An energy management system based on cloud edge-end cooperation

Abstract

The application relates to the technical field of electric communication, and discloses an energy management system based on cloud-edge-end cooperation. The system comprises the following steps: an acquisition module collects communication terminal equipment energy consumption data through channel sensing processing to obtain an original energy consumption data set. A prediction module uses a C-EPCE algorithm to perform channel sensing prediction on an edge base station to obtain predicted energy consumption parameters. An allocation module inputs the predicted parameters into a CSI-DPA algorithm to perform dynamic allocation of transmission power, thereby obtaining power control instructions. An awareness module performs energy consumption awareness routing processing on the power control instructions through a NERA algorithm, thereby obtaining an optimal routing strategy. An optimization module performs multi-level optimization on the optimal routing strategy through a cooperative decision mechanism, thereby obtaining an energy management control scheme. The application improves the energy consumption optimization effect of a communication network and the system cooperation performance.

Description

Technical Field

[0001] This application relates to the field of electrical communication technology, and in particular to an energy management system based on cloud-edge-device collaboration. Background Technology

[0002] With the rapid development of 5G / 6G networks and the widespread application of edge computing technology, cloud-edge-device collaborative energy management methods have become an important technical means for energy-saving optimization of communication networks. In existing technologies, energy management of communication networks mainly adopts a centralized control strategy, where a cloud control center uniformly schedules network resources, or a distributed control strategy, where each base station independently performs power control and resource allocation. These methods can reduce total network energy consumption to some extent, but their optimization effect remains limited in complex network environments.

[0003] However, existing cloud-edge-device collaborative energy management methods have significant shortcomings. First, energy consumption data collection lacks channel awareness capabilities, failing to accurately reflect the impact of channel quality changes on energy demand, resulting in insufficient prediction accuracy. Second, power allocation algorithms do not fully consider the dynamic changes in channel state information, and static power control strategies are difficult to adapt to fluctuations in real-time channel conditions. Furthermore, the routing process lacks an energy consumption awareness mechanism; traditional shortest path algorithms ignore the energy consumption overhead of communication paths, affecting the overall network energy consumption optimization effect.

[0004] Based on the above analysis, the problem with existing technologies lies in the lack of effective coordination mechanisms among functional modules, preventing closed-loop optimization in key areas such as channel awareness, power allocation, and routing. Specifically, the lack of a channel-aware energy consumption prediction mechanism makes accurate demand forecasting for power allocation impossible; the failure to effectively integrate power allocation with channel state information hinders dynamic power optimization control; the lack of energy consumption awareness in routing selection prevents the selection of the optimal energy-consuming path; and the lack of a collaborative optimization mechanism for multi-level decision-making prevents the achievement of global optimization in network-wide energy management. These problems are interconnected and collectively constrain the overall performance of cloud-edge-device collaborative energy management. Summary of the Invention

[0005] This application provides a cloud-edge-device collaborative energy management system, which solves the technical problems of insufficient channel awareness, lack of dynamic optimization of power allocation, and poor multi-level decision-making collaboration in existing cloud-edge-device collaborative energy management methods, thereby improving the energy consumption optimization effect and system collaborative performance of communication networks.

[0006] Firstly, this application provides a cloud-edge-device collaborative energy management system, which includes:

[0007] The data acquisition module is used to collect energy consumption data from communication terminal equipment through channel sensing processing to obtain the raw energy consumption dataset.

[0008] The prediction module is used to perform channel-aware prediction processing on the edge base station based on the original energy consumption dataset using the C-EPCE algorithm to obtain predicted energy consumption parameters.

[0009] The allocation module is used to input the predicted energy consumption parameters into the CSI-DPA algorithm to dynamically allocate the transmit power and obtain power control commands.

[0010] The sensing module is used to perform energy consumption sensing routing processing on the communication path according to the power control command and the NERA algorithm to obtain the optimal routing strategy;

[0011] The optimization module is used to perform multi-level optimization of the optimal routing strategy through a collaborative decision-making mechanism to obtain an energy management and control scheme.

[0012] The technical solution provided in this application collects energy consumption data from communication terminal equipment through channel-aware processing, accurately obtaining the original energy consumption dataset reflecting channel quality changes, providing a reliable data foundation for subsequent prediction and optimization. The C-EPCE algorithm is specifically designed for the channel variation characteristics of communication networks. By analyzing channel fading models and propagation loss characteristics, it establishes a precise mapping relationship between channel quality and energy consumption requirements, significantly improving the accuracy of energy consumption prediction for edge base stations. The CSI-DPA algorithm fully utilizes channel state information for dynamic power allocation, solving for the optimal power allocation vector using the Lagrange multiplier method, ensuring precise optimization of power control while meeting communication service quality requirements. The NERA algorithm incorporates an energy consumption awareness mechanism in the routing process, calculating the energy consumption cost of communication paths through an improved Dijkstra algorithm framework, enabling the selection of the truly optimal energy consumption path and avoiding the shortcomings of traditional shortest path algorithms that ignore energy consumption factors. The collaborative decision-making mechanism, through multi-level optimization processing, achieves organic collaboration between the cloud, edge, and terminal three-layer architecture, eliminating decision conflicts at each level and ensuring the global optimality of the network-wide energy management and control scheme.

[0013] In specific applications of telecommunications technology, the C-EPCE algorithm's channel awareness capability enables energy consumption prediction to accurately reflect the dynamic changes of wireless channels, making it particularly suitable for base station energy management in mobile communication environments. The CSI-DPA algorithm, based on a dynamic power allocation mechanism using channel state information, fully leverages the advantages of channel feedback mechanisms in telecommunications systems, achieving real-time matching of power control with channel conditions. The NERA algorithm incorporates energy consumption factors into routing metrics, fully utilizing the path diversity of communication networks to optimize energy performance while ensuring communication connectivity. The collaborative decision-making mechanism of the entire scheme reflects the advantages of modern communication network layered architectures, achieving joint optimization of computing and communication resources through cloud-edge-device three-layer collaboration. This design concept fully aligns with the development trend of 5G / 6G network architectures, providing an effective technical solution for building green communication networks. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of an embodiment of the cloud-edge-device collaborative energy management system in this application. Detailed Implementation

[0016] This application provides an energy management system based on cloud-edge-device collaboration. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0017] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the cloud-edge-device collaborative energy management system in this application includes:

[0018] The data acquisition module is used to collect energy consumption data from communication terminal equipment through channel sensing processing to obtain the raw energy consumption dataset.

[0019] The prediction module is used to perform channel-aware prediction processing on the edge base station based on the original energy consumption dataset using the C-EPCE algorithm to obtain predicted energy consumption parameters.

[0020] The allocation module is used to input the predicted energy consumption parameters into the CSI-DPA algorithm to dynamically allocate the transmit power and obtain power control commands.

