal's calculation of target environmental data relies on a pre-established environment-performance mapping model, built on a machine learning algorithm. This model achieves dynamic predictions through training with historical environmental characteristics and terminal performance data. Specifically, the model employs a three-layer neural network structure, uses the Adam optimizer and a mean square error loss function, and was trained on 1,000 sets of historical data (covering scenarios such as -40°C to 125°C temperature, 0-100% RH humidity, and 0-100mT electromagnetic interference) until convergence, ultimately achieving a 92% performance prediction accuracy on the test set.

[0056] Specifically, after receiving the target environment data, the smart terminal first normalizes the data, mapping the eigenvalues ​​of each dimension to the range [0, 1] to avoid model prediction bias due to dimensional differences. The normalized eigenvectors are then input into the environment-performance mapping model, where predicted values ​​for each performance parameter are calculated through forward propagation. After completing the performance prediction, the smart terminal verifies the predicted results for service adaptability. If the perception capability index is lower than 0.8 or the computational latency coefficient is greater than 1.5, a performance optimization mechanism is triggered. This optimization mechanism improves adaptability by dynamically adjusting terminal configuration parameters. For example, it can reduce the sensor sampling frequency from 100Hz to 60Hz to reduce computational load or disable real-time data synchronization for non-critical tasks to free up memory resources. After optimization is complete, first computational performance data is generated, including the original predicted values ​​and the optimized parameters. This data is then returned to the electronic device (edge ​​computing node) via a 5G wireless communication link (transmission rate ≥ 100Mbps), providing data support for the subsequent construction of the adaptability matrix.

[0057] Step S330 : determining a first fitness matrix between the task and the smart terminal performance according to the n first computing performance data and the m task attribute data.

[0058] The first adaptability matrix is ​​an n×m-dimensional correlation matrix representing the degree of match between task requirements and intelligent terminal performance. Each element in the matrix reflects the relevance of the execution adaptation of a specific task on the corresponding terminal. The construction of this matrix simultaneously considers real-time performance, reliability, and security. Task attribute data must extract key parameters from the dimensions of real-time performance, reliability, and security to form m task attribute vectors. The first computational performance data is converted into n terminal performance vectors. The correlation between task attributes and performance is then calculated. For example, cosine similarity, semantic similarity, or Gaussian similarity can be used, but these are not limited here.

[0059] In a possible embodiment, determining a first fitness matrix between the task and the smart terminal performance according to the n first computing performance data and the m task attribute data specifically includes the following steps:

[0060] 331. Perform feature extraction on the n first computing performance data to obtain n first performance vectors;

[0061] 332. Perform feature extraction on the m task attribute data to obtain m task attribute vectors;

[0062] 333. Determine a fit score between the first performance and the task attribute based on the n first performance vectors and the m task attribute vectors, to obtain n*m fit scores;

[0063] 334. Determine a fitness matrix according to the n*m ​​fitness scores;

[0064] 335. Determine a weight coefficient corresponding to the fitness matrix according to a mapping relationship between preset environmental factors and fitness matrix weights to obtain a first weight coefficient;

[0065] 336. Adjust the fitness matrix according to the first weight coefficient to obtain the first fitness matrix.

[0066] The first performance vector is a set of multidimensional performance indicators that characterize the intelligent terminal under environmental influences. It includes parameters such as CPU frequency decay rate, memory bandwidth utilization, storage read and write latency, and sensor sampling accuracy degradation coefficient. These parameters are standardized to eliminate dimensional differences. For example, the CPU frequency decay rate is mapped to the interval [0,1] (rated frequency corresponds to 1, maximum decay corresponds to 0). The task attribute vector is based on the real-time, reliability, and security requirements of the task. For example, the real-time dimension extracts the maximum allowable delay time of the task (unit: ms), the reliability dimension extracts the task success rate threshold (such as 99.9%), and the security dimension extracts the data encryption level or access control policy level, forming a three-dimensional attribute vector.

[0067] The mapping between environmental factors and weights is dynamically generated based on the Analytic Hierarchy Process (AHP) and historical task execution data. For example, when the ambient electromagnetic interference intensity exceeds 50mT, the computing stability of the intelligent terminal decreases. At this time, the real-time weight is automatically increased to 0.5 (from 0.3), the reliability weight is adjusted to 0.3, and the safety weight is 0.2, to prioritize on-time task completion. If the ambient temperature and humidity are within the rated operating range of the equipment, the default weights (real-time 0.3, reliability 0.4, safety 0.3) are applied. This mapping is stored in the edge computing node's weight knowledge base and can be updated online using a reinforcement learning algorithm.

[0068] Specifically, for CPU frequency data, the ratio of the current frequency to the rated frequency is calculated to obtain the frequency attenuation rate (e.g., if the current frequency is 1.2GHz and the rated frequency is 1.8GHz, the attenuation rate is 0.67); for memory bandwidth data, the ratio of the actual bandwidth to the theoretical bandwidth is measured through a benchmark program (such as STREAM) to obtain the utilization index; for the real-time vector component of the relay protection task, the real-time vector component is set to 50ms (allowing delay), the reliability is set to 99.99%, and the security is set to important encryption; for the device status analysis task, the real-time vector component is set to 1000ms, the reliability is set to 99%, and the security is set to ordinary encryption. Then, the fitness score is calculated using the weighted Euclidean distance method: for the performance vector S of terminal i, the weighted Euclidean distance method is used to calculate the fitness score. i The parameters are first normalized to the range [0, 1], and then the single-dimensional matching degree is calculated according to the formula. After the matching degree is obtained, the weights are adjusted using reinforcement learning methods to finally obtain the matching degree matrix. Specifically, when the electromagnetic interference eigenvalue in the environmental feature vector exceeds the threshold, the weight adjustment mechanism is triggered. Using a reinforcement learning model (such as DQN) based on historical task delay data, the real-time weight is increased by 0.2 and the reliability weight is reduced by 0.1. After the weight adjustment, a consistency check is performed to ensure that the weight vector meets the consistency index of the AHP (CI ≤ 0.1). Otherwise, the weight vector is recalculated. Taking a substation scenario as an example, the performance vector of terminal i1 is normalized to [0.8, 0.95, 1.0], and the attribute vector of task j1 is normalized to [0.9, 0.99, 1.0]. The default weights are [0.3, 0.4, 0.3]. The obtained real-time matching degree is: fr f = 1-(0.9-0.8) / 0.9≈0.89; the reliability matching degree is: fk = 0.95 / 0.99≈0.96; and the security matching degree is: fp = 1.0. The comprehensive score is 0.3×0.89+0.4×0.96+0.3×1.0=0.94, that is, the matrix element F[1,1] = 0.94. If the environmental electromagnetic interference increases at this time and the weight is adjusted to [0.5, 0.2, 0.3], the new score is 0.5×0.89+0.2×0.96+0.3×1.0=0.935, and the matrix element is updated to 0.935.

[0069] In one possible embodiment, the target environment data includes temperature and humidity data, dust pollution data, and electromagnetic interference data; the n first computing performance data include n computing power parameters; and after controlling the n smart terminals to execute edge decision-making according to the target task allocation strategy, the method further includes the following steps:

[0070] A1. Preprocess the temperature and humidity data, the dust pollution data, and the electromagnetic interference data to obtain first temperature and humidity data, first dust pollution data, and first electromagnetic interference data;

[0071] A2. Perform feature extraction on the first temperature and humidity data, the first dust pollution data, and the first electromagnetic interference data to obtain a first temperature and humidity feature vector, a first dust pollution feature vector, and a first electromagnetic interference feature vector;

[0072] A3. Performing feature fusion on the first temperature and humidity feature vector, the first dust pollution feature vector, and the first electromagnetic interference feature vector based on a preset feature fusion algorithm to obtain a first fused feature vector.

