Intelligent 5G edge cooperation power supply driving system

Abstract

The invention relates to the technical field of 5G edge collaborative power supply driving, and discloses an intelligent 5G edge collaborative power supply driving system, which comprises a state sensing module used for collecting multi-source data of each power supply device in the power supply driving system in real time through a multi-source sensor; the load prediction module is used for predicting a future load by using a load prediction algorithm and outputting a load prediction value; the thermal path analysis module is used for constructing a thermal path diagram, predicting a thermal diffusion trend and adjusting power output of the power supply driving system; the power dispatching optimization module is used for adjusting the power of the power supply based on the optimization objective function; and the remote cooperative control module is used for the cloud platform to generate an optimization strategy according to the real-time states of load prediction, temperature prediction and power supply scheduling. According to the method, the integrity and the time relevance of the equipment operation state expression are improved by constructing the continuous time state evolution model in a modeling mode of combining the multi-source state perception and the state space model.

Description

Technical Field

[0001] This application relates to the technical field of 5G edge collaborative power drive, and specifically provides an intelligent 5G edge collaborative power drive system. Background Art

[0002] With the rapid development of emerging industries such as intelligent manufacturing, data centers, and high-performance edge computing, the deployment of large-scale distributed power systems in industrial scenarios has become increasingly common. Such systems often include multiple power modules, power supply circuits, and temperature control components operating in parallel, with highly dynamic operating states, severe load fluctuations, and tight thermal load coupling. To ensure the efficient and safe operation of the system, there is an urgent need for a predictive, adaptive, and collaborative power drive management mechanism that can achieve real-time regulation and global perception between multiple nodes.

[0003] Currently, in most industrial power applications, power management mainly adopts fixed-threshold judgment and scheduling methods based on historical experience rules. Load changes trigger power switching through limits, and when the temperature exceeds the standard, the fan is started for cooling or manual intervention is carried out. In some relatively advanced systems, simple load prediction models, such as linear regression and BP neural networks, are also introduced for short-term power prediction.

[0004] However, the existing intelligent 5G edge collaborative power drive technology lacks a modeling mechanism for the dynamic relationships between state variables, resulting in the difficulty of using the collected state data for trend analysis and control decisions. Scheduling often relies on the current value, ignoring the change rate and future trends. Secondly, the load prediction algorithm fails to integrate multi-dimensional factors such as temperature and reliability, and the model update mechanism is also imperfect. In the face of system disturbances, the response is lagged and the error accumulation is serious. Moreover, the thermal management method still mainly relies on single-point perception, making it difficult to identify the thermal diffusion path in advance and easily causing local overheating without awareness. Therefore, the present invention provides an intelligent 5G edge collaborative power drive system to solve the deficiencies existing in the prior art. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the purpose of this application is to provide an intelligent 5G edge collaborative power drive system, which solves the problems of imperfect state perception modeling, lack of dynamic adaptability in load prediction, lack of thermal diffusion path analysis in thermal management, and lack of a global priority mechanism in collaborative scheduling in the existing power drive technology.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: An intelligent 5G edge collaborative power drive system includes: A state perception module, which is used to collect the load, output current, voltage, temperature, and device reliability information of each power device in the power drive system in real time through multi-source sensors, generate a system state vector, and construct a state space model through state space modeling technology; A load prediction module, which is used to analyze the historical change trend of the load of the power drive system based on the state space model, predict the future load by using the load prediction algorithm, and output the load prediction value; A thermal path analysis module, which is used to collect the temperature information in the power drive system in real time based on multi-source sensors, construct a thermal path map and predict the thermal diffusion trend, and adjust the power output of the power drive system; A power scheduling optimization module, which is used to receive the load prediction value and the temperature prediction result of the power drive system, and adjust the power of the power supply based on the optimization objective function; A remote collaborative control module, which is used to upload the status information of each edge node to the cloud through the 5G edge computing network. The cloud platform generates an optimization strategy according to the real-time status of load prediction, temperature prediction and power scheduling, and issues it to the edge node through the 5G network for collaborative control.

[0007] Preferably, the state perception module includes: A multi-source sensor group, including a current sensor, a voltage sensor, a temperature sensor, a load pressure sensor, and a reliability status sensor, which is used to collect the real-time operation parameters of each power supply device in the power drive system, including load data, output current, voltage, temperature, and device reliability; A state construction unit, which is used to perform normalization processing and vector quantization expression on the collected operation parameters to form a system state vector; a modeling unit, which is used to construct a state space model based on the system state vector and model the evolution trend of each operation state. The state space model satisfies the following form: X(t + 1) = A·X(t) + B·U(t); Wherein, X(t + 1) represents the system state vector at time t + 1; X(t) is the state vector; A is the state transition matrix; B is the control input matrix; U(t) is the external adjustment input vector.