[0021] The sensing module is used to perform energy consumption sensing routing processing on the communication path according to the power control command and the NERA algorithm to obtain the optimal routing strategy;

[0022] The optimization module is used to perform multi-level optimization of the optimal routing strategy through a collaborative decision-making mechanism to obtain an energy management and control scheme.

[0023] Specifically, the acquisition module collects energy consumption data from communication terminal devices through channel sensing processing to obtain a raw energy consumption dataset. Specifically, the module first performs power consumption monitoring on the communication terminal devices, calculating the instantaneous power consumption value by measuring the device's current and voltage values ​​in real time, and then obtaining the device's power consumption value through time integration. Simultaneously, based on channel quality measurement, it performs state detection processing on the communication channel, calculating the Channel Quality Indicator (CQI) value, reflecting the channel transmission quality, by measuring the signal-to-noise ratio (SNR) and bit error rate (BER) parameters of the received signal. Then, it fuses the device power consumption value and the CQI value, merging the two types of data according to a preset weight ratio using a weighted average algorithm to obtain an energy consumption state vector containing power consumption and channel information. Furthermore, the module performs load statistics processing on service traffic, calculating service load parameters reflecting network busyness by statistically analyzing the number of data packets transmitted per unit time and the data transmission rate. Finally, it integrates the energy consumption state vector and service load parameters, arranging and combining them according to a time series to form a raw energy consumption dataset containing multi-dimensional energy consumption information.

[0024] The prediction module uses the C-EPCE algorithm to perform channel-aware prediction processing on edge base stations based on the original energy consumption dataset, obtaining predicted energy consumption parameters. The C-EPCE algorithm, or Channel-Aware Energy Consumption Predictive Encoder algorithm, is specifically designed for energy consumption prediction based on the channel variation characteristics of communication networks. This algorithm first inputs the original energy consumption dataset into a multi-layer neural network for feature extraction. The neural network performs nonlinear transformations on the input data through multiple hidden layers to extract feature patterns reflecting channel state patterns, obtaining a channel feature vector. Next, based on a channel fading model, the fading coefficient is calculated on the channel feature vector. By analyzing the amplitude attenuation law of the signal during propagation, the channel fading coefficient, which quantifies the degree of channel fading, is calculated. Then, a mapping relationship is established between channel quality and energy consumption requirements based on propagation loss characteristics. By establishing a mathematical relationship between channel quality indicators and energy consumption requirements, a channel quality-energy consumption mapping table describing the correspondence between the two is obtained. Subsequently, the channel fading coefficient and the channel quality-energy consumption mapping table are input into an autoencoder for decoding and prediction processing. The autoencoder reconstructs the encoded feature information into predicted energy consumption values ​​through the decoder, obtaining the energy consumption requirements for the future period. Finally, the energy consumption demand for future periods is encapsulated by encapsulating the prediction results in a standardized format to obtain predicted energy consumption parameters that include time labels and predicted values.

[0025] The allocation module inputs the predicted energy consumption parameters into the CSI-DPA algorithm to dynamically allocate transmit power and obtain power control commands. The CSI-DPA algorithm, or Dynamic Power Allocation Algorithm Based on Channel State Information, is specifically designed to dynamically adjust base station transmit power according to channel conditions. This algorithm first fuses the predicted energy consumption parameters and real-time channel state information. Through matrix operations, it correlates the predicted energy consumption demand with the currently measured channel parameters to obtain a power optimization input matrix that comprehensively reflects both energy consumption demand and channel conditions. Next, it constructs constraints on the power optimization input matrix based on the objective function of minimizing total energy consumption. By setting service quality constraints, power limit constraints, and interference constraints, it establishes an optimization model with the goal of minimizing total energy consumption, resulting in a power allocation constraint model. Then, it solves the power allocation constraint model using the Lagrange multiplier method based on channel gain characteristics. By calculating the channel gain matrix, it establishes the Lagrange function and solves the first-order optimality conditions. KKT conditions are used for constraint processing, and a converged power allocation solution is obtained through numerical iteration. After vector normalization, the optimal power allocation vector is obtained. Subsequently, the optimal power allocation vector and path loss model are processed for power correction calculation. Based on the loss characteristics of the signal propagation path, the theoretical power value is corrected for the actual environment to obtain the corrected power parameter considering propagation loss. Finally, the corrected power parameter is encapsulated into an instruction format, and the power value is formatted and encapsulated according to the communication protocol standard to obtain a power control instruction containing the power setting value and execution time.

[0026] The sensing module, based on power control commands, performs energy-aware routing processing on communication paths using the NERA algorithm to obtain the optimal routing strategy. The NERA algorithm, or network-wide energy-aware routing algorithm, is specifically designed to select the most energy-efficient communication path across the entire network. This algorithm first maps power control commands to network topology data, matching the power settings of each base station with network node location information to establish a network-wide energy consumption state matrix reflecting the overall network energy consumption distribution. Next, based on an improved Dijkstra algorithm framework, the network-wide energy consumption state matrix is ​​initialized for path search. The improved Dijkstra algorithm adds energy consumption considerations to the traditional shortest path search, identifying all potential communication paths by traversing network nodes and links, thus obtaining a candidate path set. Then, multi-objective routing cost calculation is performed on the candidate path set based on energy consumption weighting factors. By weighting and summing multiple indicators such as path length, latency, bandwidth, and energy consumption, the comprehensive cost of each path is calculated, resulting in a path energy cost vector that quantifies the merits of each path. The path energy cost vector is then input into the shortest path algorithm for path comparison and selection. By comparing the comprehensive cost of each path, the path with the lowest cost is selected as the optimal path, resulting in the optimal energy-efficient path. Finally, load balancing verification is performed on the optimal energy-efficient path to check its load and ensure sufficient path capacity without causing network congestion, thus obtaining the verified optimal routing strategy.

[0027] The optimization module performs multi-level optimization of the optimal routing strategy through a collaborative decision-making mechanism to obtain an energy management control scheme. The collaborative decision-making mechanism is the core mechanism for achieving collaborative operation of the cloud-edge-device three-layer architecture, responsible for coordinating decisions at each level and resolving potential conflicts. This mechanism first inputs the optimal routing strategy into a decision trigger for status monitoring. The trigger monitors network performance indicators and energy consumption levels to determine whether to initiate the collaborative decision-making process, generating a collaborative decision trigger signal when an anomaly is detected. Next, based on the collaborative decision trigger signal, the system performs data aggregation processing on the cloud-edge-device three-layer status information. By collecting operational status data from the cloud control center, edge base stations, and terminal devices, information scattered across various levels is centrally integrated to obtain a multi-level status dataset containing the overall network status. Then, based on priority rules, the system coordinates decision conflicts within the multi-level status dataset. When conflicts arise between decisions at different levels, the system coordinates according to preset priority rules, prioritizing the normal operation of critical services and core equipment, resulting in collaborative decision parameters that have undergone conflict resolution. Subsequently, the collaborative decision-making parameters are input into a multi-objective optimization algorithm for global optimization. By comprehensively considering multiple objectives such as energy consumption, performance, and reliability, the optimal control strategy for the entire network is obtained, resulting in a collaborative control instruction set containing specific execution instructions for each level. Finally, the collaborative control instruction set is layered and encapsulated, with control instructions at different levels encapsulated according to corresponding communication protocols and distributed to the cloud, edge, and terminal devices for execution, resulting in a complete energy management control scheme.