[0073] A4. Determine environmental change parameters of the target power operation area based on the first fused feature vector;

[0074] A5. When the environmental change parameter is greater than a preset first threshold, dynamically adjust the first fitness matrix using a reinforcement learning method to obtain a second fitness matrix;

[0075] A6. Determine a first task allocation strategy based on the constraint condition and the second fitness matrix;

[0076] A7. Determine the maximum computing power parameter among the n computing power parameters;

[0077] A8. When the maximum computing power parameter is less than or equal to a preset second threshold, the first task allocation strategy is adjusted according to a preset task reduction strategy and the first computing performance data to obtain a second task allocation strategy.

[0078] Environmental data preprocessing ensures the accuracy of feature extraction by eliminating acquisition noise and dimensional differences. Temperature and humidity data preprocessing includes removing outliers that exceed 20% of the sensor's range (e.g., temperatures >125°C or <-40°C) and using a 5-point sliding average filter to remove random fluctuations. Dust pollution data requires median filtering to eliminate pulse interference, and then normalizing the particle concentration to the [0,1] range using maximum and minimum values. Electromagnetic interference data requires bandpass filtering to separate power frequency interference (50Hz) from high-frequency pulse interference to prevent coupling between signals in different frequency bands. Computing power parameters include CPU frequency, memory bandwidth, storage read and write rates, and other indicators. Real-time values ​​must be obtained through benchmark programs (e.g., Geekbench, IOzone) and compared with the device's rated parameters to generate a normalized computing power coefficient.

[0079] Among them, the determination of environmental change parameters combines time series analysis and anomaly detection algorithms, and the adjustment of the fitness matrix by reinforcement learning is based on the closed-loop optimization mechanism of "state-action-reward". When the environmental change parameters exceed the first threshold (such as the comprehensive feature vector change rate >30%), it indicates that the current environment has significantly affected the performance of the smart terminal and the reconstruction of the fitness matrix needs to be triggered; if the maximum computing power parameter is lower than the second threshold (such as 50% of the rated computing power), it means that the system has entered a resource-constrained state and it is necessary to reduce tasks to ensure the execution of key tasks.

[0080] Specifically, after the temperature and humidity data are collected by the SHT31 sensor, invalid values ​​with temperatures greater than 125°C or less than -40°C and humidity greater than 100% RH are first eliminated, and then the data is smoothed using a moving average filter (window size 5); after the dust pollution data are collected by the Sharp GP2Y1010AU0F sensor, sudden pollution peaks are removed through a median filter (window size 3), and then the data is normalized according to a concentration range of 0-5000 / 0.1L; after the electromagnetic interference data are collected by the Hall sensor, the interference frequency band is separated by a Butterworth bandpass filter (passband 10Hz-100kHz), and then the dimensionality effect is eliminated through Z-Score normalization. After collecting data from various sensors, feature extraction is performed on each raw data item. For temperature and humidity data, the mean, variance, and rate of change (e.g., the temperature change per minute) are calculated for a 5-minute sliding window to form a 6-dimensional temperature and humidity feature vector. For dust pollution data, the peak particle concentration, cumulative amount, and trend (e.g., hourly concentration increment) are extracted to form a 3-dimensional feature vector. For electromagnetic interference data, a 3-layer decomposition using the db4 wavelet is performed to extract the energy proportion, peak frequency, and energy entropy of each frequency band to form a 10-dimensional feature vector. After feature extraction, the resulting features need to be fused. Using the Kalman filter fusion algorithm, the weight coefficients for each feature dimension are trained based on historical data. The 3-dimensional feature vectors are fused into a 19-dimensional first fused feature vector. This vector is then reduced to 10 dimensions using principal component analysis (PCA) to eliminate redundant information. The environmental change parameter is calculated using a sliding time window (10-minute window length) and the isolation forest algorithm. The first fused feature vector sequence is sampled to generate a sliding window of 10 feature vectors. The isolation forest algorithm then calculates an anomaly score for each window (higher scores indicate more dramatic environmental changes). The maximum anomaly score within the window is used as the environmental change parameter. For example, if the high-frequency energy content of the electromagnetic interference signature increases from 20% to 60% within 10 minutes, the isolation forest algorithm identifies this window as an anomaly and outputs a higher change parameter value. The reinforcement learning model uses a deep Q-network (DQN) architecture. The state space is the first fused feature vector, and the action space is the adjustment step size (±0.1) of the weights of each dimension of the fitness matrix. The model stores state-action-reward triplets in an experience replay pool (capacity 10,000). The value network is updated every 100 iterations using the temporal difference (TD) algorithm. The weight adjustment policy is then optimized using a policy gradient algorithm to ultimately generate the second fitness matrix. The task reduction strategy first sorts the task list according to the priority in the task attributes, with the high-priority tasks retained at 100%, the medium-priority tasks retained at 50%, and the low-priority tasks retained at 20%. Then, the resource allocation of the retained tasks is linearly adjusted according to the remaining computing power parameters (for example, for every 10% decrease in computing power, the task data processing volume is reduced by 15%).For example, when the maximum computing power parameter drops to 40% of the rated value, all high-priority tasks are executed first, then 50% of the medium-priority tasks are selected for execution, and only 20% of the low-priority tasks are executed. The sampling frequency of the selected tasks is reduced by 50% to match the computing power limit.

[0081] It should be noted that the setting of environmental change parameters and computing power thresholds must be combined with the characteristics of the power operation scenario: in high-load substation areas, the first threshold is set to 0.5, and the second threshold is set to 60% of the rated computing power to trigger the adjustment mechanism in advance; in conventional power operation scenarios, the first threshold is set to 0.3, and the second threshold is set to 40%, balancing decision sensitivity and system stability. This dynamic adjustment mechanism achieves adaptive optimization of task allocation in power operation environments through real-time data processing and intelligent decision-making by electronic devices, improving the collaborative execution efficiency of distributed intelligent terminals and system robustness.

[0082] In a possible embodiment, determining the environmental change parameter of the target power operation area according to the first fused feature vector specifically includes the following steps:

[0083] A41. Sample the first fused feature vectors according to a preset window length using a sliding time window algorithm to obtain a first fused feature vector sequence; the first fused feature vector sequence includes t first fused feature vectors, where t is an integer greater than 0;

[0084] A42. Detect the first fused feature vector sequence based on a preset anomaly detection algorithm to obtain t anomaly scores;

[0085] A43. Determine the maximum anomaly score among the t anomaly scores;

[0086] A44. Determine a first fused feature vector corresponding to the maximum anomaly score to obtain a second fused feature vector;

[0087] A45. Input the second fused feature vector into a preset machine learning algorithm to obtain the environmental change parameter.

[0088] The length of the sliding time window is determined based on sensitivity to environmental changes and computational efficiency. The anomaly detection algorithm used is the Isolation Forest algorithm, which isolates sample points by constructing a binary tree and uses the path length of these sample points to assess the degree of anomaly. Compared to traditional statistical methods, it is more suitable for real-time detection of high-dimensional feature vectors. The Lightweight Gradient Boosting Tree (LightGBM) or the Random Forest algorithm are preferred machine learning algorithms because they can rank feature importance and automatically select feature dimensions that are sensitive to environmental changes. The anomaly score represents the degree to which the environmental state deviates from the normal distribution and ranges from [0 to 1]. Higher scores indicate more drastic environmental changes. The second fused feature vector, representing the environmental state at the moment of maximum anomaly, is reduced to less than 10 dimensions through principal component analysis (PCA) to prevent overfitting of the machine learning model. The environmental change parameter is ultimately quantized to a value between 0 and 1, with 0 indicating a stable environment and 1 indicating an extreme anomaly. This parameter can be used to directly trigger the task reallocation mechanism (e.g., dynamic adjustment is initiated when the parameter is greater than 0.5).