[0008] Preferably, the load prediction module includes: A historical load analysis unit, which is used to extract the load change characteristics from the historical state vector provided by the state perception module; A prediction model training unit, which is used to train the load prediction model based on the extracted features, and the model supports adaptive adjustment of the prediction accuracy under different working conditions; A prediction output unit, which is used to generate the load prediction value in the future time period according to the current state space model and the trained model, and transfer the prediction value to the power scheduling optimization module.

[0009] Preferably, training the load prediction model based on the extracted features includes: Performing denoising, normalization and time series segmentation processing on the historical load data; Dynamic model parameter updates within the sliding window; The newly generated prediction model is tested on recent load data for error, and the model parameters are adjusted based on the error feedback using the following formula: in, is the predicted load value; L i is the actual load value; is the mean square error loss function; n is the total number of samples.

[0010] Preferably, the thermal path analysis module includes: Temperature information acquisition unit, used to extract real-time temperature information of various parts of the system from the status perception module; The heat path construction unit is used to build a heat path diagram model based on the equipment structure and temperature sensor layout, and mark the heat transfer relationship; The diffusion trend prediction unit is used to infer the future heat flow diffusion trend based on the current thermal path diagram and temperature status, and output the prediction results to drive the power scheduling optimization module to dynamically adjust the power.

[0011] Preferably, the power scheduling optimization module includes: A strategy generation unit, configured to receive data provided by the load prediction module and the thermal path analysis module and construct a power regulation plan based on the optimization goal; a policy execution unit, configured to adjust the output power of the power supply device and execute power switching according to the power regulation scheme; The feedback processing unit is used to collect execution results and device response status, build local execution feedback information, and upload the feedback data to the remote collaborative control module.

[0012] Preferably, constructing a power regulation scheme based on the optimization target includes the following steps: Read load prediction value and temperature prediction results; Establish scheduling constraints based on power caps, equipment reliability, and temperature thresholds; Evaluate power utilization efficiency and system thermal load balance under various power distribution schemes; Filter the optimal scheduling plan and transfer it to the execution unit to issue scheduling instructions.

[0013] Preferably, the remote collaborative control module includes: Edge information collection unit, used to collect state vectors, prediction data and scheduling feedback results from each edge node; The global state modeling unit is used to aggregate the information of each node and construct a system-level global state map; A global policy optimization unit for dynamically calculating a cross-node scheduling optimization policy based on the current global state graph; A policy distribution and synchronization unit for distributing the global policy to each edge node and coordinating the power output and load matching among the nodes.

[0014] Preferably, the cross-node scheduling optimization policy dynamically calculated based on the current global state graph adopts the following formula: where S i is the scheduling priority score of the i-th edge node; L i is the current predicted load of node i; L max represents the maximum predicted load among all nodes; T i is the current temperature prediction value of node i; T max is the maximum temperature prediction value among all nodes; α and β are weight coefficients used to regulate the influence weights of load and temperature in the priority score, and satisfy α + β = 1.

[0015] There is also provided an intelligent 5G edge collaborative power drive method, including the following steps: Real-time collect the load, output current, voltage, temperature and device reliability information of each power device in the power drive system through multi-source sensors, process the collected multi-dimensional operation parameters, generate a system state vector and construct a state space model; extract the historical load component from the state space model, construct a load change feature set, train an adaptive load prediction model based on the feature set, and combine the current state vector for real-time input to generate load prediction values within multiple future time steps; Based on the obtained temperature states of each component, construct a heat path graph model on the basis of the power drive system topology structure, and combine the current temperature state and material thermal property parameters to calculate the heat transfer rate and trend on different paths, and predict the spatio-temporal distribution of heat diffusion; Based on the load prediction results and the heat diffusion prediction trend, dynamically construct a power regulation optimization objective function, and establish multiple scheduling constraint conditions including power upper limit, temperature threshold and device reliability lower limit; Based on the constructed optimization objective function and constraint conditions, use a multi-scheme evaluation method to calculate the optimal power distribution strategy, and adjust the output power of each power device in real time according to the calculation results; Upload the state vectors, load prediction results, heat diffusion trends and local scheduling results of each edge node to the cloud through the 5G network.

[0016] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present invention adopts a modeling method combining multi-source state perception and state space model. By uniformly normalizing and modeling multiple parameters such as current, voltage, temperature, load pressure, and reliability, a continuous-time state evolution model is constructed, fundamentally improving the integrity and time correlation of the expression of the device operating state. Compared with the monitoring method in the prior art that monitors multiple parameters in isolation and lacks a state evolution mechanism, it effectively solves the problems of fragmented device states and unfavorable prediction modeling.