[0028] In one specific embodiment, the acquisition module is used for:

[0029] Power consumption monitoring is performed on communication terminal equipment to obtain the equipment power consumption value;

[0030] Based on channel quality measurement, state detection processing is performed on the communication channel to obtain the Channel Quality Indicator (CQI) value;

[0031] The device power consumption value and the channel quality indicator (CQI) value are fused together to obtain an energy consumption state vector.

[0032] Perform load statistics processing on business traffic to obtain business load parameters;

[0033] The energy consumption state vector and the service load parameters are integrated and processed to obtain the original energy consumption dataset.

[0034] Specifically, the power consumption monitoring and processing in the acquisition module involves installing current and voltage sensors at the power input of the communication terminal equipment to collect the instantaneous current and voltage values ​​of the equipment in real time. The current sensor, based on the Hall effect principle, converts the current flowing through the conductor into a corresponding voltage signal, while the voltage sensor directly measures the voltage difference across the equipment. The system multiplies the collected current and voltage values ​​to calculate the instantaneous power, and then accumulates the power consumption within a certain time window through time integration to obtain the equipment power consumption value reflecting the actual energy consumption of the equipment. Channel quality measurement is based on the signal quality assessment mechanism in wireless communication, measuring the strength and quality parameters of the received signal through the RF front-end circuit of the receiver. The system measures the Received Signal Strength Indicator (RSSI) value, calculates the signal-to-noise power ratio (SNR) to obtain the signal-to-noise ratio (SNR), and simultaneously calculates the bit error rate (BER) based on the number of erroneous bits during data transmission. These parameters are input into the channel quality assessment algorithm, and a standardized Channel Quality Indicator (CQI) value is obtained through weighted calculation. This value typically ranges from 0 to 15, with higher values ​​indicating better channel quality.

[0035] The data fusion processing employs a weighted average algorithm to correlate device power consumption and Channel Quality Indicator (CQI) values. The system first normalizes both types of data: power consumption is standardized according to the device's rated power, and CQI values ​​are normalized to their maximum value of 15. Then, a power consumption weighting coefficient of 0.6 and a channel weighting coefficient of 0.4 are set, and a comprehensive evaluation value is obtained through weighted summation. Finally, this comprehensive evaluation value is combined with auxiliary information such as timestamps and device identifiers to form an energy consumption state vector containing multi-dimensional information.

[0036] Traffic monitoring is achieved through a packet counter at the network interface layer. The system counts the number of packets passing through the network interface within a specified time window, recording the size and transmission timestamp of each packet. Throughput is calculated by measuring the total data transmission per unit time, and the service arrival rate is obtained by calculating the distribution of packet arrival time intervals. Simultaneously, the proportional distribution of services with different priorities is also statistically analyzed. These statistical results are combined to calculate a service load parameter reflecting network busyness. This parameter is typically represented by a value between 0 and 1, where 0 indicates idle status and 1 indicates full load.

[0037] The data integration and processing organizes the energy consumption status vector and service load parameters according to a time series. The system establishes a data structure containing timestamp, device identifier, energy consumption status vector, and service load parameter fields. The collected data is stored sequentially in an array structure according to time order, forming a time series dataset. An indexing mechanism is also established to facilitate subsequent data querying and processing based on time range or device type. The final result is a raw energy consumption dataset containing complete energy consumption and network status information, which serves as the input data source for subsequent prediction and optimization algorithms.

[0038] In one specific embodiment, the prediction module is used for:

[0039] The original energy consumption dataset is input into a multi-layer neural network for feature extraction to obtain a channel feature vector.

[0040] The fading coefficients are calculated based on the channel fading model to obtain the channel fading coefficients.

[0041] Based on the propagation loss characteristics, a mapping relationship is established for channel quality to obtain a channel quality energy consumption mapping table.

[0042] The channel fading coefficient and the channel quality energy consumption mapping table are input into the autoencoder for decoding and prediction processing to obtain the energy consumption demand for future periods.

[0043] The predicted energy consumption parameters are obtained by encapsulating the energy consumption demand for the future period.

[0044] Specifically, the multi-layer neural network in the prediction module adopts a three-layer fully connected neural network structure, including an input layer, a hidden layer, and an output layer. The input layer receives multi-dimensional features from the original energy consumption dataset, including device power consumption values ​​(unit: watts), channel quality indicator values ​​(range 0 to 15), service load parameters (unit: megabits per second), and an energy consumption state vector calculated by the acquisition module through weighted summation. The energy consumption state vector is calculated as follows: first, the device power consumption value is normalized according to the device's rated power to obtain a power consumption normalized value; the channel quality indicator value is normalized according to the maximum value of 15 to obtain a channel normalized value; then, a weighted average algorithm is used, with the power consumption weight coefficient set to 0.6 and the channel weight coefficient set to 0.4, and a comprehensive evaluation value is obtained through weighted summation as the energy consumption state vector. The complete parameters of the input layer include four dimensions: device power consumption value, channel quality indicator value, service load parameters, and energy consumption state vector, forming a 4-dimensional input vector with a fixed length of 4. The hidden layer employs a modified linear unit activation function for nonlinear transformation. The activation function is applied after a linear combination of the weight matrix and bias vector to extract nonlinear feature patterns from the data. The network trains the weight parameters using backpropagation and minimizes the mean squared error loss function using gradient descent. The trained neural network can identify temporal patterns and spatial distribution characteristics in energy consumption data. When the device power consumption value or channel quality indicator value in the input parameters changes, the corresponding energy consumption state vector also changes accordingly. The network outputs a corresponding prediction result based on the learned weight parameter mapping relationship, ensuring that changes in input parameters are accurately reflected in subsequent outputs. The output layer generates a fixed-dimensional feature vector containing key information reflecting the channel state change patterns, forming the channel feature vector.