[0089] Specifically, the window sliding step is set to 50% of the window length (for example, a 5-minute window corresponds to a 2.5-minute sliding step), ensuring the data overlap rate to capture continuous changes, using a circular buffer to store the feature vector sequence, and automatically removing the earliest sample when a new sample is written, maintaining a fixed window length t = 12. Then, 100 isolated trees are constructed, and each tree randomly selects 30% of the feature dimensions and sample points for splitting. Taking LightGBM as an example, the 10-dimensional principal components of the second fused feature vector and label data (expert annotation levels corresponding to historical environmental change events) are first input into the model for training, setting the learning rate to 0.1, the tree depth to 6, and the number of iterations to 100. The optimal hyperparameters are selected through cross-validation. The following example illustrates this: in a substation main transformer area scenario, when the transformer is overloaded and the oil temperature rises suddenly, the sliding window collects 12 fused feature vectors. The temperature change rate of the eighth vector's temperature and humidity features reaches 2°C / min (exceeding the normal 0.5°C / min), and the proportion of high-frequency energy in the electromagnetic features increases from 15% to 40%. The isolation forest's anomaly score for this vector is 0.85 (threshold 0.5), which is the maximum value in the window. The LightGBM model outputs an environmental change parameter of 0.78 based on the vector's temperature and humidity change rate (0.7) and electromagnetic high-frequency energy proportion (0.8), triggering the task reallocation mechanism.

[0090] It should be noted that the preset window length and anomaly detection threshold need to be dynamically adjusted based on the type of power equipment. For enclosed equipment such as GIS switchgear, where temperature and humidity change slowly, the window length can be set to 10 minutes, with an anomaly threshold of 0.6. For open-air transmission lines, which are significantly affected by weather, the window length can be set to 5 minutes, with an anomaly threshold of 0.4. These are not set here.

[0091] This mechanism combines time series analysis with machine learning to achieve quantitative characterization of environmental changes, provide accurate environmental state input for subsequent reinforcement learning to adjust the fitness matrix, and ensure the decision-making accuracy of distributed intelligent terminals in dynamic environments.

[0092] In a possible embodiment, dynamically adjusting the first fitness matrix using a reinforcement learning method to obtain a second fitness matrix specifically includes the following steps:

[0093] A51. Obtain m pieces of first task status information corresponding to the first fitness matrix;

[0094] A52. Generate m first state weights corresponding to the m first task state information according to the m first task state information using the reinforcement learning method;

[0095] A53. Extracting the fitness scores corresponding to the m pieces of first task status information from the first fitness matrix to obtain m first fitness scores;

[0096] A54. Adjust the m first fitness scores according to the m first state weights to obtain m second fitness scores;

[0097] A55. Adjust the first fitness matrix according to the m second fitness scores to obtain the second fitness matrix.

[0098] The first task state information is a multidimensional parameter set representing the task execution environment and requirements. It includes the task's real-time delay threshold, reliability success rate requirements, security encryption level, and the terminal performance fluctuation coefficient in the current environment. This state information is collected in real time by sensors deployed on smart terminals, forming the state vectors of m tasks. A reinforcement learning approach employs a deep deterministic policy gradient algorithm, combining a policy network with a value network to construct an end-to-end weight generation model. This model can output the optimal state weight vector in a continuous state space. The first state weight reflects the impact of the task state information on the fitness matrix, ranging from 0 to 1, and is dynamically optimized through a reinforcement learning reward mechanism. For example, when the real-time delay threshold of a task approaches the terminal's computing power limit, the corresponding state weight is automatically increased to prioritize the task's assigned fitness. If ambient temperature and humidity cause a general decline in terminal performance, the reliability-related state weight is increased to ensure the execution success rate of high-reliability tasks. The second fitness score is dynamically modified by weighting, allowing the matrix to adapt to shifts in the match between task requirements and terminal performance caused by environmental changes.

[0099] Step S340: determining a target task allocation strategy based on preset constraints and the first fitness matrix.

[0100] Among them, the preset constraints cover the dual limitations of task requirements and terminal resources, including task real-time thresholds (such as relay protection tasks requiring a response of ≤100ms), reliability thresholds (task success rate ≥99.9%), security protocols, and terminal resource constraints (CPU utilization ≤80%, remaining memory ≥200MB, power ≥20%). These constraints are dynamically parsed by the policy engine of the edge computing node and converted into hard constraints (such as invalid task allocation when security is not met) and soft constraints (such as real-time performance is flexibly adjusted through weight coefficients) that can be executed by the algorithm. The first fitness matrix serves as the optimization objective function, and its element values ​​represent the degree of matching between the task and the terminal. It is necessary to combine the constraints to screen the feasible solution space to ensure that the allocation strategy meets both business needs and does not exceed the upper limit of terminal resources.

[0101] The target task allocation strategy is generated through an iterative search using a heuristic optimization algorithm, employing a combination of Hungarian algorithm initialization and particle swarm optimization (PSO) refinement. The Hungarian algorithm solves the bipartite graph matching problem between tasks and terminals, rapidly generating an initial feasible solution. The PSO algorithm simulates the social collaboration of a swarm of particles to search for the global optimal solution within the solution space. By dynamically adjusting the weight coefficients of the fitness matrix (for example, the real-time weight increases with increasing environmental electromagnetic interference), it balances the priorities of multi-dimensional constraints, ensuring that high-priority tasks (such as equipment failure warnings) are preferentially assigned to the terminals with the highest fitness.

[0102] In a possible embodiment, determining the target task allocation strategy based on the preset constraint conditions and the first fitness matrix specifically includes the following steps:

[0103] 341. Determine the allocation relationships between the n intelligent terminals and the m job tasks based on a preset Hungarian algorithm and the first fitness matrix to obtain a third task allocation strategy; the third task allocation strategy includes k first task allocation relationships; the k first task allocation relationships include allocating one of the m job tasks to at least one of the n intelligent terminals; k is less than or equal to n*m;

[0104] 342. Generate k particles according to the third task allocation strategy, and initialize the initial position and initial velocity of each of the k particles; each particle corresponds to a second task allocation relationship;

[0105] 343. Iteratively optimize the k particles according to the constraint condition and a preset iterative formula, stop the iteration when the k particles meet the preset iterative condition, and obtain a global optimal solution at the time of stopping the iteration;

[0106] 344. Adjust the first task allocation relationship in the second task allocation strategy according to the global optimal solution to obtain the target task allocation strategy.

[0107] The Hungarian algorithm prioritizes highly compatible task-terminal pairs by finding the maximum weighted match in the matrix. For example, it assigns relay protection tasks with high real-time requirements to terminals with the lowest computational latency. The third task allocation strategy, serving as the initial solution, must meet hard constraints (such as terminal security protocol matching and remaining battery power ≥ 20%). Allocation relationships that do not meet these constraints are automatically eliminated (by setting the corresponding matrix elements to 0). The PSO algorithm searches for the global optimal solution in the solution space by simulating the social behavior of a swarm of particles. Each particle represents a possible task allocation strategy, and its position and velocity are dynamically updated during the iteration process. Preset iteration conditions include a maximum number of iterations (e.g., 100), a fitness function convergence threshold (e.g., a fitness change of <1% after 10 consecutive iterations), or computational resource exhaustion. The global optimal solution is the allocation strategy that maximizes the fitness function value among all particles. This function comprehensively considers the fitness matrix score, task priority, and resource constraint satisfaction. The constraints include task real-time thresholds (such as ≤100ms), terminal CPU utilization ≤80%, and memory ≥200MB. These are integrated into the fitness calculation through a penalty function mechanism, which imposes a penalty on allocation strategies that violate the constraints, reducing their fitness values ​​and guiding the algorithm to search for feasible solutions.