[0017] 2. The present invention constructs a lightweight load prediction model with a sliding window for training, and combines an error feedback mechanism and a weighted loss function to dynamically update the model parameters, enabling high prediction accuracy to be maintained under multi-condition and variable load conditions. Compared with the prior art where traditional static load prediction algorithms have deficiencies in model generalization ability and inability to quickly adapt to system mutations, this solution overcomes the problems of response lag and prediction error accumulation.

[0018] 3. The present invention introduces a thermal path graph modeling mechanism, calculates the thermal diffusion trend between nodes based on graph topology and thermal coupling relationships, and can discover potential hot spot areas in advance by establishing a temperature-driven thermal migration estimation formula. Compared with previous thermal control strategies that rely on single-point temperature warnings and lack spatial correlation, it effectively alleviates the hidden danger of local failure caused by system thermal imbalance and improves the operating thermal safety.

[0019] 4. The present invention constructs a cross-node priority scoring model in the multi-edge node collaborative scheduling scenario, dynamically introduces load intensity and temperature level factors, completes global perception and scheduling control through the 5G network, and combines a delay factor to guide the scheduling landing point, solving the problems of many scheduling blind spots and rough resource allocation in existing distributed systems. In scenarios with intensive tasks or abnormal fluctuations, the system still has strong resource collaboration and controllability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the system architecture diagram of the present application; Figure 2 is the flowchart of the method steps of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following will further elaborate on the present application in conjunction with the attached Figure 1 - attached Figure 2 , and make a more detailed description of the present application.

[0022] Please refer to the attached Figure 1 , an intelligent 5G edge collaborative power drive system, comprising: A state perception module, configured to collect the load, output current, voltage, temperature, and device reliability information of each power device in the power drive system in real time through multi-source sensors, generate a system state vector, and construct a state space model through state space modeling technology; A load prediction module, which is used to analyze the historical change trend of the load of the power drive system based on the state space model, predict the future load by using the load prediction algorithm, and output the load prediction value; A thermal path analysis module, which is used to collect the temperature information in the power drive system in real time based on multi-source sensors, construct a thermal path map and predict the thermal diffusion trend to adjust the power output of the power drive system; A power scheduling optimization module, which is used to receive the load prediction value and the temperature prediction result of the power drive system, and adjust the power of the power supply based on the optimization objective function; A remote collaborative control module, which uploads the status information of each edge node to the cloud through the 5G edge computing network, generates a global power scheduling optimization strategy according to the global information, and the cloud platform generates an optimization strategy according to the real-time status of load prediction, temperature prediction and power scheduling, and issues it to the edge nodes through the 5G network for collaborative control.

[0023] For the state perception module, in this embodiment, the state perception module serves as the information entry part of the system and constitutes the technical starting point of the entire operation process, playing a key supporting role in subsequent modeling, load prediction, scheduling optimization, etc. of the system state. This module is mainly responsible for collecting the key operating parameters of each power supply device in the power drive system, preprocessing the collected original multi-dimensional data, vector construction and modeling, laying a foundation for the system to build a state space model. Generally, a data communication method in the form of a state vector is adopted between this module and the subsequent prediction and modeling module to achieve the unification of parameter structures and interface standardization. In some implementation manners, in order to improve the model accuracy and perception response speed, this module is deployed near the edge node side to realize local preprocessing by using the edge computing ability.

[0024] In this embodiment, the state perception module includes a multi-source sensor group, a state construction unit and a modeling unit, and the three work together to gradually complete the conversion process from perception to modeling.

[0025] Specifically, the multi-source sensor group includes but is not limited to the following categories: A current sensor, which is used to collect the output current signal of the power supply device; A voltage sensor, which is used to measure the voltage level of each node in the system; A temperature sensor, which is used to detect the local temperature rise and thermal distribution characteristics of the device; A load pressure sensor, which is used to reflect the current bearing state of the load device; A reliability status sensor, which is used to monitor the stability, aging trend or potential failure of the power supply device.

[0026] In a possible implementation, the above sensors are arranged between the internal control unit and the output port of the power module, and unified sampling is achieved through a synchronous clock system. The sampling period can be set in the range of 1 ms to 10 ms.

[0027] As an option, the state construction unit is used to normalize the original operating parameters from the multi-source sensor group and express them in a standardized vector form. The normalization uses the maximum-minimum standard deviation method, which can still maintain the contrast and relative change trend of parameter expression under dynamic working conditions.