[0045] The channel fading model is based on the physical propagation characteristics of wireless channels and uses the Rayleigh fading model to describe signal fading phenomena in multipath propagation environments. The system extracts the amplitude component from the channel feature vector and calculates fading parameters by analyzing the statistical distribution characteristics of the signal envelope. The Rayleigh fading model assumes that the channel gain follows a Rayleigh distribution, and the channel gain is obtained by calculating the ratio of received signal power to transmitted signal power. The system statistically analyzes the changes in channel gain within a certain time window, calculates its mean and variance, and solves for the scale parameter of the Rayleigh distribution using the maximum likelihood estimation method. This scale parameter reflects the severity of channel fading; a smaller value indicates more severe fading. Finally, a channel fading coefficient is obtained, ranging from 0 to 1, to quantify the degree of channel fading and describe the degree of channel quality degradation. The propagation loss characteristic model comprehensively considers the combined effects of free-space propagation loss and the channel fading coefficient. By using the channel fading coefficient as a correction factor in the propagation loss calculation, a mathematical mapping relationship between channel quality indicators and energy consumption requirements is established. Specifically, the system calculates the actual path loss value based on the propagation loss formula and the channel fading coefficient. This path loss value directly affects the transmit power required to maintain communication quality, thereby determining the corresponding energy consumption requirement. When the channel fading coefficient is close to 1, it indicates good channel quality, corresponding to lower path loss and energy consumption requirements; when the channel fading coefficient is close to 0, it indicates severely degraded channel quality, corresponding to higher path loss and energy consumption requirements. The system uses a piecewise linear interpolation method to establish a correspondence table between channel quality indication values ​​and energy consumption requirements. The channel quality indication value is divided into 16 intervals (corresponding to values ​​from 0 to 15). Each interval, combined with the corresponding channel fading coefficient range, determines a specific energy consumption requirement value. The mapping table includes four fields: channel quality indication value, corresponding channel fading coefficient range, calculated path loss value, and corresponding energy consumption requirement power value. By calibrating the interpolation parameters and fading coefficient correction coefficients using actual measurement data, a channel quality-energy consumption mapping table that accurately reflects the relationship between channel quality, fading characteristics, and energy consumption requirements is obtained. This mapping table, along with the real-time calculated channel fading coefficient, serves as an input parameter in subsequent encoder processing to achieve accurate energy consumption prediction based on the current channel state.

[0046] The autoencoder employs an encoder-decoder architecture for energy consumption prediction. The encoder takes channel fading coefficients and a channel quality-energy-consumption mapping table as input, compressing the high-dimensional input data into a low-dimensional latent representation through a multi-layer fully connected network. The decoder reconstructs the latent representation into a predictive output, generating predicted energy demand values ​​for multiple future time steps. The network uses Long Short-Term Memory (LSTM) units to process time-series data, selectively memorizing and forgetting historical information through a gating mechanism. The input gate controls the writing of new information, the forget gate controls the retention of historical information, and the output gate controls the output of the current state. Through this mechanism, the network can learn the long-term dependencies and periodic change patterns of energy consumption data. The training process uses mean squared error as the loss function, and the network parameters are optimized using a time backpropagation algorithm. After training, the decoder outputs predicted energy demand values ​​for future periods, including power demand predictions for multiple time points.

[0047] The parameter encapsulation process organizes future energy consumption demand according to a standardized data format. The system establishes a data structure for the prediction results, including fields such as prediction timestamp, prediction time window, predicted energy consumption value, and confidence interval. The timestamp identifies the start time of the prediction, the time window specifies the time range for the prediction, the predicted energy consumption value is the specific predicted power demand, and the confidence interval reflects the reliability of the prediction results. The system also adds metadata information, including the prediction model version, input data source, prediction accuracy, and other auxiliary information. This information is serialized in JSON format to generate a standardized data packet. Finally, the predicted energy consumption parameters, containing complete prediction information and metadata, are obtained and serve as input data for subsequent power allocation algorithms.

[0048] In one specific embodiment, the allocation module is used for:

[0049] The fusion unit is used to perform data fusion processing on the predicted energy consumption parameters and real-time channel state information to obtain a power optimization input matrix.

[0050] The construction unit is used to perform constraint condition construction processing on the power optimization input matrix based on the objective function of minimizing total energy consumption, so as to obtain a power allocation constraint model;

[0051] The solution unit is used to solve the power allocation constraint model using the Lagrange multiplier method based on the channel gain characteristics to obtain the optimal power allocation vector.

[0052] The correction unit is used to perform power correction calculations on the optimal power allocation vector and the path loss model to obtain the corrected power parameters.

[0053] The encapsulation unit is used to encapsulate the modified power parameters into an instruction format to obtain the power control instruction.

[0054] Specifically, the data fusion processing in the fusion unit uses matrix operations to correlate and calculate the predicted energy consumption parameters and real-time channel state information. The predicted energy consumption parameters include power demand forecasts for multiple future time points, represented in vector form. = [ , ,..., ], where each element represents the predicted power demand at the corresponding time point. Real-time channel state information includes parameters such as channel gain, signal-to-noise ratio, and multipath delay for each base station, organized as a channel state matrix H = [ , ,..., This matrix is ​​an M×N dimensional matrix, where M represents the number of base stations and N represents the number of channel parameter types. The elements in the matrix... This represents the nth type of channel parameter value for the m-th base station. The value of m ranges from 1 to M (M is the total number of base stations in the system, typically 10 to 100), and the value of n ranges from 1 to N (N is the total number of channel parameter types, including four types: channel gain, signal-to-noise ratio, multipath delay, and Doppler shift; therefore, N equals 4). The rows of the matrix represent the numbers of different base stations, and the columns represent different types of channel parameters. The fusion process first performs time alignment on the two types of data to ensure that the predicted data and real-time data are consistent in the time dimension. Then, a weighted fusion algorithm is used, and the fusion result F = α is calculated by setting the prediction weight α and the real-time weight β. + β Finally, the fusion results are reorganized into a matrix according to the base station number and time series to form a power optimization input matrix that includes power demand prediction and channel state. This matrix provides complete input data for subsequent optimization algorithms.

[0055] The building blocks are constructed using a mathematical optimization model based on the objective function of minimizing total energy consumption. The objective function is defined as minimizing the total energy consumption of the entire network, expressed as: ,in Let represent the transmit power of the i-th base station, and N be the total number of base stations. Constraints include three categories: Quality of Service (QoS) constraints, power limitation constraints, and interference constraints. QoS constraints ensure that the signal reception quality for each user meets the minimum requirements, denoted as [equation missing]. ≥ ,in Let be the signal-to-interference-plus-noise ratio (SIR) for the i-th user.

[0056] in The calculation method is as follows:

[0057]

[0058] in This represents the base station transmit power providing service to the i-th user. This represents the channel gain from the serving base station to the i-th user. This represents the transmit power of the j-th interfering base station. This represents the channel gain from the j-th interfering base station to the i-th user. This represents the noise power, and the formula is used to calculate the ratio of the useful signal power received by the i-th user to the sum of the interfering signal power and the noise power. The power limit constraint stipulates that the transmit power of each base station cannot exceed the maximum value, expressed as 0 ≤ ≤ Interference constraints limit the level of mutual interference between base stations. The system converts these constraints into standard linear inequality constraints, constructs a complete optimization model containing the objective function and constraints, and obtains the power allocation constraint model.

