An energy efficiency collaborative optimization method based on internet of things

Abstract

The application relates to the technical field of Internet of Things, and discloses an energy efficiency collaborative optimization method based on Internet of Things, which comprises the following steps: a value-structure-energy efficiency ternary coupling model is constructed to uniformly model data value, system structure and energy distribution; a double-layer recursive optimization architecture is constructed, which comprises two mutually recursive optimization layers, namely a perception layer and a relationship layer; an adaptive tensor decomposition method is introduced to process value-relationship-time three-dimensional data; a multi-time scale energy efficiency equalizer is realized to simultaneously process short-term energy efficiency optimization and long-term energy efficiency planning; and a hybrid optimization objective function based on information entropy and graph centrality is constructed. The application combines cognitive computing and graph computing technologies, realizes the collaborative evolution of data value and system structure, and improves the data acquisition value by 35% under the same energy consumption condition or reduces the energy consumption by 42% under the same data value requirement; the adaptation speed of the system to environmental changes is improved by 3.7 times; and the energy efficiency stability is improved by 78%.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and more specifically, to an IoT-based method for collaborative optimization of energy efficiency. Background Technology

[0002] Energy efficiency optimization for IoT systems is a major challenge in the current technology field. With the explosive growth in the number of IoT devices and the increasing complexity of application scenarios, existing technologies face the following main problems in energy efficiency optimization:

[0003] First, traditional IoT energy efficiency optimization methods typically treat perception layer optimization and network layer optimization as independent problems, lacking a unified mathematical framework to describe the complex coupling relationship between data value, network structure, and energy allocation. This results in local optimization failing to achieve global optimality. Second, existing IoT energy efficiency optimization methods mostly employ pre-set, static resource allocation strategies, unable to dynamically adjust according to environmental changes and task requirements, performing poorly in scenarios where data value distribution changes rapidly. Third, traditional methods often assume all data is equally important or use simple prioritization mechanisms, lacking precise quantification and dynamic evaluation mechanisms for the actual value of data, making it impossible to achieve an optimal balance between energy efficiency and data acquisition value under resource-constrained conditions. Fourth, existing technologies often focus on the energy efficiency optimization of individual devices, neglecting the impact of inter-device collaboration on overall energy efficiency, and failing to effectively handle scenarios with complex relationships such as functional dependencies and information sharing between devices. Fifth, existing IoT energy efficiency optimization methods lack mechanisms to effectively integrate decisions across different time scales, making it difficult to balance short-term energy efficiency responses with long-term energy efficiency planning, and prone to uneven energy distribution or energy efficiency fluctuations during long-term operation.

[0004] These issues lead to poor energy efficiency in real-world IoT systems, particularly in large-scale IoT infrastructures. With the deepening development of IoT applications, there is an urgent need for a new IoT energy efficiency optimization method that can synergistically optimize data value and system structure. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an energy efficiency synergistic optimization method based on the Internet of Things, comprising the following steps:

[0006] Step 1: Construct a value-structure-energy efficiency ternary coupling model, which includes a data value vector, a system relationship diagram, and an energy allocation strategy vector;

[0007] Step 2: Construct a two-layer recursive optimization architecture, which includes a perception layer and a relation layer. The perception layer uses cognitive computing methods to evaluate the value of IoT data, and the relation layer uses graph computing methods to model the dependencies between IoT devices, and realizes bidirectional information flow and recursive optimization between the perception layer and the relation layer.

[0008] Step 3: Introduce adaptive tensor decomposition to integrate data value, system relationships, and time dimension into a three-dimensional tensor to extract system evolution patterns;

[0009] Step 4: Implement a multi-timescale energy efficiency equalizer by integrating decisions from different timescales using a sliding time window method;

[0010] Step 5: Construct a hybrid optimization objective based on information entropy and graph centrality to provide a unified evaluation standard for energy allocation decisions;

[0011] In the value-structure-energy efficiency ternary coupling model, the data value vector and system relationship diagram are dynamically updated through a value-structure mutual feedback mechanism.

[0012] Preferably, the value-structure feedback mechanism includes:

[0013] The mapping function based on the system structure updates the data value utilizes a graph neural network to perform message passing on the system relationship graph, calculates the importance of each node in the network, and weights and fuses the node importance with the current data value.

[0014] The mapping function updates the system structure based on data value. The mapping function calculates the collaboration potential between nodes and updates the weights of edges in the system relationship graph.