[0028] Generally, the normalized data will be organized into a structured vector form to form the system state vector X(t), which contains the following elements: Real-time load parameter L t ; Current output current I t ; Output voltage V t ; Device surface temperature T t ; Reliability evaluation parameter R t .

[0029] The above vector form can adjust the element composition according to the device type, but the dimension consistency should be satisfied.

[0030] In this embodiment, the modeling unit uses the state space method for modeling. This method can perform time evolution modeling on the system operating state and establish the causal relationship between regulation and state. The basic form of the state space model is as follows: X(t + 1) = A·X(t) + B·U(t); Where, X(t) is the system state vector at time t; X(t + 1) represents the system state vector at time t + 1; A is the state transition matrix, reflecting the evolution relationship between the independent variables of the system, with a dimension of n×n; U(t) is the external regulation input vector, including system control quantities such as power regulation signals, with a dimension of m×1; B is the input control matrix, defining the influence degree of the input variables on the system state, with a dimension of n×m.

[0031] In a typical configuration, the numerical value of matrix A is preset by the device characteristics or obtained by fitting the training data. Matrix B can be defined individually according to the power supply type, power supply topology structure, and load coupling degree.

[0032] In some embodiments, to improve the state expression accuracy, the modeling unit also introduces a weighting coefficient or a transfer factor to fine-tune the model parameters through fitting the historical state data. The system state vector is not only used for time series prediction but also serves as the input basis for subsequent load prediction and scheduling optimization algorithms.

[0033] It should be noted that the modeling process is a real-time online modeling process. That is, as new state data is input, the model can be updated and adjusted dynamically in a rolling manner, without relying on a fixed training cycle.

[0034] As an extended approach, in a distributed deployment, each edge node can independently run the above-mentioned state awareness process and construct a local state space model. This can significantly improve the modularity and fault isolation ability of the system.

[0035] For the load prediction module, in this embodiment, after the state awareness module completes the basic system state collection and state modeling, the load prediction module is immediately started to run, which is used to conduct a forward-looking analysis and numerical prediction of the future load situation of the system. Generally, this module directly receives the state vector data output from the state modeling unit, and combines the historical load evolution trajectory to realize the prediction and deduction of the short-term or medium-term load trend by constructing a load prediction model. In some embodiments, the load prediction module not only predicts the trend of power load, but also can capture the change rate and periodic fluctuation characteristics to enhance the response ability of the model to dynamic working conditions.

[0036] In this embodiment, the load prediction module mainly includes a historical load analysis unit, a prediction model training unit, and a prediction output unit, and the three parts form a structured prediction calculation link.

[0037] Specifically, the historical load analysis unit is used to extract time series load data from the state vector obtained by the state awareness module and perform multi-stage preprocessing operations on this data.

[0038] In a possible implementation manner, the original load data is first segmented by a sliding window. The window length can be set to w, and the step size is set to s. Common values are, for example, w = 60 and s = 15. After segmentation, the following data normalization processing is adopted: where L i is the i-th piece of original load data; L min , L max respectively represent the minimum and maximum load values in the sample set; is the normalization result. This processing method can avoid interference from data with different dimensions to model training.

[0039] As an option, the prediction model training unit adopts a deep network model based on a time series modeling structure, such as a lightweight LSTM structure. The model encodes the input state sequence through an autoregressive mechanism and predicts the load values within a certain number of future time steps. In some implementations, an attention mechanism is introduced to enhance the model's ability to identify key periods.

[0040] Generally, to ensure the stability and generalization ability of the training model, an error feedback mechanism needs to be set up, and a loss function is defined for backpropagation adjustment. In the present invention, the loss function of the prediction model is defined as follows: where, is the predicted load value; L i is the actual load value; is the mean squared error loss function; n represents the total number of samples. This function is used to measure the prediction deviation.

[0041] In a possible design, the model also introduces a weighted loss strategy to enhance the sensitivity to high load intervals. Its weighted version can be expressed as: where, is the weighted mean squared error; w i is a weighted factor that is positively correlated with the load amplitude and is defined as: where the parameter γ is a coefficient for adjusting the weight strength; is the average load value of the training samples; L represents the total load value in the training sample set.

[0042] In this embodiment, the prediction output unit generates a load prediction sequence for the next k time steps according to the trained model and the current state input. This sequence is output to the scheduling optimization module in vector form, with a dimension of [k×1], where k is usually set between 15 and 30.

[0043] In some embodiments, to further improve the robustness of the load prediction, the system also incorporates a confidence estimation module to give a confidence interval for each prediction point. For example, the MC Dropout method is used for sample perturbation and probabilistic regression modeling to obtain the upper and lower bound estimates.