[0059] The solution unit uses the Lagrange multiplier method to numerically solve the power allocation constraint model. First, the channel gain matrix G is calculated based on channel measurement data, where the elements... Let L(P, λ, μ) represent the channel gain from base station j to user i. The channel gain is calculated using the path loss formula, considering the combined effects of range fading, shadowing fading, and fast fading. Then, the Lagrangian function L(P, λ, μ) = Σ is constructed. +Σ ( - ) + Σ ( - ), where λ and μ are Lagrange multiplier vectors. Taking the partial derivatives of the Lagrange function, we obtain the first-order optimality condition ∂L / ∂. = 0, forming a system of equations containing power and multiplier variables. Inequality constraints are addressed using KKT conditions, including complementary relaxation conditions. ( - ) = 0 and nonnegativity condition ≥ 0, ≥ 0. Numerical iterative solutions are performed using the interior-point method, updating variable values ​​using Newton's method until convergence. Convergence is determined based on the gradient norm being less than a preset threshold ε, typically set to 0. The power allocation solution that satisfies all constraints is obtained, and the optimal power allocation vector is obtained after vector normalization.

[0060] The correction unit combines the optimal power allocation vector with the path loss model for real-world environment correction. The path loss model uses the Okumura-Hata model to describe the signal attenuation characteristics in the actual propagation environment. The loss calculation formula is PL = A + Blog(d) + C, where A is the intercept parameter, B is the distance attenuation coefficient, C is the environmental correction factor, and d is the propagation distance. The system calculates the actual propagation distance based on the geographical location information of the base station and the user, and determines the model parameter values ​​by combining environmental parameters such as terrain and building density. For the power allocation value obtained from theoretical optimization, correction is performed through path loss compensation, calculated using the formula Pcorrected = Poptimal + PLcompensation, where PLcompensation is the path loss compensation value, Pcorrected is the corrected power value, and Poptimal is the optimal power value. It should be noted that the path loss compensation value is the path loss power compensation amount calculated based on the Okumura-Hata model, used to compensate for signal attenuation loss in the actual propagation environment. The corrected power value is the final transmit power setting value after path loss compensation, and the optimal power value is the theoretically optimal transmit power allocation value obtained by solving the Lagrange multiplier method. The path loss compensation value generally ranges from 0.5 to 5 watts. When the propagation distance is within 1 kilometer and it is a line-of-sight propagation environment, the compensation value is usually 0.5 to 1 watt. When the propagation distance is 1 to 5 kilometers and there are building obstructions, the compensation value is 1 to 3 watts. When the propagation distance exceeds 5 kilometers or it is in a dense urban environment, the compensation value can reach 3 to 5 watts. This correction formula ensures that the theoretically optimized power allocation value can adapt to the loss effects of the actual propagation environment.

[0061] Simultaneously considering the impact of hardware factors such as antenna gain and feeder loss, the power value is further adjusted. The correction process also includes power quantization, which quantizes continuous power values ​​into discrete power levels supported by the hardware. The final result is a corrected power parameter that takes into account the actual propagation environment and hardware limitations.

[0062] The encapsulation unit encapsulates the power parameter correction instructions into a standardized instruction format. The system uses an XML-based instruction format, comprising a header, parameter body, and checksum. The header includes metadata such as instruction type identifier, version number, timestamp, and source device identifier. The parameter body contains specific power control parameters, organized in key-value pairs, including fields such as base station identifier, power setpoint, execution time, and duration. The base station identifier uses a unique device ID, the power setpoint is expressed in dBm, the execution time uses UTC time format, and the duration is in seconds. The checksum includes a CRC checksum and a digital signature to ensure the integrity and security of the instruction transmission. The CRC checksum is obtained by performing cyclic redundancy check on the instruction content, and the digital signature uses the RSA algorithm to encrypt and sign the instruction. After encapsulation, the instruction is verified to check the correctness of the format and the rationality of the parameters. Finally, a power control instruction conforming to the communication protocol standard is obtained, which is then sent to the corresponding base station equipment via the network to perform the power adjustment operation.

[0063] In one specific embodiment, the solving unit is used for:

[0064] The channel gain matrix is ​​obtained by performing matrix calculations on the channel measurement data.

[0065] The channel gain matrix is ​​input into the Lagrange function for partial derivative calculation to obtain a set of first-order optimality condition equations.

[0066] The first-order optimality condition equations are constrained according to the KKT conditions to obtain the Lagrange multiplier solution set.

[0067] Substituting the Lagrange multiplier solution set into the power allocation formula and performing numerical iterative calculations yields a convergent power solution, where the power allocation formula is:

[0068]

[0069] in, This represents the optimal transmit power of base station i; This indicates the minimum signal-to-interference-plus-noise ratio requirement; Indicates noise power; Represents the Lagrange multiplier corresponding to base station i; This represents the channel gain from base station i to its serving users; This represents the interference channel gain of base station j to the users served by base station i; This represents the transmit power of base station j;

[0070] The converged power solution is vector normalized to obtain the optimal power allocation vector.

[0071] Specifically, the channel gain matrix calculation in the solution unit is based on channel measurement data between each base station and user terminal. Channel measurement data includes parameters such as Received Signal Strength Indication (RSSI), Reference Received Power (RSRP), and Reference Received Quality (RSRQ). The system first preprocesses this measurement data, removing outliers and noise interference, and smoothing data fluctuations using a moving average filter. Then, path loss is calculated based on the wireless propagation model, considering the combined effects of free space loss, multipath fading, and shadowing effects. Free space loss is calculated based on transmit / receive distance and operating frequency; multipath fading is described using Rayleigh or Rice distribution models; and shadowing effects are modeled using a log-normal distribution. The path loss values ​​are combined with the measured power to calculate the channel gain values ​​for each link. Finally, the channel gains from all base stations to users are organized in matrix form, with row indices representing user numbers, column indices representing base station numbers, and matrix elements representing the channel gain values ​​of the corresponding links, forming a complete channel gain matrix.

[0072] The calculation of partial derivatives of the Lagrange function involves mathematical processing of the objective function and constraints. The system inputs the channel gain matrix as a known parameter into the Lagrange function, which contains the original power optimization objective function and a linear combination of all constraints. Partial derivatives are calculated for each power variable in the Lagrange function to obtain the gradient vector with respect to power allocation. Partial derivatives are calculated for each Lagrange multiplier variable to recover the original constraints. The partial derivative calculation process requires handling the nonlinear expression of the signal-to-interference-plus-noise ratio (SINR), using the chain rule to decompose the derivative of the composite function into a product of the derivatives of simpler functions. The system employs automatic differentiation technology for accurate gradient calculation, avoiding the accuracy loss caused by numerical differentiation. The calculated partial derivative equations are grouped according to variable type, forming a system of first-order optimality condition equations containing power variable equations and multiplier variable equations.