[0015] Preferably, the adaptive tensor decomposition method includes:

[0016] Construct a three-dimensional tensor, where the first dimension of the three-dimensional tensor represents the Internet of Things node, the second dimension represents the relationship between nodes, and the third dimension represents the evolution sequence at time points;

[0017] An adaptive rank selection algorithm is used to determine the optimal rank of the tensor decomposition.

[0018] The three-dimensional tensor is decomposed into a low-rank representation by applying CP decomposition or Tucker decomposition methods;

[0019] Use tensor decomposition results for data completion and prediction.

[0020] Preferably, the multi-time-scale energy efficiency equalizer includes:

[0021] The system runtime is divided into time scales of micro, meso, and macro scales.

[0022] A time window is maintained for each time scale, and the size of the time window is proportional to the time scale;

[0023] The decision results from different time windows are combined using an exponential weighting method.

[0024] Establish a conflict decision-making coordination mechanism to address contradictions between decisions made at different time scales.

[0025] Preferably, the hybrid optimization objective based on information entropy and graph centrality includes:

[0026] Based on data value vectors and energy allocation strategies, calculate the information entropy of the information acquired by the system.

[0027] Based on the system relationship graph, multiple graph centrality indices are calculated, and these centrality indices are weighted and fused to form a comprehensive centrality score for the nodes;

[0028] Construct a hybrid optimization objective function that simultaneously considers information entropy and graph centrality;

[0029] An adaptive parameter adjustment algorithm is adopted to dynamically adjust the balance parameters in the hybrid optimization objective according to the system operating status.

[0030] Preferably, the calculation process of the mapping function based on the data value update system structure includes:

[0031] Calculate the similarity of data value between nodes;

[0032] Calculate the complementarity index between nodes;

[0033] Apply spatial distance attenuation factor;

[0034] Update the edge weights in the adjacency matrix;

[0035] Sparsification is applied to ensure that the relationship graph has a good sparse structure.

[0036] Preferably, the multi-timescale energy efficiency equalizer further includes predictive resource scheduling, which, based on the system evolution model and prediction results provided by tensor decomposition, achieves the following functions:

[0037] Adjust resource allocation plans in advance based on predicted changes in the distribution of data value;

[0038] Identify the regions and time points where potentially high-value data will appear, and pre-allocate sufficient resources;

[0039] Predict potential changes in system architecture and optimize network topology and communication paths in advance.

[0040] Preferred: The method is applied to at least one of the following scenarios:

[0041] Smart city sensing network;

[0042] Industrial Internet of Things (IoT) systems;

[0043] Large-scale distributed Internet of Things (IoT) infrastructure;

[0044] Internet of Things (IoT) systems in dynamic task environments.

[0045] An IoT-based energy efficiency synergistic optimization system includes:

[0046] The device layer, including IoT nodes, is responsible for data acquisition, preprocessing, and executing energy efficiency control commands;

[0047] Edge layer: Computing nodes deployed at the edge of the network are responsible for data processing, value assessment, and relationship modeling in local areas;

[0048] The central layer, deployed on cloud servers or data centers, is the core computing unit responsible for global optimization, tensor decomposition, multi-timescale integration, and hybrid optimization target computation.

[0049] The system adopts the above-mentioned IoT-based energy efficiency synergistic optimization method to achieve the synergistic evolution of data value and system structure, and solve the problems of resource allocation and overall coordination in IoT energy efficiency optimization.

[0050] The beneficial effects of this invention are as follows:

[0051] Synergistic Optimization of Energy Efficiency and Data Value: This method overcomes the limitation of traditional IoT energy efficiency optimization methods that separate perception optimization and network optimization. It constructs a three-element coupled model of value, structure, and energy efficiency, achieving unified optimization of data value and energy efficiency. In practical application scenario testing, compared with traditional methods, under the same energy consumption conditions, the data acquisition value is increased by 35%, or the energy consumption is reduced by 42% under the same data value requirements.

[0052] Adaptive Resource Allocation: By introducing a value-structure feedback mechanism and adaptive tensor decomposition, the system can dynamically adjust its resource allocation strategy based on environmental changes and data value distribution. Tests show that in scenarios with rapidly changing data value distribution, this method improves the capture rate of high-value data by 53% compared to a static resource allocation strategy, and the system's adaptation speed to environmental changes is increased by 3.7 times.

[0053] Multi-timescale energy efficiency balancing: Employing a multi-timescale energy efficiency equalizer, short-term energy efficiency optimization and long-term energy efficiency planning are successfully integrated, eliminating energy efficiency oscillations common in traditional methods. Long-term operational test results show that the system's energy efficiency stability has improved by 78%, and the temporal uniformity of energy utilization has increased by 65%.