[0044] Some edge nodes can also deploy local micro-models to achieve local fast prediction response. When the deviation between the prediction result of the main model and the local actual load exceeds the threshold, the system will automatically trigger model retraining or structural fine-tuning.

[0045] For the thermal path analysis module, in this embodiment, after completing the modeling of the system operating state and the prediction of future load trends, it is necessary to further introduce an analysis mechanism for the temperature behavior of the device to form a structured understanding of the thermal state. Generally, power devices generate heat during continuous operation, and its diffusion in the system structure has obvious directionality and hierarchy. If not evaluated, it is extremely easy to cause local overheating of the device, thereby affecting the system stability. Therefore, the present invention sets up a thermal path analysis module, as an important part of the thermal state modeling, to dynamically model the thermal diffusion trend inside the system and output it for use by the subsequent power adjustment logic.

[0046] In this embodiment, the thermal path analysis module mainly includes a temperature information acquisition unit, a thermal path construction unit, and a diffusion trend prediction unit, and the three cooperate to complete the whole process from temperature perception to thermal behavior inference.

[0047] Specifically, the temperature information acquisition unit is responsible for extracting the temperature parameters of key structural parts from the state perception module. As an option, temperature sampling points are set at multiple key nodes such as the device surface, the center position of the power module, and the thermal coupling part, ensuring that the thermal distribution data is comprehensive and representative.

[0048] In a possible implementation, the sampling frequency is set between 1Hz - 5Hz, and the acquisition results include but are not limited to the instantaneous temperature T i , the temperature rise rate ΔT i / Δt and the equilibrium difference where represents the average temperature value of the adjacent structures near the current device.

[0049] The thermal path construction unit constructs a heat propagation graph model based on the device physical topology structure and the sensor layout position. Generally, this model can be represented as an undirected weighted graph G=(V, E, W), where; V represents the set of thermal nodes; corresponding to the sensor installation positions; E represents the thermal connection edges, indicating that there is thermal coupling between nodes; W represents the thermal conduction weight of the edges, which depends on the material thermal conductivity, structural distance, and ventilation conditions.

[0050] In some embodiments, the thermal weight w ij can be calculated as follows: where k ij is the thermal conductivity of the medium between nodes i and j; d ij is the structural distance between nodes; φ ij is an adjustment factor used to consider non-structural thermal effects such as air-cooled heat dissipation.

[0051] The diffusion trend prediction unit estimates the migration trend of heat between nodes based on the current heat path map and in combination with real-time temperature data. In a typical implementation, this process can be approximated as a two-dimensional discrete heat diffusion process, satisfying the following deviation fitting relationship: where represents the set of adjacent nodes of node i; Δt is the time step; T i (t) is the current temperature at time t; w ij is the heat path weight; T i (t + Δt) represents the temperature value of node i at time t + Δt; T j (t) represents the temperature value of node j at time t.

[0052] In some embodiments, to enhance the adaptability of the heat diffusion model, the system introduces a structure-aware factor, enabling the heat path modeling to be scalable and self-adaptive under different device structure types.

[0053] Generally, the output of the heat path analysis module will include multiple characteristic indicators, such as parameters like the maximum heat flux path, hot spot migration direction, heat equilibrium time, etc. These indicators can be used to determine whether to perform early load shedding or power reallocation.

[0054] For the power supply scheduling optimization module, in this embodiment, the load prediction module and the heat path analysis module respectively complete the prediction of the system power demand trend and the modeling analysis of the device temperature evolution. As the subsequent processing link, based on receiving the output results of the above two modules, the power supply scheduling optimization module comprehensively considers the operating state of the system and resource constraints, and constructs an optimal allocation strategy for the power supply power. Generally, the power supply scheduling optimization module not only needs to satisfy the dynamic response ability to the total system load, but also needs to control the device heat load level within an acceptable range to avoid the risks of overload or local thermal runaway.

[0055] In this embodiment, the power supply scheduling optimization module includes a strategy generation unit, a strategy execution unit, and a feedback processing unit. The three parts form a closed-loop control mechanism to realize the calculation, execution, and feedback of power scheduling.

[0056] Specifically, the input of the strategy generation unit is the predicted future load data and the heat diffusion trend results corresponding to each power supply device. As an option, the system first constructs a resource constraint vector for all available power nodes, including the following information: Node power upper limit The current predicted temperature value T i ; Safe temperature threshold Current device health factor R i 。

[0057] In a possible implementation, based on the above parameters, the system constructs the following scheduling constraint set Among them, P i represents the power value of node i; represents the set of power supply nodes; ΔT i is the temperature rise value within the prediction time window; R min is the minimum reliability requirement.