[0073] KKT condition processing transforms inequality-constrained optimization problems into equality-constrained systems. The system applies the four components of KKT conditions to the first-order optimality condition equations: stationarity conditions, primal feasibility conditions, dual feasibility conditions, and complementary relaxation conditions. The stationarity condition requires the gradient of the Lagrange function to be zero, a necessary condition to ensure the solution is at the optimum. The primal feasibility condition ensures that the power allocation solution satisfies all primal constraints, including power limits and quality of service requirements. The dual feasibility condition requires all Lagrange multipliers to be non-negative, reflecting the unidirectional nature of inequality constraints. The complementary relaxation condition requires the product of the constraint and its corresponding multiplier to be zero, ensuring that only the active constraint corresponds to a non-zero multiplier. The system processes these conditions numerically, using the interior-point method to transform inequality constraints into equality constraints with a penalty function. During the solution process, the penalty function parameters are dynamically adjusted to gradually approximate the optimal solution of the original problem. Finally, a Lagrange multiplier solution set satisfying all KKT conditions is obtained.

[0074] Numerical iterative computation employs Newton's method to solve the nonlinear equation system. The system substitutes the Lagrange multiplier solution set into the optimality condition equations for power allocation, forming a nonlinear equation system about the power variables. Newton's method iteratively updates the system by linearizing the equation system near the current point and calculating the search direction and step size. Each iteration requires calculating the Jacobian matrix of the equation system, i.e., the partial derivative matrix of all equations with respect to all variables. The system uses sparse matrix techniques to handle large-scale problems, leveraging the sparse structure of the equation system to improve computational efficiency. Convergence criteria are set during iteration, including multiple indicators such as the change in function value, gradient norm, and change in variables. The algorithm stops iterating when all convergence indicators simultaneously meet preset thresholds. To avoid getting trapped in local optima, the system adopts a multi-starting-point strategy, running the algorithm from different initial values ​​and comparing the results. The iterative process also includes a step-size control mechanism, using backtracking to ensure that the objective function value decreases in each iteration. Finally, a numerical solution for power allocation that satisfies the convergence conditions is obtained, i.e., the convergent power solution.

[0075] Vector normalization ensures that the optimal power allocation vector meets the constraints of the actual system. The system first checks for negative values ​​or elements exceeding power limits in the converged power solution, and performs boundary projection processing on outliers. Negative power values ​​are directly set to zero, and values ​​exceeding the maximum power limit are truncated to the maximum allowable power. Then, the norm of the power vector is calculated, including different metrics such as the first norm, second norm, and infinite norm. Based on the system's total power budget constraint, an appropriate normalization method is selected to scale the power vector. The normalization process maintains the relative proportion of power allocation for each base station, ensuring that the optimality obtained by the optimization algorithm is not compromised. Finally, the normalized power values ​​are quantized, mapping continuous power values ​​to discrete power levels supported by the hardware. The quantization process uses nearest neighbor rounding to select the discrete level closest to the continuous value. After normalization and quantization, an optimal power allocation vector that satisfies both mathematical optimality and practical constraints is obtained.

[0076] In one specific embodiment, the sensing module is used for:

[0077] The power control commands and network topology data are correlated and mapped to obtain the energy consumption status matrix of the entire network.

[0078] Based on the improved Dijkstra algorithm framework, the entire network energy consumption state matrix is ​​initialized by path search to obtain a set of candidate paths.

[0079] The candidate path set is processed by multi-objective routing cost calculation based on the energy consumption weight factor to obtain the path energy consumption cost vector.

[0080] The path energy consumption overhead vector is input into the shortest path algorithm for path comparison and selection to obtain the optimal energy consumption path.

[0081] The optimal energy consumption path is subjected to load balancing verification to obtain the optimal routing strategy.

[0082] Specifically, the association mapping process in the perception module spatially and logically associates power control commands with network topology data. Power control commands include information such as the power setpoints, execution times, and durations of each base station, while network topology data includes network structure information such as base station location coordinates, coverage areas, adjacency relationships, and link capacity. The system first establishes a mapping table between base station identifiers and geographical locations, matching the base station IDs in the power control commands with the node coordinates in the topology data. Then, it calculates the actual coverage area based on the power setpoints and coverage radii of each base station, and uses a Geographic Information System (GIS) algorithm to determine the coverage overlap between adjacent base stations. Next, it analyzes the signal interference levels between base stations, calculates the co-channel interference intensity based on the power setpoints and propagation distances, and establishes an interference relationship matrix between base stations. Finally, it organizes the power information, location information, coverage information, and interference information according to network nodes and links to construct a network topology map containing energy consumption status, forming a network-wide energy consumption status matrix reflecting the energy consumption distribution and connectivity relationships of the entire network. The improved Dijkstra algorithm framework adds energy consumption perception capabilities to the traditional shortest path algorithm. The system first performs a data structure transformation on the network energy consumption state matrix, converting the matrix-form network topology into a graph structure representation of an adjacency list or adjacency matrix. The initialization process includes setting the source and destination nodes, creating a distance array to record the shortest distances from the source node to each node, and creating a predecessor array to record the previous hop node of the optimal path. Unlike the traditional Dijkstra algorithm, the improved algorithm considers not only hop count or physical distance but also the energy consumption state of each node along the path. The algorithm maintains a priority queue, sorting nodes to be visited according to their comprehensive cost, and prioritizing the node with the lowest cost for expansion. The path search process employs a breadth-first search strategy, expanding outwards layer by layer from the source node, updating the shortest distances to each node through relaxation operations. When a shorter path to a node is found, the distance value and predecessor node information of that node are updated. During the search, all possible path combinations are recorded, forming a candidate path set from the source node to the destination node.

[0083] Multi-target routing cost calculation comprehensively considers multiple performance indicators such as latency, bandwidth, packet loss rate, and energy consumption. The system assigns a weight coefficient to each performance indicator, with the energy consumption weight factor typically set between 0.4 and 0.6 to reflect the importance of energy optimization. For each path in the candidate path set, the values ​​of each performance indicator are calculated separately. Latency calculation includes the sum of propagation latency, queuing latency, and processing latency. Propagation latency is calculated based on physical distance and signal propagation speed, queuing latency is estimated based on node load and buffer status, and processing latency is determined based on the node's processing capacity. Bandwidth calculation selects the bottleneck link bandwidth on the path as the available bandwidth for the entire path. Packet loss rate is obtained by accumulating the packet loss rates of each node on the path. Energy consumption calculation is estimated based on the power consumption values ​​of each node on the path and the data transmission volume. After normalizing each indicator, a weighted sum is used to obtain the comprehensive cost value of each path, forming a path energy consumption cost vector that quantifies the path's quality.