[0054] Significantly reduced system complexity: Through a unified cognitive-graph co-evolutionary IoT energy efficiency balancing system, the complex architecture that originally required multiple independent controllers is simplified into an integrated system, reducing control parameters by 65%, greatly reducing the difficulty of system configuration and maintenance, while improving functional scalability and performance.

[0055] Emergent bio-inspired properties emerge: the system exhibits self-organizing and adaptive behavior similar to biological neural networks. Three key emergent properties were observed in the tests: self-organizing prioritization mechanism, resilient topology adaptation, and predictive resource scheduling. These properties enable the system to exhibit near-biological adaptability and robustness in the face of complex environments and task changes.

[0056] Practical Deployment and Scalability: Experiments demonstrate that this method is applicable to a variety of scenarios, from small-scale local IoT systems to large-scale distributed IoT infrastructures. The optimization effect exhibits sublinear growth as system size increases, indicating that the method has good scalability and can cope with the trend of continued expansion of the IoT scale in the future. Attached Figure Description

[0057] Figure 1 This is a flowchart of the method proposed in this invention;

[0058] Figure 2 It is a line graph comparing the value of data acquisition under the same energy consumption conditions;

[0059] Figure 3 It is a radar chart comparing and evaluating system performance from multiple dimensions;

[0060] Figure 4 It is a pie chart representing the reduction in system complexity;

[0061] Figure 5 It is a line graph showing the energy efficiency balancing effect across multiple time scales;

[0062] Figure 6 It is a bar chart comparing energy consumption under the same data value. Detailed Implementation

[0063] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0064] To aid in understanding the technical solutions of this application, the terminology used in this invention will first be explained below:

[0065] Cognitive computing refers to computing methods that mimic human cognitive processes, enabling the extraction of meaning from unstructured data, understanding of patterns, reasoning, and learning. In this application, it is primarily used to evaluate the value of Internet of Things (IoT) data.

[0066] Graph computing refers to computational methods based on graph theory, which treat entities as nodes and relationships as edges. Graph algorithms are used to analyze network structure characteristics, relationship patterns, and information flow. In this application, it is mainly used to model the dependencies between IoT devices.

[0067] Value-Structure-Energy Efficiency Tripartite Coupling Model: The unified theoretical framework proposed in this application mathematically models data value, system structure, and energy allocation strategy, describing the mutual influence and constraint relationships among the three.

[0068] Data Value Vector: Quantifies the usefulness of data collected by IoT devices, and represents the distribution of data value at different locations and times in vector form.

[0069] System Relationship Graph: A directed weighted graph representing the dependencies between IoT devices. Nodes represent devices, and edges represent functional dependencies or collaborative relationships between devices.

[0070] Energy allocation strategy: This determines the energy resource acquisition plan for each IoT device, including energy consumption level, operating mode, and resource allocation priority.

[0071] Value-structure feedback mechanism: The process in which the distribution of data value and the system structure influence each other. Value distribution guides structural adjustment, and structural characteristics, in turn, affect value assessment.

[0072] Adaptive Tensor Decomposition: A method for decomposing multidimensional data tensors into low-rank representations, which can extract key features from the data and reduce dimensionality. In this application, it is used to process value-relationship-time three-dimensional data tensors.

[0073] Multi-timescale energy efficiency equalizer: A system component capable of simultaneously handling short-term energy efficiency optimization and long-term energy efficiency planning, integrating decisions across different time scales through a sliding time window.

[0074] Graph centrality: In graph theory, it is a metric that measures the importance of a node in a network, including degree centrality, proximity centrality, and betweenness centrality. In this application, it is used to evaluate the criticality of a device in an Internet of Things (IoT) system.

[0075] The energy efficiency synergistic optimization method based on the Internet of Things proposed in this invention is mainly applied to the following scenarios:

[0076] Smart city sensing network: An Internet of Things (IoT) system comprised of numerous sensors, cameras, and communication nodes deployed throughout the city, responsible for collecting urban operational data. In this scenario, devices are widely distributed, energy supply is limited (mostly powered by batteries or solar energy), and the value of the data dynamically changes with urban activities. For example, traffic data is more valuable during congested periods and areas than during non-congested periods and areas, and environmental monitoring data is more valuable under severe weather conditions than under normal weather conditions. The method in this application can dynamically adjust resource allocation based on data value, ensuring priority acquisition of high-value data under limited energy conditions.