[0108] Specifically, the first fitness matrix is ​​first converted into a cost matrix (1-fitness score), and the Kuhn-Munkres (KM) algorithm is used to solve the maximum weight matching of the bipartite graph to generate the initial allocation relationship. For example, after the fitness matrix of 3 tasks and 4 terminals is processed by the KM algorithm, the initial allocation of task j1→terminal i2, j2→i1, and j3→i3 is obtained. Then, the particle swarm optimization algorithm is used to obtain the global optimal solution. In the particle swarm optimization algorithm, each particle is encoded as a vector of length m, and the element value is the terminal index (1~n). For example, particle [2,1,3] represents j1→i2, j2→i1, and j3→i3; the initial position is randomly generated, but must meet hard constraints (such as the remaining resources of terminal i ≥ the requirements of task j), and the initial speed is set to [0,0,…,0] or a random decimal. The speed update formula, position update formula, and fitness calculation formula are as follows:

[0109]

[0110]

[0111] in, represents the velocity of particle i in d dimension at t iterations, w represents the inertia weight, c1, c2 represent the learning factor, r1, r2 represent the random number [0, 1], p id represents the individual optimal position of particle i, p gd represents the global optimal position, represents the position of particle i in dimension d at iteration t, represents the position of particle i in the d dimension at the t-1 iteration, F represents the fitness score, m represents the total number of tasks, F'[x j ,j] indicates that task j is assigned to terminal x j The fitness score, prj j represents the priority of task j, λ represents the penalty coefficient, and penalty is the number of constraint violations.

[0112] Specifically, the optimal solution is adjusted through constraints. If the optimal solution still violates the soft constraint (such as terminal CPU utilization of 75% to 80%), resources are released by dynamically adjusting the task sampling frequency; if it violates the hard constraint (such as utilization > 80%), task priority preemption is triggered, the allocation relationship of low-priority tasks is eliminated, and they are reallocated to other terminals. Taking a substation scenario as an example, when m = 5 tasks (j1-j5) and n = 4 terminals (i1-i4), the Hungarian algorithm generates the initial allocation: j1→i2, j2→i1, j3→i3, j4→i4, j5→i1 (i1 executes j2 and j5 simultaneously). If i1's remaining memory (150MB) is less than j5's required memory (200MB), a hard constraint violation is detected, and j5→i1 is eliminated, with k=4. During the PSO iteration, particles explore new allocations through velocity updates, such as adjusting the particle from [2,1,3,4,1] to [2,1,3,4,2], with j5→i2. At the 60th iteration, the fitness function converges to the global optimal solution: j1→i2, j2→i1, j3→i3, j4→i4, j5→i2. The resource utilization of all terminals is less than 80%, and the overall fitness score improves by 18%.

[0113] In a possible embodiment, after adjusting the first task allocation relationship in the second task allocation strategy according to the global optimal solution to obtain the target task allocation strategy, the method further includes the following steps:

[0114] B1. Determine the priority parameters of the m job tasks according to the target task allocation strategy to obtain m priority parameters;

[0115] B2. Obtain computing resource parameters required for the m job tasks to obtain m first computing resource parameters;

[0116] B3. Obtain the remaining computing resource parameters of the n smart terminals to obtain n remaining computing resource parameters;

[0117] B4. Determine an execution sequence of the job tasks corresponding to the m priority parameters to obtain a first execution sequence;

[0118] B5. Determine the total number of computing resources based on the m first computing resource parameters and the n remaining computing resource parameters, respectively, to obtain a second computing resource parameter and a total remaining computing resource parameter;

[0119] B6. If the total remaining computing resource parameter is greater than the second computing resource parameter, executing all the job tasks in the first execution sequence on the n intelligent terminals according to the first execution sequence;

[0120] B7. If the total remaining computing resource parameter is less than or equal to the second computing resource parameter, determining the number of job tasks that need to be reduced based on the total remaining computing resource parameter and the m first computing resource parameters to obtain the number of tasks to be reduced;

[0121] B8. Reduce the tasks in the first execution sequence by reducing a number of operation tasks to obtain a second execution sequence;

[0122] B9. Execute all the job tasks in the second execution sequence in the n intelligent terminals according to the second execution sequence.

[0123] The priority parameter represents the business importance of the job task. The first computing resource parameter includes metrics such as the number of CPU cycles, memory capacity, and storage bandwidth required for task execution. These parameters are statically acquired using a task profile analysis tool (e.g., a relay protection task requires a 200MHz CPU and 100MB of memory). Remaining computing resource parameters are acquired through real-time reporting from smart terminals and include remaining CPU frequency, available memory, and remaining battery power. For example, terminal i1 reports a remaining CPU frequency of 1.2GHz, 300MB of memory, and 30% battery power.

[0124] The execution sequence is generated based on a "priority first" principle, with high-priority tasks prioritized. The total remaining computing resource parameter is the cumulative value of the remaining resources of n terminals (e.g., total remaining CPU frequency 4.8 GHz = 1.2 GHz x 4 terminals). The second computing resource parameter is the sum of the resources required by m tasks (e.g., total CPU requirement 5.2 GHz). The number of tasks to be reduced is calculated using a "greedy strategy," accumulating resource requirements starting with low-priority tasks and determining the number of tasks to be reduced when remaining resources are insufficient, ensuring that high-priority tasks are executed first.

[0125] Specifically, weights are calculated based on normalized task attributes. The task list is sorted in descending order by priority parameter. For tasks of the same priority, they are sorted in ascending order by real-time requirements (lower latency thresholds give priority), thus creating a strict priority queue. Next, the first execution sequence is traversed, starting with the lowest-priority task: if the cumulative resource demand + the current task demand ≤ the total remaining resources, the accumulation continues; otherwise, the current task is recorded as the first task that cannot be executed, and the number of tasks removed equals the total number of subsequent tasks. For example, if the total remaining CPU resources are 4.8GHz, the total task demand is 5.2GHz, and the cumulative low-priority task demand is 0.8GHz (exceeding 0.4GHz), the number of tasks removed is 1 (retaining only the first three high / medium priority tasks). Then, according to the principle of "keeping the high and discarding the low," for example, in the original sequence [Task 1 (high), Task 2 (high), Task 3 (medium), Task 4 (low)], if the number of tasks removed is 1, Task 4 is removed, resulting in the new sequence [Task 1, Task 2, Task 3]. To illustrate, consider a substation scenario with four terminals with a total remaining CPU resource of 4.8GHz and 1.2GB of memory. There are five task requirements: Task 1 (high, CPU 1.5 GHz), Task 2 (high, 1.3 GHz), Task 3 (medium, 1.0 GHz), Task 4 (medium, 0.8 GHz), and Task 5 (low, 0.6 GHz). The total demand = 1.5 + 1.3 + 1.0 + 0.8 + 0.6 = 5.2 GHz > 4.8 GHz. Then, starting with low-priority Task 5, the frequency is accumulated: 0.6 GHz ≤ 4.8 GHz. Continuing, Task 4 accumulates to 1.4 GHz ≤ 4.8 GHz, Task 3 accumulates to 2.4 GHz, Task 2 accumulates to 3.7 GHz, and Task 1 accumulates to 5.2 GHz > 4.8 GHz. The reduction number = 1 (Task 5). The second execution sequence is [Task 1, Task 2, Task 3, Task 4], with a total demand of 4.6 GHz ≤ 4.8 GHz. During execution, the sampling frequency of Task 4 is dynamically adjusted (from 100 Hz to 80 Hz), releasing 0.2 GHz of resources to ensure full execution.