[0084] The shortest path algorithm employs an improved A* algorithm for path comparison and selection. The system uses the path energy cost vector as input to a heuristic function, and the A* algorithm combines actual cost and heuristic estimation for path search. Actual cost represents the known optimal cost from the starting node to the current node, while heuristic estimation represents the estimated cost from the current node to the target node. The algorithm maintains open and closed lists; the open list contains nodes to be expanded, and the closed list contains processed nodes. In each iteration, the node with the minimum overall cost in the open list is selected for expansion, and its neighboring nodes are added to the open list or their cost values ​​are updated. The heuristic function uses a weighted combination of energy distance and geometric distance to ensure the algorithm's acceptability and consistency. During path search, the parent node pointers of nodes are dynamically updated to record the composition of the optimal path. When the algorithm reaches the target node, the complete optimal path is reconstructed by backtracking the parent node pointers. The comparison and selection process also includes path deduplication, eliminating redundant paths that pass through the same set of nodes. Finally, the optimal energy-efficient path with the minimum overall cost is obtained.

[0085] Load balancing verification ensures the actual availability and network stability of the selected optimal path. The system first checks the current load status of each node on the optimal energy-efficient path, including metrics such as CPU utilization, memory utilization, buffer occupancy, and link utilization. By comparing these metrics with preset load thresholds, it determines whether there are overloaded nodes or links on the path. If overload is detected, the system initiates a load balancing mechanism, reallocating some service traffic to less loaded alternative paths. The load balancing algorithm employs a weighted round-robin strategy, dynamically adjusting weights based on node processing capacity and current load status. The verification process also includes path reachability testing, verifying the connectivity and transmission quality of each link on the path by sending probe packets. Test results include parameters such as round-trip latency, packet loss rate, and bandwidth utilization, used to evaluate the actual performance of the path. If the verification results show that the path performance does not meet requirements, the system selects a suboptimal path from the candidate path set for re-verification. The verification process also considers dynamic network changes, setting path update trigger conditions to automatically recalculate the routing strategy when the network status changes significantly. Finally, the optimal routing strategy, verified by load balancing and meeting performance requirements, is obtained.

[0086] In one specific embodiment, the optimization module is used to:

[0087] The optimal routing strategy is input into the decision trigger for state monitoring processing to obtain the collaborative decision trigger signal;

[0088] Based on the collaborative decision-making trigger signal, the cloud-edge-device three-layer state information is aggregated and processed to obtain a multi-level state dataset.

[0089] Based on priority rules, the multi-level state dataset is subjected to decision conflict coordination processing to obtain collaborative decision parameters;

[0090] The collaborative decision parameters are input into a multi-objective optimization algorithm for global optimization to obtain a collaborative control instruction set.

[0091] The energy management control scheme is obtained by performing layered distribution and encapsulation processing on the collaborative control instruction set.

[0092] Specifically, the decision trigger in the optimization module determines whether to initiate a collaborative decision-making process by monitoring network performance metrics and energy consumption levels in real time. The system sets multiple monitoring thresholds, including key indicators such as network latency threshold, throughput threshold, energy consumption growth rate threshold, and load balancing threshold. After receiving the optimal routing strategy, the decision trigger compares and analyzes it with the current network operating state. The system calculates the expected performance metrics after the routing strategy is implemented and evaluates the impact of strategy execution on overall network performance through simulation models. When the predicted performance change exceeds a preset threshold range, the trigger generates an anomaly detection signal. Anomaly detection includes three dimensions: performance degradation detection, energy consumption anomaly detection, and load imbalance detection. Performance degradation detection is determined by comparing the deviation of current latency and throughput from historical benchmark values; energy consumption anomaly detection is identified by analyzing the rate of change of power consumption per unit time; and load imbalance detection is measured by calculating the standard deviation of the load on each node. When the detection result of any dimension triggers an early warning condition, the system generates a collaborative decision trigger signal containing the anomaly type, severity, and information about the involved nodes.

[0093] Data aggregation processing centrally collects and integrates state information from the cloud-edge-device three-layer architecture based on collaborative decision-making trigger signals. State information at the cloud control center layer includes macro-level data such as network-wide topology changes, resource allocation strategies, historical performance statistics, and long-term optimization goals. State information at the edge base station layer includes meso-level data such as local load status, neighbor node information, channel quality parameters, and power control execution results. State information at the terminal device layer includes micro-level data such as device energy consumption status, service request types, mobility information, and service quality experience. The system employs a distributed data collection mechanism, with nodes at each level periodically reporting state information to the aggregation node. The data aggregation process includes three steps: time synchronization, format standardization, and integrity verification. Time synchronization ensures that data from different levels has a unified time reference; format standardization converts state information in different formats into a unified data structure; and integrity verification verifies the accuracy and completeness of data transmission. The aggregated data is categorized and organized according to hierarchical affiliation, time series, and data type, forming a multi-level state dataset containing network-wide state information.

[0094] The decision conflict coordination and handling system resolves potential conflicts between decisions at different levels based on predefined priority rules. The system establishes a three-tier priority system: high priority corresponds to decision requirements for critical services and core equipment; medium priority corresponds to performance assurance requirements for general services; and low priority corresponds to non-critical optimization and improvement requirements. Priority rules include three dimensions: service type rules, equipment importance rules, and resource scarcity rules. Service type rules allocate priorities based on the real-time requirements and importance of the service, with emergency communication and security-related services having the highest priority. Equipment importance rules determine priorities based on the criticality of equipment in the network, prioritizing the fulfillment of decision requirements for core routing nodes and key base stations. Resource scarcity rules prioritize the efficient utilization of scarce resources in resource contention situations. The conflict coordination algorithm uses the analytic hierarchy process (AHP) to calculate the weights of decision requirements in multi-level state datasets, determining the relative importance of each decision requirement through the construction of a judgment matrix and consistency checks. When a decision conflict is detected, the system reallocates resources according to priority weights, ensuring resource guarantees for high-priority decisions and making appropriate compromises for low-priority decisions. The coordination process also includes dynamic adjustments to conflict resolution strategies, optimizing priority rules based on network state changes and decision execution effects. Ultimately, collaborative decision parameters that eliminate conflicts and meet priority requirements are obtained.

[0095] The global optimization solution employs a multi-objective genetic algorithm to perform network-wide optimization calculations on the collaborative decision parameters. This algorithm simultaneously considers three optimization objectives: minimizing energy consumption, maximizing performance, and ensuring reliability. It seeks the optimal balance between these objectives through a Pareto-optimal solution set. The algorithm initialization process generates a random population of decision variables, with each individual representing a possible combination of control strategies. A fitness function comprehensively evaluates each individual's performance across multiple objective dimensions, and a weighted summation method is used to transform the multi-objective problem into a single-objective optimization. Genetic operations include three basic steps: selection, crossover, and mutation. The selection operation uses a tournament selection strategy, choosing superior individuals for the next generation based on their fitness values. The crossover operation generates new individual combinations through single-point or multi-point crossover, maintaining population diversity. The mutation operation randomly perturbs some genes of individuals to prevent the algorithm from getting trapped in local optima. During iteration, algorithm parameters, including key parameters such as crossover probability, mutation probability, and population size, are dynamically adjusted. The convergence criterion is based on the improvement margin of the optimal solution within consecutive generations; the algorithm stops when the improvement margin is less than a preset threshold. The optimization process also includes a constraint handling mechanism to ensure that the generated control strategy meets the physical constraints and performance requirements of the actual system. The final result is a coordinated control instruction set containing specific control instructions for each level.