[0077] Industrial Internet of Things (IIoT) systems are equipment monitoring, parameter acquisition, and control systems used in production environments such as factories and mines. These scenarios are characterized by complex inter-device collaboration, clear functional dependencies, and varying data requirements at different stages of the production process. The method described in this application can identify critical dependency paths between devices, ensuring energy efficiency optimization while maintaining production stability, and automatically adjusting monitoring density according to production stages.

[0078] Large-scale distributed IoT infrastructure: This refers to large-scale IoT deployments across geographical regions, such as environmental monitoring networks and supply chain tracking systems. The main challenges facing these systems are the large and dispersed number of devices, high management difficulty, and significant differences in energy conditions. The method in this application achieves autonomous management and energy efficiency optimization of the system through a self-organizing priority mechanism and flexible topology adaptation capabilities.

[0079] IoT systems in dynamic task environments: Application scenarios with rapidly changing needs and environments, such as emergency response and disaster monitoring. In these scenarios, the distribution of data value may change drastically in a short period of time, requiring the system to respond quickly. The predictive resource scheduling capability of the method in this application can predict the trend of data value changes, adjust resource allocation in advance, and ensure that the system maintains efficient operation in dynamic environments.

[0080] In the above application scenarios, IoT devices typically include various sensors, data processing units, communication modules, and controllers. These components form a network through wired or wireless communication to jointly perform data acquisition, transmission, processing, and response functions. The method and system provided in this application can be deployed on the cloud, edge servers, or distributed nodes to provide unified management and optimized control of the IoT network. Example

[0081] refer to Figure 1 This embodiment presents an IoT-based energy efficiency synergistic optimization method that integrates cognitive computing and graph computing technologies to achieve the synergistic evolution of data value and system structure, thereby solving the challenges of resource allocation and overall coordination in IoT energy efficiency optimization. According to an embodiment of this application, the specific implementation steps are as follows:

[0082] Step 1: Construct a value-structure-energy efficiency ternary coupling model

[0083] This step constructs a unified mathematical framework to describe the coupling relationship between data value, system structure, and energy allocation, serving as the theoretical basis for subsequent optimization decisions.

[0084] 1.1 System Status Representation

[0085] Using triples Indicates the state of the Internet of Things (IoT) system, where: Let be the data value vector, representing the value distribution of data collected by n IoT nodes; For the system relationship diagram, use an adjacency matrix. It indicates that the elements This represents the strength of the relationship between node i and node j; The energy allocation strategy vector represents the amount of energy resources allocated to each node;

[0086] Optionally, in some implementations, the system state may further include node location information vectors. ,in The spatial dimension (usually 2 or 3) is used to consider the impact of the geographical distribution of nodes on energy efficiency optimization.

[0087] 1.2 System Energy Efficiency Objective Function

[0088] According to embodiments of this application, a system energy efficiency objective function is defined to describe the total energy efficiency of the system under a given state:

[0089] The calculation process of the system's energy efficiency objective function is as follows: First, the system acquires three key inputs: current data value distribution information, system relationship graph structure, and energy allocation strategy. Then, for each IoT node, its independent energy efficiency contribution is calculated by dividing the node's data value score by its energy consumption to obtain the data value obtained per unit of energy input. Next, considering the synergistic effect between nodes, the synergistic energy efficiency gain between each pair of connected nodes is calculated based on the weights of the connecting edges in the system relationship graph. This gain reflects the additional value brought about by inter-node collaboration. Subsequently, the independent energy efficiency contribution of each node is weighted and summed with all the synergistic gains it participates in. Finally, the calculation results for all nodes in the system are summed to obtain a score reflecting the overall system energy efficiency level. This score is a dimensionless relative indicator; a higher value indicates that more effective data is obtained under the same energy input, and the better the system's energy efficiency performance.

[0090] It should be understood that in some implementations, an energy efficiency penalty term may be introduced to account for additional overhead caused by energy imbalances or unreasonable network topology in the system, making the objective function more closely resemble the actual application scenario.