[0126] Step S350: Control the n smart terminals to execute edge decision-making according to the target task allocation strategy.

[0127] The edge decision-making process is based on a distributed collaborative mechanism, implementing policies through bidirectional communication between edge computing nodes and smart terminals. The target task allocation strategy includes mappings between tasks and terminals, resource allocation parameters, and execution priorities, which must be converted into a set of instructions executable by the terminal. When executing edge decisions, smart terminals must respond to environmental changes and resource fluctuations in real time, dynamically adjusting execution plans through task scheduling middleware to ensure the reliability of high-priority tasks. As the decision-making hub, edge computing nodes are responsible for policy issuance, status monitoring, and exception handling, forming a closed-loop control system of "decision-execution-feedback."

[0128] For easier understanding, see Figure 4 , Figure 4 This is a flow chart of another method for allocating tasks to smart terminals based on AI edge decision-making provided by an embodiment of the present application. Figure 4 As can be seen, first, real-time environmental data such as electromagnetic interference, temperature and humidity fluctuations, and dust pollution in the power operation environment are collected. This type of data is a key factor affecting the performance and task execution of smart terminals. For example, electromagnetic interference intensity can affect terminal communication reliability, and temperature and humidity affect hardware heat dissipation efficiency. The collected data is transmitted to the edge computing node, where it undergoes preprocessing such as filtering, denoising, and normalization to provide high-quality input for subsequent terminal performance prediction. This addresses the issues of high noise and inconsistent dimensionality in environmental data and ensures the accuracy of the underlying decision-making data. Next, the preprocessed environmental feature vectors are input into a pre-established "environmental factor-terminal performance" mapping model to predict the perception capabilities and computing performance of different smart terminals. This model is trained using historical data to quantify the impact of the environment on terminals. Next, for computing tasks in power operations, requirements are analyzed from multiple dimensions, including real-time performance, reliability, and security. Key task attributes are extracted, and combined with the terminal performance prediction results, a weighted calculation is used to construct a fitness matrix between the task and the terminal. This quantifies the degree of match between the two, providing a quantitative basis for subsequent optimization and allocation, achieving a structured mapping between task requirements and terminal capabilities. Then, a heuristic optimization algorithm is employed, using the fitness matrix as the optimization objective. Task requirements and terminal resource constraints are incorporated to search for an initial task-to-terminal allocation strategy. Through population iteration, the algorithm searches for optimal solutions within the solution space that satisfy these constraints. For example, this ensures that high-real-time tasks are assigned to terminals with the lowest computational latency and highest reliability. This solves the combinatorial optimization challenge of task allocation under multiple constraints and generates a preliminary, feasible task allocation plan. Simultaneously, the system continuously monitors the dynamics of the power operation environment. When environmental factors (such as sudden changes in electromagnetic interference or dramatic fluctuations in temperature and humidity) change significantly, a policy adjustment mechanism is triggered. Terminal performance is re-predicted, the fitness matrix is ​​updated, and heuristic optimization is performed again, enabling real-time adjustments to the task allocation strategy. Through reinforcement learning, the intelligent agent interacts with the environment, using task execution efficiency and resource utilization as reward signals to learn the optimal scheduling strategy for different environments. The learned strategy guides subsequent task allocation, enabling the system to self-optimize, improve task-terminal matching efficiency over the long term, and adapt to the long-term dynamic evolution of the power scenario. Finally, when terminal resources (such as CPU, memory, and power) no longer meet all task requirements, a threshold-based task reduction strategy is initiated. Dynamically adjust the task execution order and resource allocation based on task priority and remaining terminal resources to prioritize the execution of high-priority tasks.

[0129] For easier understanding, see Figure 5 , Figure 5This is a structural diagram of a task allocation method for an intelligent terminal based on AI edge decision-making provided by an embodiment of the present application, such as Figure 5As can be seen, the environmental data preprocessing module serves as data input for collecting and preprocessing environmental information for power operations. This module deploys multiple sensor types (e.g., electromagnetic sensors, temperature and humidity sensors, and particulate matter sensors) to collect real-time dynamic environmental data, targeting environmental variables such as electromagnetic interference (EMI), temperature and humidity fluctuations, and dust pollution in power scenarios. Subsequently, the module cleans and structures the raw environmental data through algorithms such as filtering and denoising (e.g., adaptive median filtering to address EMI noise), normalization, and outlier removal, generating an environmental feature vector. For example, EMI intensity is quantified as an interference index on a scale of 0-100, and temperature and humidity are converted to standard environmental parameters. This provides high-quality, highly reliable input data for subsequent analysis by edge computing nodes, addressing the high noise and dimensionality heterogeneity of power environmental data and strengthening the foundation for task allocation decisions. The edge computing node, acting as the data processing and transmission hub, receives the feature vectors output by the environmental data preprocessing module and integrates resource status data from smart terminals (e.g., CPU power, memory usage, remaining battery life, and communication link quality). Based on the lightweight computing capabilities of the edge, this node predicts the perception and computing performance of smart terminals in the current environment based on an "environment-terminal performance" mapping model, generating dynamic terminal capability information. Secondly, it constructs a task-terminal fitness matrix. Taking into account the real-time, reliability, and security requirements of power operation tasks, it quantifies the degree of match between tasks and terminals through weighted calculations, achieving a structured mapping of task requirements and terminal resources, providing a quantitative analysis basis for the AI ​​edge decision-making module. Based on the fitness matrix output by the edge computing node and constraints (such as terminal resource limits and task deadlines), the AI ​​edge decision-making module first searches for an initial task allocation scheme using heuristic optimization algorithms (such as genetic algorithms and particle swarm optimization algorithms), then searches for optimal solutions in the solution space that satisfy multiple constraints (e.g., ensuring that high-real-time tasks are assigned to low-latency, high-reliability terminals). Furthermore, a reinforcement learning mechanism (such as the DDPG algorithm) is introduced, using task execution efficiency (such as resource utilization and task completion rate) and environmental adaptability (such as the strategy's responsiveness to environmental changes) as reward signals. This allows the decision-making system to continuously interact with the power operation environment and learn optimal scheduling strategies for different scenarios. Generate a task allocation module as a policy implementation outlet, and convert the optimization strategy output by the AI ​​edge decision module into instructions executable by the smart terminal. Based on the decision results, this module generates a detailed allocation plan that includes task-terminal mapping relationships, resource allocation parameters (such as CPU computing power ratio, memory allocation), and execution priority (such as high-priority task preemption scheduling), and accurately pushes it to the corresponding smart terminal through the edge communication network (such as 5G private network, edge LAN). At the same time, establish a policy feedback mechanism to collect the terminal task execution status (such as task delay, resource occupancy rate) in real time, and transmit it back to the AI ​​edge decision module to ensure that the task allocation strategy is continuously optimized in a dynamic environment and that power operation tasks are executed efficiently and reliably.