[0058] Generally, the policy generation unit constructs multiple candidate power scheduling schemes based on the constraints and uses the total power deviation and thermal load balance as the optimization objective function. In this embodiment, the objective function can be expressed as: Among them, P i represents the power allocated to the i-th node; L pred is the total predicted load; is the average system temperature; λ1 and λ2 are objective weighting coefficients used to balance the load satisfaction and thermal balance; n represents the number of nodes in the system; T represents the average temperature of the system.

[0059] As an option, in some implementations, methods such as genetic algorithms, particle swarm optimization, or reinforcement learning are used to solve the objective function to obtain the global optimal scheduling strategy.

[0060] After receiving the optimal scheduling scheme, the policy execution unit issues a power adjustment instruction to the corresponding power supply device through the edge control interface to achieve dynamic adjustment of the output power. In a possible implementation, the control interface supports millisecond-level response and has an automatic fault tolerance mechanism to handle execution deviations.

[0061] The feedback processing unit is responsible for real-time collecting the device response status and power output situation, and comparing the actual execution result with the expected value. In some embodiments, the feedback result will be used to adjust the priority of the subsequent scheduling strategy or re-evaluate the accuracy of the current state model to form an adaptive update mechanism.

[0062] In addition, in a multi-node deployment environment, the feedback processing unit can also collect high-frequency characteristic indicators such as scheduling delay, device power volatility, and temperature mutation rate as reference parameters for dynamically adjusting the window length or prediction step of the scheduling system.

[0063] For the remote collaborative control module, in this embodiment, after completing local prediction, thermal analysis, and power scheduling calculations, to achieve a wider and more coordinated operation control, a remote collaboration mechanism with a global perspective needs to be introduced. Generally, the system is deployed in a multi-node environment, and different power supply devices are distributed in multiple edge regions, with state differences and physical distribution dispersion. Relying solely on local decisions is difficult to ensure overall efficiency and system stability. Therefore, the present invention sets up a remote collaborative control module, which aggregates the status data of each node, generates a global scheduling strategy, and performs edge synchronization control based on the 5G network, constituting an important part of the system-level closed-loop scheduling.

[0064] In this embodiment, the remote collaborative control module mainly includes an edge information collection unit, a global status modeling unit, a global policy optimization unit, and a policy distribution and synchronization unit. The four cooperate with each other to build a dynamic perception and control mechanism covering all network nodes.

[0065] Specifically, the edge information collection unit collects status vectors, load prediction data, temperature evaluation results, and local scheduling feedback information from each power node in real time through the 5G network. In a possible implementation, the collected parameters include: the current power output value P of each node i ; the current temperature value T i and its predicted value the current predicted load L i ; the local execution delay τ i ; the device operation stability index R i .

[0066] As an option, to improve communication stability and scheduling accuracy, the collection unit adopts an edge caching mechanism. When the data sampling frequency is greater than the transmission frequency, key data with a larger change amplitude is preferentially uploaded.

[0067] The global status modeling unit then aggregates the above data from different nodes to construct a multi-dimensional global status map. In some embodiments, the status map uses a graph neural network structure to model the correlation between nodes, forming a dynamic status graph with a topological structure and status labels where: V represents the set of edge nodes; E represents the possible load migration or thermal coupling paths between nodes; A represents the set of attributes of each node, including predicted load, current power, temperature, response time, etc.

[0068] Generally, the status map will be used as the input of the optimizer for subsequent policy inference and scheduling sorting.

[0069] The global policy optimization unit is responsible for scheduling scoring and priority sorting of all edge nodes based on the constructed global state graph. In this embodiment, the scoring mechanism is based on the following priority calculation formula: Among them, S i is the scheduling priority score of the i-th node; L i is the current predicted load of the node; T i is the current temperature prediction value of the node; L max is the maximum value of the load prediction values among all nodes; T max is the maximum value of the temperature prediction values among all nodes; α and β are weighting coefficients used to balance the scheduling weights of load pressure and thermal safety, satisfying α + β = 1.

[0070] In some embodiments, the system introduces a node response ability adjustment factor θ i , and makes a secondary correction to the scheduling result: S i ′ = S i ·θ i , θ i = exp(-γ·τ i ); Among them, S i ′ represents the response ability of the i-th node after adjustment; S i represents the initial response ability of the i-th node; θ i is the adjustment factor of the i-th node; τ i is the scheduling response delay of the node; γ is the delay penalty coefficient. This mechanism is used to improve the execution reliability of scheduling instructions.