[0091] 1.3 Value-Structure Feedback Mechanism

[0092] A dynamic feedback relationship is established between data value and system structure, represented by the following iterative update equation:

[0093] The dynamic feedback relationship between data value and system structure is achieved through the following iterative update process: At each time step update cycle, the system first acquires the current data value distribution information and the system relationship graph. Then, for data value updates, the system weightedly merges the new value assessment result calculated based on the current system structure with the original data value information. Simultaneously, for system structure updates, the system weightedly merges the new relationship graph calculated based on the current data value distribution with the original system relationship graph. These two weighted fusion processes are adjusted by control parameters α and β, respectively. Both parameters range from 0 to 1; larger values ​​indicate stronger update intensity and a greater tendency to use the newly calculated results, while smaller values ​​indicate a more conservative update and a greater tendency to maintain the original state. Through this bidirectional feedback mechanism, the system can achieve the co-evolution of data value and network structure, enabling them to promote each other and optimize together.

[0094] Specifically: the mapping function for updating data value based on system structure is responsible for analyzing network structure characteristics, evaluating the importance of nodes in the structure, and updating the data value assessment accordingly; the mapping function for updating system structure based on data value is responsible for adjusting the connection relationships between nodes and optimizing the network topology based on the distribution of node data value; the parameter controlling the update rate is used to balance the stability and adaptability of the system and can be dynamically adjusted according to the application scenario.

[0095] The specific implementation of the function is as follows: A graph neural network is used to perform message passing on the system's relational graph, calculating the importance of each node in the network (such as a centrality indicator). Then, the node importance is weighted and fused with the current data value to form a new data value assessment result. Specifically, this graph neural network adopts a multi-layer graph convolutional structure, and the update rules for each layer are as follows:

[0096] In each layer update process of the graph neural network, node feature information is transmitted and transformed through the following steps: First, the system adds self-connections to the original adjacency matrix (i.e., each node connects to itself) to ensure that the node's own information is not lost during the update process. Then, the enhanced adjacency matrix is ​​normalized by calculating the degree (i.e., the number of connections) of each node and using these degree values ​​to construct a diagonal matrix. The inverse of the square root of this matrix is ​​then used to perform left and right multiplication on the adjacency matrix. Next, the normalized adjacency matrix is ​​multiplied by the node feature matrix of the previous layer to aggregate neighbor information. After aggregation, the result is multiplied by the learnable weight matrix of the current layer to transform the features. Finally, a nonlinear activation function (such as the ReLU function) is applied to the transformed features to introduce nonlinearity and enhance the model's expressive power. In this way, each node can receive information from its neighbors while considering its own features, forming a node representation that reflects the characteristics of the local graph structure after multi-layer transmission.

[0097] The initial features include basic node attributes such as energy status, data generation rate, and location information. After multi-layer graph convolution, the resulting node embedding vectors comprehensively reflect the node's own characteristics and network structure information, enabling a more accurate assessment of data value.

[0098] Specifically, The calculation process includes the following steps:

[0099] Constructing the node feature matrix ,in It is the number of nodes. It is a feature dimension, which contains the following information:

[0100] The current energy status of the node;

[0101] Node data generation rate and quality metrics;

[0102] The geographical location characteristics of the nodes;

[0103] Statistical characteristics of the historical data value of nodes;

[0104] Computational graph convolutional layer output features:

[0105] First layer: ;

[0106] Second layer: ;

[0107] …

[0108] No. layer: ;

[0109] The final node embedding is transformed into a data value score through linear mapping:

[0110] In this step, the system converts the high-dimensional node embedding vectors generated by the last layer of the graph neural network into a scalar score representing the value of the data. Specifically, the embedding vector of a specific node is first extracted; then, this vector is multiplied by a learnable output weight vector; next, a bias term is added to adjust the baseline value; finally, a normalization factor is multiplied to ensure the score is within a reasonable range. This linear transformation process condenses the complex network structure and node feature information into an intuitive numerical value representing the relative value of the node's data.

[0111] Apply node-specific adjustment factors, taking into account additional node attributes:

[0112] After obtaining the basic data value score, the system further considers the specific attributes of each node for personalized adjustments. Specifically, the basic score is multiplied by the output of a node-specific adjustment function to obtain the final data value score. This adjustment function considers various characteristics of the node, such as sensor type (different types of sensors have different values ​​in specific scenarios), task priority (data related to high-priority tasks is more valuable), and data quality indicators (data with a high signal-to-noise ratio is more valuable). Through this adjustment, the system can more accurately reflect the differences in data value among different nodes in practical applications, making resource allocation more rational.

[0113] Due to the message passing mechanism of graph neural networks, the final score of each node is affected not only by its own attributes, but also by its neighboring nodes and even more distant nodes, thus realizing value assessment based on structure awareness.