[0130] For easier understanding, see Figure 6 , Figure 6 This is an architecture diagram of a task allocation model for an intelligent terminal based on AI edge decision-making provided by an embodiment of the present application, such as Figure 6 As can be seen, the data preprocessing model targets environmental data and smart terminal data in power scenarios. Through operations such as denoising, normalization, and outlier removal, it generates a high-quality, structured "environment-terminal" feature dataset. The LightBGM model constructs a mapping between the environment and smart terminal performance based on this preprocessed data. Leveraging the efficiency of the gradient boosting tree algorithm, this model uses environmental features as input and terminal perception capabilities (such as sensor data acquisition accuracy) and computing performance (such as task execution latency) as output to train an "environment-terminal performance" prediction model. Through multiple rounds of iterative optimization of the decision tree structure, it accurately captures the impact of environmental factors on terminal performance. The reinforcement learning model introduces a dynamic decision-making and environmental interaction mechanism. Based on the terminal performance profiles output by the LightBGM model and incorporating the real-time, reliability, and safety requirements of power operation tasks, an intelligent decision-making framework for task allocation is constructed. The reinforcement learning agent uses task execution efficiency (such as resource utilization and task completion rate) and environmental adaptability (such as the speed at which the strategy responds to environmental changes) as reward signals. It continuously explores and learns within the environment to generate a dynamic task allocation strategy. The particle swarm optimization model complements global optimization by fine-tuning the strategies output by the reinforcement learning model. This model simulates the collaborative optimization behavior of a particle swarm, uses the task-terminal fitness matrix as the optimization objective, incorporates terminal resource constraints and task demand constraints, and searches for the global optimal solution within the solution space. Through particle position updates (speed formula adjustment) and fitness evaluation (weighted calculation of the fitness matrix), the detailed task allocation plan is optimized.

[0131] For easier understanding, see Figure 7 , Figure 7 This is a schematic diagram of the structure of a smart terminal provided by an embodiment of the present application. Figure 7It can be seen that the smart terminal 700 includes: a sensor 710, a CPU 720, an Ubuntu 20.04 operating system 730, and a communication unit 740. Among them, the sensor 710 serves as the data collection entrance of the smart terminal and assumes the function of sensing the power operation environment and equipment status information. In response to the needs of power scenarios, multiple types of sensors can be integrated, such as electromagnetic sensors for capturing the intensity of environmental electromagnetic interference, temperature and humidity sensors for monitoring temperature and humidity changes in the operating area, dust sensors for collecting the concentration of polluted particles, and power-specific sensors (such as current / voltage sensors, partial discharge sensors) to achieve accurate collection of power equipment operating parameters. Sensor 710 converts analog signals of the physical world (such as electromagnetic waveforms, temperature changes) into digital signals through analog-to-digital conversion (ADC) technology, providing basic data for subsequent CPU processing. As the computing core of the smart terminal, CPU720 is responsible for edge-side processing of received sensor data. Based on its computing power, it executes data preprocessing algorithms (such as filtering and denoising, feature extraction), cleans and structures the raw data collected by sensor 710. For example, it uses fast Fourier transform (FFT) to analyze the frequency characteristics of electromagnetic interference signals and extract the changing trends of temperature and humidity data. At the same time, CPU720 carries the task execution logic of the smart terminal, runs applications related to power operations (such as equipment status monitoring algorithms and fault warning models), and makes local lightweight decisions based on the processed data, such as determining whether the equipment has abnormal temperature rise or whether the electromagnetic interference exceeds the threshold. Ubuntu 20.04 operating system 730, this operating system is based on the Linux kernel and has the characteristics of open source, stability and efficiency. It can adapt to the hardware resources of smart terminals (such as CPU computing power and memory space) and allocate system resources (such as process scheduling and memory management) to applications running on CPU 720. At the same time, Ubuntu 20.04 operating system 730 integrates rich development tools and libraries (such as Python development environment and machine learning library), supports smart terminals to deploy AI edge decision algorithms (such as reinforcement learning reasoning and fitness matrix calculation), and realizes software environment support from data processing to intelligent decision-making. As the interaction outlet of the smart terminal, the communication unit 740 is responsible for data interaction with edge computing nodes, other smart terminals and power operation and maintenance platforms. It supports multiple communication protocols and network standards, such as 5G to achieve high-speed and low-latency data transmission, LoRa to meet the communication requirements of wide coverage and low power consumption, and Ethernet to ensure stable connection within the local area network. The communication unit 740 encapsulates the data processed by the CPU 720 into a network data packet and uploads it to the edge computing node through an adapted communication protocol for the formulation of the task allocation strategy; at the same time, it receives the task allocation instructions issued by the edge computing node and drives the smart terminal to adjust the task execution logic.

[0132] It can be seen that by executing the above-mentioned task allocation method of intelligent terminals based on AI edge decision-making, the accuracy and robustness of task allocation of distributed terminal decisions can be improved in power operations.

[0133] The above mainly introduces the solution of the embodiment of the present application from the perspective of the execution process of the method side. It is understandable that, in order to realize the above functions, the electronic device includes a hardware structure and / or software module corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiment provided herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software driven hardware manner depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0134] In the case of dividing each functional module into corresponding functional modules, Figure 8 This is a functional module block diagram of a task allocation device for an intelligent terminal based on AI edge decision-making provided in an embodiment of the present application. The task allocation device 800 for an intelligent terminal based on AI edge decision-making is applied to an electronic device connected to n intelligent terminals in a target power operation area, where n is an integer greater than 1. The device includes:

[0135] An acquisition module 810 is configured to acquire environmental data in the target power operation area and task attribute data of the power operation task, thereby obtaining target environmental data and m task attribute data; the power operation task includes m operation tasks, where m is an integer greater than 1;

[0136] a calculation module 820 configured to send the target environment data to the n smart terminals respectively, and have the n smart terminals perform calculations based on the target environment data respectively to obtain n first computing performance data, and return the n first computing performance data to the electronic device;

[0137] A determination module 830 is configured to determine a first fitness matrix between tasks and smart terminal performance based on the n first computing performance data and the m task attribute data; and determine a target task allocation strategy based on preset constraints and the first fitness matrix.

[0138] The control module 840 is used to control the n intelligent terminals to execute edge decision-making according to the target task allocation strategy.

[0139] In a possible embodiment, the calculation module 820, in determining the first fitness matrix between the task and the smart terminal performance according to the n first computing performance data and the m task attribute data, is specifically configured to:

[0140] Performing feature extraction on the n first computing performance data to obtain n first performance vectors;

[0141] Performing feature extraction on the m task attribute data to obtain m task attribute vectors;

[0142] Determining a fit score between the first performance and the task attribute based on the n first performance vectors and the m task attribute vectors, to obtain n*m fit scores;

[0143] Determine a fitness matrix according to the n*m ​​fitness scores;

[0144] Determining a weight coefficient corresponding to the fitness matrix according to a mapping relationship between preset environmental factors and fitness matrix weights to obtain a first weight coefficient;

[0145] The fitness matrix is ​​adjusted according to the first weight coefficient to obtain the first fitness matrix.

[0146] In a possible embodiment, the target environment data includes: temperature and humidity data, dust pollution data, and electromagnetic interference data; the n first computing performance data include n computing power parameters; after controlling the n intelligent terminals to execute edge decision according to the target task allocation strategy, the computing module 820 is further specifically configured to:

[0147] Preprocessing the temperature and humidity data, the dust pollution data, and the electromagnetic interference data respectively to obtain first temperature and humidity data, first dust pollution data, and first electromagnetic interference data;

[0148] performing feature extraction on the first temperature and humidity data, the first dust pollution data, and the first electromagnetic interference data respectively to obtain a first temperature and humidity feature vector, a first dust pollution feature vector, and a first electromagnetic interference feature vector;

[0149] Performing feature fusion on the first temperature and humidity feature vector, the first dust pollution feature vector, and the first electromagnetic interference feature vector based on a preset feature fusion algorithm to obtain a first fused feature vector;

[0150] determining an environmental change parameter of the target power operation area according to the first fused feature vector;

[0151] When the environmental change parameter is greater than a preset first threshold, dynamically adjusting the first fitness matrix using a reinforcement learning method to obtain a second fitness matrix;

[0152] Determine a first task allocation strategy according to the constraint condition and the second fitness matrix;

[0153] Determine a maximum computing power parameter among the n computing power parameters;

[0154] When the maximum computing power parameter is less than or equal to a preset second threshold, the first task allocation strategy is adjusted according to a preset task reduction strategy and the first computing performance data to obtain a second task allocation strategy.