[0096] The layered encapsulation and processing of collaborative control command sets categorizes, encapsulates, and transmits them according to a three-layer cloud-edge-device architecture. The system first categorizes the collaborative control command sets based on their execution level. Cloud commands mainly include macro-level control commands such as network-wide resource scheduling, topology optimization, and policy updates; edge commands mainly include meso-level control commands such as power adjustment, route switching, and load balancing; and terminal commands mainly include micro-level control commands such as service migration, sleep / wake-up, and parameter configuration. The encapsulation process employs a layered protocol stack structure, selecting appropriate communication protocols and encapsulation formats for commands at different levels. Cloud commands use TCP to ensure reliable transmission, edge commands use UDP to ensure real-time performance, and terminal commands select either WiFi or cellular network protocols based on device type. Command encapsulation includes three processing steps: security encryption, integrity verification, and priority identification. Security encryption uses the AES algorithm to encrypt and protect sensitive control commands; integrity verification uses Message Authentication Code (MAC) to ensure commands are not tampered with during transmission; and priority identification assigns execution priorities to commands to ensure timely execution of critical commands. The transmission process employs a reliable transmission mechanism, including acknowledgment, timeout retransmission, and error recovery safeguards. The system also establishes a command execution monitoring mechanism to track the execution status and effect feedback of commands at each level. This ultimately results in an energy management control scheme that includes a complete control strategy and execution mechanism.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An energy management system based on cloud-edge-device collaboration, characterized in that, The system includes: The data acquisition module is used to collect energy consumption data from communication terminal equipment through channel sensing processing to obtain the raw energy consumption dataset. The prediction module is used to perform channel-aware prediction processing on edge base stations based on the original energy consumption dataset using the C-EPCE algorithm to obtain predicted energy consumption parameters. Specifically, it is used to: input the original energy consumption dataset into a multi-layer neural network for feature extraction processing to obtain a channel feature vector; calculate the fading coefficient of the channel feature vector based on a channel fading model to obtain a channel fading coefficient; establish a mapping relationship between channel quality and propagation loss characteristics to obtain a channel quality energy consumption mapping table; input the channel fading coefficient and the channel quality energy consumption mapping table into an autoencoder for decoding and prediction processing to obtain the energy consumption demand for the future period; and encapsulate the energy consumption demand for the future period to obtain the predicted energy consumption parameters. The allocation module, used to input the predicted energy consumption parameters into the CSI-DPA algorithm, dynamically allocates the transmit power to obtain a power control command, specifically includes: a fusion unit, used to perform data fusion processing on the predicted energy consumption parameters and real-time channel state information to obtain a power optimization input matrix; a construction unit, used to perform constraint condition construction processing on the power optimization input matrix based on the objective function of minimizing total energy consumption to obtain a power allocation constraint model; a solution unit, used to perform Lagrange multiplier method solution processing on the power allocation constraint model according to the channel gain characteristics to obtain an optimal power allocation vector; a correction unit, used to perform power correction calculation processing on the optimal power allocation vector and the path loss model to obtain a corrected power parameter; and an encapsulation unit, used to encapsulate the corrected power parameter into an instruction format to obtain the power control command. The perception module is used to perform energy-aware routing processing on the communication path according to the power control command and the NERA algorithm to obtain the optimal routing strategy. Specifically, it is used to: perform association mapping processing on the power control command and network topology data to obtain a network-wide energy consumption state matrix; perform path search initialization processing on the network-wide energy consumption state matrix based on the improved Dijkstra algorithm framework to obtain a candidate path set; perform multi-objective routing cost calculation processing on the candidate path set according to the energy consumption weight factor to obtain a path energy consumption cost vector; input the path energy consumption cost vector into the shortest path algorithm for path comparison and selection processing to obtain the optimal energy consumption path; and perform load balancing verification processing on the optimal energy consumption path to obtain the optimal routing strategy. The optimization module is used to perform multi-level optimization of the optimal routing strategy through a collaborative decision-making mechanism to obtain an energy management and control scheme.

2. The cloud-edge-device collaborative energy management system according to claim 1, characterized in that, The acquisition module is used for: Power consumption monitoring is performed on communication terminal equipment to obtain the equipment power consumption value; Based on channel quality measurement, state detection processing is performed on the communication channel to obtain the Channel Quality Indicator (CQI) value; The device power consumption value and the channel quality indicator (CQI) value are fused together to obtain an energy consumption state vector. Perform load statistics processing on business traffic to obtain business load parameters; The energy consumption state vector and the service load parameters are integrated and processed to obtain the original energy consumption dataset.

3. The cloud-edge-device collaborative energy management system according to claim 1, characterized in that, The solving unit is used for: The channel gain matrix is ​​obtained by performing matrix calculations on the channel measurement data. The channel gain matrix is ​​input into the Lagrange function for partial derivative calculation to obtain a set of first-order optimality condition equations. The first-order optimality condition equations are constrained according to the KKT conditions to obtain the Lagrange multiplier solution set. Substituting the Lagrange multiplier solution set into the power allocation formula and performing numerical iterative calculations yields a convergent power solution, where the power allocation formula is: ； in, This represents the optimal transmit power of base station i; This indicates the minimum signal-to-interference-plus-noise ratio requirement; Indicates noise power; Represents the Lagrange multiplier corresponding to base station i; This represents the channel gain from base station i to its serving users; This represents the interference channel gain of base station j to the users served by base station i; This represents the transmit power of base station j; The converged power solution is vector normalized to obtain the optimal power allocation vector.

4. The cloud-edge-device collaborative energy management system according to claim 1, characterized in that, The optimization module is used for: The optimal routing strategy is input into the decision trigger for state monitoring processing to obtain the collaborative decision trigger signal; Based on the collaborative decision-making trigger signal, the cloud-edge-device three-layer state information is aggregated and processed to obtain a multi-level state dataset. Based on priority rules, the multi-level state dataset is subjected to decision conflict coordination processing to obtain collaborative decision parameters; The collaborative decision parameters are input into a multi-objective optimization algorithm for global optimization to obtain a collaborative control instruction set. The energy management control scheme is obtained by performing layered distribution and encapsulation processing on the collaborative control instruction set.

Images