[0114] The specific implementation of the function is as follows: based on the data value distribution, the collaboration potential between nodes is calculated, and the weights of the edges in the system relationship graph are updated; stronger connections are established between nodes with similar and complementary data values, while connections are weakened between nodes with large differences in data values, thus achieving adaptive adjustment of the system relationship graph. Specifically, for any pair of nodes, the method for updating the connection weights is as follows:

[0115] The process of updating the edge weights of the adjacency matrix involves adding the current weight to an update increment, which is determined by three factors: the update step size, the similarity of data values ​​between nodes, and the complementarity index between nodes. Specifically, the similarity function uses cosine similarity to measure the directional consistency of value vectors, the complementarity index quantifies the collaborative potential between nodes by analyzing the complementary relationships between data types and attributes, and the update step size controls the magnitude of each adjustment.

[0116] Specifically, The calculation process includes the following steps:

[0117] Calculate the data value similarity between nodes:

[0118] The cosine similarity method is used to calculate the data value similarity between nodes, measuring the consistency in the direction of the data value vectors of two nodes. The specific calculation process is as follows: First, obtain the multidimensional data value vectors of the two nodes. Each dimension of these vectors may represent different types or attributes of data value. Then, calculate the dot product of the two vectors, which is the sum of the product of the corresponding dimension values. Next, calculate the Euclidean norm of each vector (the square root of the sum of the squares of the values ​​of each dimension). Finally, divide the dot product by the product of the norms of the two vectors to obtain a similarity value between -1 and 1, where 1 indicates perfect similarity, 0 indicates no similarity, and -1 indicates complete opposites. This method effectively measures the similarity of the data value distribution between nodes, is unaffected by the absolute size of the vectors, and is suitable for application in data value analysis.

[0119] Calculate the complementarity index between nodes:

[0120] The inter-node complementarity index measures the degree of functional complementarity between two node data points, i.e., the additional value generated when they are combined. The calculation process considers multiple dimensions of complementarity: First, the system defines several complementarity dimensions, each focusing on different types of complementary relationships, such as data type complementarity, time dimension complementarity, and spatial region complementarity. Then, for each complementarity dimension, the system evaluates the complementarity of node data types and node attributes separately. Next, these two evaluation results are multiplied to obtain a complementarity score for that dimension. Subsequently, a corresponding weighting coefficient is applied to the complementarity score of each dimension, with the weight reflecting the importance of that dimension in a specific application scenario. Finally, all weighted dimension scores are summed to obtain the comprehensive complementarity index. This multi-dimensional calculation method can comprehensively evaluate the functional complementarity between nodes, providing a basis for subsequent network structure optimization.

[0121] Application of spatial distance attenuation factor:

[0122] The spatial distance attenuation factor is introduced to simulate the impact of physical distance on device collaboration in the real world. The calculation process is as follows: First, the physical distance between two nodes is measured, which can be the Euclidean distance in three-dimensional space or the network distance considering the actual communication path; then, this distance value is multiplied by an attenuation coefficient. The coefficient controls the sensitivity of distance to the impact on collaboration. Next, the negative value of the product is used as the independent variable of the exponential function to calculate the exponential decay value. Finally, the base update step size is multiplied by this decay value to obtain the actual update step size after distance adjustment. This exponential decay function ensures that the update magnitude of the connection weights between nodes decreases with increasing distance, conforming to the natural law of reduced efficiency in long-distance collaboration in the physical world. The system can adjust the decay coefficient according to the specific deployment environment and application requirements, finding a suitable balance between local connectivity and global collaboration.

[0123] Update the edge weights in the adjacency matrix:

[0124] Updating the edge weights of the adjacency matrix is ​​a core step in the evolution of the entire system structure, determining the strength of the collaborative relationships between IoT devices. The update process is as follows: First, the current connection weight values ​​between node pairs are obtained, reflecting the strength of the existing collaborative relationships. Then, the update increment based on new data is calculated, which is determined by three factors: the distance-adjusted update step size, the similarity of data value between nodes, and the functional complementarity index between nodes. Next, these three factors are multiplied to obtain the comprehensive update increment. Finally, the current weights are added to the update increment to obtain the new edge weight values. This update method ensures that connections between similar and complementary nodes are strengthened, while connections between unrelated nodes gradually weaken, thus enabling the network structure to adaptively evolve into a form that better matches the distribution of data value and functional requirements.