[0155] In a possible embodiment, in determining the environmental change parameter of the target power operation area according to the first fused feature vector, the calculation module 820 is further configured to:

[0156] The first fused feature vector is sampled according to a preset window length using a sliding time window algorithm to obtain a first fused feature vector sequence; the first fused feature vector sequence includes t first fused feature vectors, where t is an integer greater than 0;

[0157] Detecting the first fused feature vector sequence based on a preset anomaly detection algorithm to obtain t anomaly scores;

[0158] Determining a maximum anomaly score among the t anomaly scores;

[0159] Determine a first fused feature vector corresponding to the maximum anomaly score to obtain a second fused feature vector;

[0160] The second fused feature vector is input into a preset machine learning algorithm to obtain the environmental change parameter.

[0161] In a possible embodiment, the calculation module 820, in dynamically adjusting the first fitness matrix using the reinforcement learning method to obtain the second fitness matrix, is further configured to:

[0162] Obtaining m pieces of first task status information corresponding to the first fitness matrix;

[0163] Generating m first state weights corresponding to the m first task state information according to the m first task state information by the reinforcement learning method;

[0164] Extracting the fitness scores corresponding to the m first task state information from the first fitness matrix to obtain m first fitness scores;

[0165] Adjusting the m first fitness scores according to the m first state weights to obtain m second fitness scores;

[0166] The first fitness matrix is ​​adjusted according to the m second fitness scores to obtain the second fitness matrix.

[0167] In a possible embodiment, the determining module 830, in determining the target task allocation strategy according to the preset constraint conditions and the first fitness matrix, is specifically configured to:

[0168] Determine, based on a preset Hungarian algorithm and the first fitness matrix, an allocation relationship between the n intelligent terminals and the m job tasks, and obtain a third task allocation strategy; the third task allocation strategy includes k first task allocation relationships; the k first task allocation relationships include allocating one of the m job tasks to at least one of the n intelligent terminals; k is less than or equal to n*m;

[0169] Generate k particles according to the third task allocation strategy, and initialize the initial position and initial velocity of each of the k particles; each particle corresponds to a second task allocation relationship;

[0170] Iteratively optimizing the k particles according to the constraint conditions and a preset iterative formula, stopping the iteration when the k particles meet the preset iterative conditions, and obtaining a global optimal solution at the time of stopping the iteration;

[0171] The first task allocation relationship in the second task allocation strategy is adjusted according to the global optimal solution to obtain the target task allocation strategy.

[0172] In a possible embodiment, after adjusting the first task allocation relationship in the second task allocation strategy according to the global optimal solution to obtain the target task allocation strategy, the determination module 830 is further configured to:

[0173] Determining the priority parameters of the m job tasks according to the target task allocation strategy to obtain m priority parameters;

[0174] Obtain computing resource parameters required for the m job tasks to obtain m first computing resource parameters;

[0175] Obtaining the remaining computing resource parameters of the n intelligent terminals to obtain n remaining computing resource parameters;

[0176] Determine an execution sequence of the job tasks corresponding to the m priority parameters to obtain a first execution sequence;

[0177] Determine the total number of computing resources according to the m first computing resource parameters and the n remaining computing resource parameters, and obtain a second computing resource parameter and a total remaining computing resource parameter;

[0178] If the total remaining computing resource parameter is greater than the second computing resource parameter, executing all the job tasks in the first execution sequence on the n intelligent terminals according to the first execution sequence;

[0179] If the total remaining computing resource parameter is less than or equal to the second computing resource parameter, determining the number of job tasks that need to be reduced based on the total remaining computing resource parameter and the m first computing resource parameters to obtain the number of tasks to be reduced;

[0180] Reducing the tasks in the first execution sequence by reducing a number of operation tasks to obtain a second execution sequence;

[0181] All job tasks in the second execution sequence are executed in the n intelligent terminals according to the second execution sequence.

[0182] It should be noted that the specific functional implementation of the task allocation device 800 of the intelligent terminal based on AI edge decision-making can be found in the above Figure 3 The description of a task allocation method for an intelligent terminal based on AI edge decision-making is shown, for example, the acquisition module 810 is used to implement the relevant content of executing S310, and the control module 840 is used to implement the relevant content of executing S350, which will not be repeated. The various units or modules in the task allocation device 800 for an intelligent terminal based on AI edge decision-making can be individually or completely merged into one or several other units or modules to form a structure, or one (some) of the units or modules can be further divided into multiple functionally smaller units or modules to form a structure, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present invention. The above-mentioned units or modules are divided based on logical functions. In actual applications, the functions of one unit (or module) are implemented by multiple units (or modules), or the functions of multiple units (or modules) are implemented by one unit (or module).

[0183] It can be seen that the task allocation device of an intelligent terminal based on AI edge decision-making described in the embodiment of the present application is applied to an electronic device, wherein the electronic device is connected to n intelligent terminals in the target power operation area, where n is an integer greater than 1, and obtains the target environment data and m task attribute data by acquiring the environmental data and task attribute data in the target power operation area. The target environment data is sent to the n intelligent terminals respectively, and the n intelligent terminals respectively perform calculations based on the target environment data to obtain n first computing performance data, and return the n first computing performance data to the electronic device, determine the first fitness matrix of the task and the intelligent terminal performance based on the n first computing performance data and the m task attribute data, determine the target task allocation strategy based on the preset constraints and the first fitness matrix; and control the n intelligent terminals to execute edge decision-making based on the target task allocation strategy. In the power operation scenario, the accuracy and robustness of task allocation of distributed terminal decisions can be improved.

[0184] An embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, and the computer program enables a computer to execute part or all of the steps of any method described in the above method embodiments, and the above computer includes an electronic device.

[0185] The present application also provides a computer program product comprising a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. The computer program product may be a software installation package, and the computer may comprise an electronic device.

[0186] It should be noted that, for the above-mentioned various embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations. Those skilled in the art should know that this application is not limited by the order of the actions described, because some steps in the embodiments of the present application can be performed in other orders or simultaneously. In addition, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions, steps, modules or units involved are not necessarily required by the embodiments of the present application.

[0187] In the above embodiments, the embodiments of the present application have different focuses on the description of each embodiment. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0188] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0189] The steps of the method or algorithm described in the embodiments of the present application can be implemented in hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC. In addition, the ASIC can be located in a terminal device or a management device. Of course, the processor and storage medium can also be present in a terminal device or a management device as discrete components.

[0190] The modules / units included in the devices and products described in the above embodiments may be software modules / units, hardware modules / units, or partly software modules / units and partly hardware modules / units. For example, for the devices and products applied to or integrated in the chip, the modules / units included therein may all be implemented in the form of hardware such as circuits, or at least part of the modules / units may be implemented in the form of software programs, which run on the processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits; for the devices and products applied to or integrated in the chip module, the modules / units included therein may all be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (such as chip, circuit module, etc.) or different components of the chip module, or at least part of the modules / units may be It is implemented in the form of a software program, which runs on the processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated in the terminal equipment, the various modules / units contained therein can be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or different components in the terminal equipment, or, at least some modules / units can be implemented in the form of a software program, which runs on the processor integrated inside the terminal equipment, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits.