[0071] The policy distribution and synchronization unit distributes the finally sorted scheduling instructions and resource allocation policies to the edge nodes through the 5G URLLC channel respectively. In a possible implementation manner, this module supports a multicast scheduling push mechanism to synchronize linkage instructions for nodes in the same area or with similar structures, reducing the total communication cost.

[0072] In addition, after the policy is distributed, the module also has a status confirmation mechanism to confirm whether the scheduling instruction is successfully executed through heartbeat packets or return codes. The unconfirmed part will enter the retransmission channel and trigger an exception report.

[0073] In some embodiments, the remote collaborative control module also has a status evolution learning function. By comparing historical scheduling records and execution results, online training and updating of scheduling policies are carried out to improve the self-adaptability of global policies.

[0074] An intelligent 5G edge collaborative power driving method described below can be correspondingly referred to with an intelligent 5G edge collaborative power driving system described above.

[0075] Please refer to the attached Figure 2 , the present invention also provides an intelligent 5G edge collaborative power driving method, including the following steps: collecting in real time the load, output current, voltage, temperature, and device reliability information of each power device in the power driving system through multi-source sensors, processing the collected multi-dimensional operation parameters, generating a system state vector, and constructing a state space model; extracting the historical load component from the state space model, constructing a load change feature set, training an adaptive load prediction model based on the feature set, and combining the current state vector for real-time input to generate load prediction values within multiple future time steps; Based on the temperature states of each component obtained, constructing a heat path diagram model on the basis of the power driving system topology structure, and combining the current temperature state and material heat property parameters to calculate the heat transfer rate and trend on different paths, and predicting the spatio-temporal distribution of heat diffusion; Based on the load prediction results and the heat diffusion prediction trend, dynamically constructing a power regulation optimization objective function, and establishing multiple scheduling constraint conditions including a power upper limit, a temperature threshold, and a device reliability lower limit; Based on the constructed optimization objective function and constraint conditions, using a multi-scheme evaluation method to calculate the optimal power distribution strategy, and adjusting the output power of each power device in real time according to the calculation results; Uploading the state vectors, load prediction results, heat diffusion trends, and local scheduling results of each edge node to the cloud through the 5G network.

[0076] The method of this embodiment can be used to execute the above system embodiment, and its principle and technical effects are similar, which will not be elaborated here.

[0077] The embodiments of this specific implementation manner are all preferred embodiments of this application, and do not limit the protection scope of this application accordingly. The same components are denoted by the same reference numerals. Therefore: All equivalent changes made according to the structure, shape, and principle of this application should be covered within the protection scope of this application.

Claims

1. An intelligent 5G edge collaborative power drive system, characterized in that, Including: A state perception module, which is used to collect the load, output current, voltage, temperature and equipment reliability information of each power supply device in the power drive system in real time through multi-source sensors, generate a system state vector, and construct a state space model through state space modeling technology; A load prediction module, which is used to analyze the historical change trend of the load of the power drive system based on the state space model, predict the future load by using a load prediction algorithm, and output a load prediction value; A thermal path analysis module, which is used to collect the temperature information in the power drive system in real time based on multi-source sensors, construct a thermal path map and predict the thermal diffusion trend, and adjust the power output of the power drive system; A power supply scheduling optimization module, which is used to receive the load prediction value and the temperature prediction result of the power drive system, and adjust the power of the power supply based on an optimization objective function; A remote collaborative control module, which is used to upload the state information of each edge node to the cloud through a 5G edge computing network. The cloud platform generates an optimization strategy according to the real-time state of load prediction, temperature prediction and power supply scheduling, and issues it to the edge node through the 5G network for collaborative control.

2. The intelligent 5G edge collaborative power drive system according to claim 1, wherein The state perception module includes: A multi-source sensor group, including a current sensor, a voltage sensor, a temperature sensor, a load pressure sensor, and a reliability status sensor, which is used to collect the real-time operation parameters of each power supply device in the power drive system, including load data, output current, voltage, temperature and equipment reliability; A state construction unit, which is used to perform normalization processing and vectorization expression on the collected operation parameters to form a system state vector; A modeling unit, which is used to construct a state space model based on the system state vector and model the evolution trend of each operation state. The state space model satisfies the following form: X(t + 1) = A·X(t) + B·U(t); Where, X(t + 1) represents the system state vector at time t + 1; X(t) is the state vector; A is the state transition matrix; B is the control input matrix; U(t) is the external adjustment input vector.