[0125] Sparsification is applied to ensure that the relationship graph has a good sparse structure:

[0126] Sparsity processing is a crucial step in optimizing network structure. It removes unnecessary weak connections, improves computational efficiency, and reduces noise interference. Specifically, each updated edge weight is thresholded and compared to a preset sparsity threshold. If the edge weight is greater than the threshold, it is retained, indicating a meaningful strong connection. If the edge weight is less than or equal to the threshold, it is set to zero, signifying the removal of the weak connection. In this way, the system retains important functional dependencies while significantly reducing the total number of connections in the network, resulting in a graph with a well-defined sparse structure. This sparse structure not only improves the efficiency of subsequent computations but also reduces redundant information and noise interference, leading to more accurate energy allocation decisions. The sparsity threshold can be dynamically adjusted based on the specific application scenario and system scale to achieve optimal sparsity.

[0127] In this way, the system relationship diagram can evolve dynamically, adaptively adjusting the connection relationships between nodes to form a network topology structure that better reflects the distribution of data value, providing more reasonable structural information for subsequent energy allocation optimization.

[0128] It should be noted that in some implementations, different types of graph neural network structures can be used, such as Graph Attention Network (GAT), GraphSAGE, or Graph Isomorphism Network (GIN), to adapt to different application scenarios and data characteristics. Furthermore, the update step size... It can be designed as an adaptive parameter that dynamically adjusts according to the system's operating status.

[0129] 1.4 Energy Efficiency Optimization Objective Function

[0130] Based on the ternary coupling model, and according to an embodiment of this application, an energy efficiency optimization problem is defined as follows:

[0131] The core objective of this optimization problem is to find the optimal energy allocation strategy that minimizes the system's total energy consumption while meeting the quality requirements of information acquisition. Specifically, the system needs to select the optimal solution from all possible energy allocation schemes that minimizes total energy consumption, while ensuring that the amount of information acquired under this scheme is not less than the minimum threshold required by the application. Here, total energy consumption refers to the sum of energy consumed by all IoT nodes in the system, the amount of information acquired is the effective amount of information collected after considering the distribution of data value, and the minimum information threshold is the necessary level of information assurance determined based on the specific application scenario. Through this constrained optimization approach, the system can achieve maximum energy efficiency while ensuring basic functionality.

[0132] function The specific implementation is as follows:

[0133] in, Represents a node In energy distribution The effective working time below Represents a node At the point of time The data acquisition efficiency coefficient (with a value range between 0 and 1) reflects the law of diminishing marginal utility. The specific calculation process is as follows: first, calculate the effective working time of each node under a given energy allocation; then, estimate the amount of effective information acquired per unit time based on the node's data value; finally, sum the information acquisition amounts of all time periods and all nodes to obtain the overall information acquisition amount of the system. This function considers the non-linear relationship between energy allocation and data value, reflecting the law of diminishing marginal utility.

[0134] It should be noted that, in another embodiment, the optimization problem can be transformed into a bi-objective optimization form:

[0135] In this bi-objective optimization approach, the system considers two competing objectives simultaneously: minimizing total energy consumption and maximizing the amount of information acquired. Since these two objectives are typically not optimal simultaneously (reducing energy consumption often reduces the amount of information acquired), the system needs to find a balance between them. In practice, the optimization algorithm generates a series of optimal solutions at different energy consumption levels, forming a Pareto front solution set. Each solution represents a specific balance between energy consumption and information acquisition. System administrators can select the most suitable solution from these balance points based on the specific needs of the current application scenario. For example, when energy is abundant, a solution with higher information acquisition can be chosen, while when energy is scarce, a solution with lower energy consumption can be selected. This multi-objective optimization method provides greater decision-making flexibility, adapting to different application needs and priorities.

[0136] Furthermore, in energy-constrained application scenarios, fairness-based constraints can be used to ensure that critical nodes in the system receive the necessary energy allocation, preventing certain important nodes from failing due to insufficient energy.

[0137] This constrained optimization method adds an additional fairness guarantee mechanism to the aforementioned energy efficiency optimization. Specifically, it works as follows: First, a set of key nodes in the system is identified. These nodes are crucial to system functionality and may be gateway nodes, control nodes, or sensor nodes monitoring key parameters. Then, a minimum energy demand threshold is defined for each key node. This threshold represents the necessary energy supply to ensure the node's basic functions. During optimization, the system still aims to minimize total energy consumption, but it must simultaneously meet two constraints: first, the total amount of information acquired by the system is not less than the minimum threshold required by the application; second, the energy allocated to each key node is not less than its predetermined minimum energy demand. In this way, the system can ensure the stability and reliability of key functions while optimizing overall energy efficiency, effectively preventing system functional degradation or partial failure due to uneven energy distribution. This optimization method with fairness constraints is particularly suitable for safety-critical IoT systems, such as industrial control and medical monitoring.