[0191] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above description is only a specific implementation method of the embodiments of the present application and is not intended to limit the scope of protection of the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the scope of protection of the embodiments of the present application.

Claims

1. A task allocation method for intelligent terminals based on AI edge decision-making, characterized in that: Applied to an electronic device, the electronic device being connected to n smart terminals in a target power operation area, where n is an integer greater than 1, the method comprising: Acquire environmental data in the target power operation area and task attribute data of the power operation task to obtain target environmental data and m task attribute data; the power operation task includes m operation tasks, where m is an integer greater than 1; Sending the target environment data to the n smart terminals respectively, and having the n smart terminals perform calculations based on the target environment data respectively to obtain n first computing performance data, and returning the n first computing performance data to the electronic device; Determining a first fitness matrix between the task and the smart terminal performance according to the n first computing performance data and the m task attribute data; Determining a target task allocation strategy based on preset constraints and the first fitness matrix; Control the n intelligent terminals to execute edge decision-making according to the target task allocation strategy.

2. The method according to claim 1, wherein The determining, based on the n first computing performance data and the m task attribute data, a first fitness matrix between the task and the smart terminal performance includes: Performing feature extraction on the n first computing performance data to obtain n first performance vectors; Performing feature extraction on the m task attribute data to obtain m task attribute vectors; Determining a fit score between the first performance and the task attribute based on the n first performance vectors and the m task attribute vectors, to obtain n*m fit scores; Determine a fitness matrix according to the n*m ​​fitness scores; Determining a weight coefficient corresponding to the fitness matrix according to a mapping relationship between preset environmental factors and fitness matrix weights to obtain a first weight coefficient; The fitness matrix is ​​adjusted according to the first weight coefficient to obtain the first fitness matrix.

3. The method according to claim 1 or 2, wherein: The target environment data includes: temperature and humidity data, dust pollution data, and electromagnetic interference data; the n first computing performance data include n computing power parameters; after controlling the n smart terminals to execute edge decision-making according to the target task allocation strategy, the method further includes: Preprocessing the temperature and humidity data, the dust pollution data, and the electromagnetic interference data respectively to obtain first temperature and humidity data, first dust pollution data, and first electromagnetic interference data; performing feature extraction on the first temperature and humidity data, the first dust pollution data, and the first electromagnetic interference data respectively to obtain a first temperature and humidity feature vector, a first dust pollution feature vector, and a first electromagnetic interference feature vector; Performing feature fusion on the first temperature and humidity feature vector, the first dust pollution feature vector, and the first electromagnetic interference feature vector based on a preset feature fusion algorithm to obtain a first fused feature vector; determining an environmental change parameter of the target power operation area according to the first fused feature vector; When the environmental change parameter is greater than a preset first threshold, dynamically adjusting the first fitness matrix using a reinforcement learning method to obtain a second fitness matrix; Determine a first task allocation strategy according to the constraint condition and the second fitness matrix; Determine a maximum computing power parameter among the n computing power parameters; When the maximum computing power parameter is less than or equal to a preset second threshold, the first task allocation strategy is adjusted according to a preset task reduction strategy and the first computing performance data to obtain a second task allocation strategy.

4. The method according to claim 3, wherein The determining of the environmental change parameter of the target power operation area according to the first fused feature vector includes: The first fused feature vector is sampled according to a preset window length using a sliding time window algorithm to obtain a first fused feature vector sequence; the first fused feature vector sequence includes t first fused feature vectors, where t is an integer greater than 0; Detecting the first fused feature vector sequence based on a preset anomaly detection algorithm to obtain t anomaly scores; Determining a maximum anomaly score among the t anomaly scores; Determine a first fused feature vector corresponding to the maximum anomaly score to obtain a second fused feature vector; The second fused feature vector is input into a preset machine learning algorithm to obtain the environmental change parameter.

5. The method according to claim 3, wherein The step of dynamically adjusting the first fitness matrix using a reinforcement learning method to obtain a second fitness matrix includes: Obtaining m pieces of first task status information corresponding to the first fitness matrix; Generating m first state weights corresponding to the m first task state information according to the m first task state information by the reinforcement learning method; Extracting the fitness scores corresponding to the m first task state information from the first fitness matrix to obtain m first fitness scores; Adjusting the m first fitness scores according to the m first state weights to obtain m second fitness scores; The first fitness matrix is ​​adjusted according to the m second fitness scores to obtain the second fitness matrix.

6. The method according to claim 1 or 2, wherein: The determining of the target task allocation strategy according to the preset constraint conditions and the first fitness matrix includes: Determine, based on a preset Hungarian algorithm and the first fitness matrix, an allocation relationship between the n intelligent terminals and the m job tasks, and obtain a third task allocation strategy; the third task allocation strategy includes k first task allocation relationships; the k first task allocation relationships include allocating one of the m job tasks to at least one of the n intelligent terminals; k is less than or equal to n*m; Generate k particles according to the third task allocation strategy, and initialize the initial position and initial velocity of each of the k particles; each particle corresponds to a second task allocation relationship; Iteratively optimizing the k particles according to the constraint conditions and a preset iterative formula, stopping the iteration when the k particles meet the preset iterative conditions, and obtaining a global optimal solution at the time of stopping the iteration; The first task allocation relationship in the second task allocation strategy is adjusted according to the global optimal solution to obtain the target task allocation strategy.

7. The method according to claim 6, wherein After adjusting the first task allocation relationship in the second task allocation strategy according to the global optimal solution to obtain the target task allocation strategy, the method further includes: Determining the priority parameters of the m job tasks according to the target task allocation strategy to obtain m priority parameters; Obtain computing resource parameters required for the m job tasks to obtain m first computing resource parameters; Obtaining the remaining computing resource parameters of the n intelligent terminals to obtain n remaining computing resource parameters; Determine an execution sequence of the job tasks corresponding to the m priority parameters to obtain a first execution sequence; Determine the total number of computing resources according to the m first computing resource parameters and the n remaining computing resource parameters, and obtain a second computing resource parameter and a total remaining computing resource parameter; If the total remaining computing resource parameter is greater than the second computing resource parameter, executing all the job tasks in the first execution sequence on the n intelligent terminals according to the first execution sequence; If the total remaining computing resource parameter is less than or equal to the second computing resource parameter, determining the number of job tasks that need to be reduced based on the total remaining computing resource parameter and the m first computing resource parameters to obtain the number of tasks to be reduced; Reducing the tasks in the first execution sequence by reducing a number of operation tasks to obtain a second execution sequence; All job tasks in the second execution sequence are executed in the n intelligent terminals according to the second execution sequence.

8. A task allocation device for intelligent terminals based on AI edge decision-making, characterized in that: Applied to an electronic device, the electronic device is connected to n smart terminals in a target power operation area, where n is an integer greater than 1, and the device includes: an acquisition module, configured to acquire environmental data in the target power operation area and task attribute data of the power operation task, to obtain target environmental data and m task attribute data; the power operation task includes m operation tasks, where m is an integer greater than 1; a calculation module, configured to send the target environment data to the n smart terminals respectively, and have the n smart terminals perform calculations based on the target environment data respectively to obtain n first calculation performance data, and return the n first calculation performance data to the electronic device; a determination module, configured to determine a first fitness matrix between tasks and intelligent terminal performance based on the n first computing performance data and the m task attribute data; and determine a target task allocation strategy based on preset constraints and the first fitness matrix; A control module is used to control the n intelligent terminals to execute edge decision-making according to the target task allocation strategy.

9. An electronic device, characterized in that: include: a processor, a memory, a communication interface, and one or more programs; The one or more programs are stored in the memory and configured to be executed by the processor, wherein the programs include instructions for executing the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 7.

Images