3. An intelligent 5G edge collaborative power drive system according to claim 1, characterized in that The load prediction module includes: A historical load analysis unit, which is used to extract load change characteristics from the historical state vector provided by the state perception module; A prediction model training unit, which is used to train a load prediction model based on the extracted characteristics. The model supports adaptive adjustment of the prediction accuracy under different working conditions; A prediction output unit, which is used to generate a load prediction value for a future time period according to the current state space model and the trained model, and transfer the prediction value to the power supply scheduling optimization module.

4. An intelligent 5G edge collaborative power drive system according to claim 3, characterized in that, The training of the load prediction model based on the extracted characteristics includes: Performing denoising, normalization and time series segmentation processing on historical load data; Performing dynamic model parameter update within a sliding window; Verifying the error of the newly generated prediction model in recent load data, and adjusting the model parameters according to the error feedback. The following formula is used: Among them, is the predicted load value; L i is the actual load value; is the mean square error loss function; n represents the total number of samples.

5. An intelligent 5G edge collaborative power drive system according to claim 1, characterized in that, The thermal path analysis module includes: A temperature information collection unit, which is used to extract the real-time temperature information of each part of the system from the state perception module; A thermal path construction unit, which is used to establish a thermal path map model based on the device structure and the layout of temperature sensors, and mark the heat transfer relationship; A diffusion trend prediction unit, which is used to infer the future heat flow diffusion trend based on the current heat path map and temperature state, and output the prediction result to drive the power scheduling optimization module to perform power dynamic adjustment.

6. The intelligent 5G edge collaborative power drive system according to claim 1, characterized in that The power scheduling optimization module includes: A policy generation unit, which is used to receive the data provided by the load prediction module and the heat path analysis module, and construct a power adjustment plan based on the optimization goal; A policy execution unit, which is used to adjust the output power of the power supply device according to the power adjustment plan and perform power switching; A feedback processing unit, which is used to collect the execution result and the device response status, construct local execution feedback information, and upload the feedback data to the remote collaborative control module.

7. An intelligent 5G edge collaborative power drive system according to claim 6, characterized in that Constructing the power adjustment plan based on the optimization goal includes the following steps: Read the load prediction value and the temperature prediction result; Construct scheduling constraint conditions based on the power upper limit, device reliability, and temperature threshold; Evaluate the power utilization efficiency and system heat load balance under multiple power distribution schemes; Select the optimal scheduling plan and transfer it to the execution unit to issue a scheduling instruction.

8. An intelligent 5G edge collaborative power drive system according to claim 1, characterized in that The remote collaborative control module includes: An edge information collection unit, which is used to collect the state vector, prediction data, and scheduling feedback results from each edge node; A global state modeling unit, which is used to summarize the information of each node to construct a system-level global state map; A global policy optimization unit, which is used to dynamically calculate the cross-node scheduling optimization policy based on the current global state map; A policy distribution and synchronization unit, which is used to distribute the global policy to each edge node and coordinate the power output and load matching between nodes.

9. An intelligent 5G edge collaborative power drive system according to claim 8, characterized in that, Dynamically calculating the cross-node scheduling optimization policy based on the current global state map adopts the following formula: Among them, S i is the scheduling priority score of the i-th edge node; L i is the current predicted load of node i; L max represents the maximum predicted load among all nodes; T i is the current temperature prediction value of node i; T max is the maximum temperature prediction value among all nodes; α and β are weight coefficients used to regulate the influence weights of load and temperature in the priority score, satisfying α + β = 1.

10. A smart 5G edge collaborative power driving method, applied to a smart 5G edge collaborative power driving system according to any one of claims 1-9, characterized in that, Includes the following steps: Real-time collect the load, output current, voltage, temperature, and device reliability information of each power supply device in the power drive system through multi-source sensors, process the collected multi-dimensional operation parameters, generate a system state vector, and construct a state space model; Extract the historical load component from the state space model, construct a load change feature set, train an adaptive load prediction model based on the feature set, and combine the real-time input of the current state vector to generate the load prediction value within multiple future time steps; Based on the temperature state of each component obtained, construct a heat path map model on the basis of the power drive system topology structure, and combine the current temperature state and material heat characteristic parameters to calculate the heat transfer rate and trend on different paths, and predict the spatio-temporal distribution of heat diffusion; Based on the load prediction result and the heat diffusion prediction trend, dynamically construct a power adjustment optimization objective function, and establish multiple scheduling constraint conditions including the power upper limit, temperature threshold, and device reliability lower limit; Based on the constructed optimization objective function and constraint conditions, use a multi-scheme evaluation method to calculate the optimal power distribution strategy, and adjust the output power of each power supply device in real time according to the calculation result; Upload the state vector, load prediction result, heat diffusion trend, and local scheduling result of each edge node to the cloud through the 5G network.

Images