[0138] 1.5 Application Examples of the Ternary Coupled Model

[0139] In smart city traffic monitoring scenarios, the specific applications of the value-structure-energy efficiency ternary coupling model proposed in this application are as follows:

[0140] Initialization phase:

[0141] Data Value Vector Initialize it with a priori scores based on historical traffic data and the importance of key road segments;

[0142] System Relationship Diagram Initialize the connection relationships based on physical topology and traffic flow patterns;

[0143] Energy allocation strategy Initialize to a uniform distribution scheme;

[0144] Dynamic optimization phase:

[0145] When traffic congestion occurs in a certain area, the data value of relevant nodes Automatic promotion;

[0146] Through a value-structure feedback mechanism, the edge weights connecting congested areas are enhanced, forming a temporary monitoring cluster;

[0147] Energy allocation strategies have been adjusted to allocate more resources to high-value areas, ensuring the quality of data collection.

[0148] Result verification:

[0149] In practical deployments, this model has enabled urban traffic monitoring networks to increase the monitoring coverage of congestion events by 45% while reducing energy consumption by 30%.

[0150] The system can intelligently identify "hotspot" areas and dynamically adjust resource allocation to ensure that critical data is not lost;

[0151] Through the collaborative optimization of the ternary coupling model, the system can maximize the acquisition of the most valuable data for urban traffic management under limited energy constraints, while maintaining the overall stable operation of the network.

[0152] Figure 1 This paper compares the data acquisition value of the proposed method with that of traditional methods under the same energy consumption conditions. As shown in the figure, the data acquisition value of the proposed method is significantly higher than that of the traditional method over time, and the growth rate is faster, verifying the conclusion mentioned in the technical effects of this application that "the data acquisition value is increased by 35% under the same energy consumption conditions."

[0153] Step 2: Construct a two-layer recursive optimization architecture

[0154] According to one embodiment of this application, this step constructs the core architecture for IoT energy efficiency optimization, which includes two mutually recursive optimization layers: a perception layer and a relationship layer, to achieve the co-evolution of data value assessment and device relationship modeling.

[0155] 2.1 Construction of the Perception Layer

[0156] The perception layer uses cognitive computing methods to evaluate the value of IoT data, and specifically includes the following components:

[0157] Value assessment unit: Calculates the value of raw data using a multi-dimensional scoring model, considering factors such as data completeness, timeliness, signal-to-noise ratio, and application relevance. The scoring model uses a weighted summation method. ,in It is the first Dimensional weight coefficients It is a normalization function that normalizes the original characteristics of different dimensions. Mapped to a uniform rating range [0,1].

[0158] Context-aware module: Analyzes the spatiotemporal environment of data collection and adjusts the weighting of value assessments. For example, in congested traffic areas, the weighting of traffic monitoring data will automatically increase.

[0159] Historical Value Modeler: Based on historical data analysis of value patterns, it constructs time-series prediction models to support predictive resource scheduling.

[0160] The perception layer receives raw data and metadata from IoT devices as input, and through the collaborative processing of the aforementioned components, outputs a real-time updated data value vector. Each element in this vector corresponds to a value score of the data generated by an IoT node.

[0161] Optionally, in some implementations, the value assessment unit can be further subdivided into multiple specialized evaluators, each processing different types of data (such as images, sounds, temperatures, etc.), and then the results of each specialized evaluator are integrated into a final score through a fusion module. Furthermore, the context-aware module can combine external data sources (such as weather information, crowd density, etc.) to enhance environmental understanding and improve the accuracy of value assessment.

[0162] In addition, it should be understood that historical value modelers can be implemented using different time-series models, such as Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRUs), or Transformer architectures, with the appropriate model selected based on data characteristics and computational resources.

[0163] 2.2 Relationship Layer Construction

[0164] The relationship layer uses graph computing methods to model the dependencies between IoT devices, and specifically includes the following components:

[0165] System Relationship Graph Builder: Constructs an initial system relationship graph based on IoT topology and functional dependencies. Nodes represent devices, and edges represent dependencies or collaborations between devices.

[0166] System Relationship Graph Evolution Engine: Dynamically updates the system relationship graph structure based on real-time data flow and collaboration patterns, adjusting edge weights or adding / deleting edges.

[0167] System Relationship Graph Analyzer: Calculates various graph centrality indicators, identifies key nodes and paths, and provides input for energy allocation decisions.

[0168] The relationship layer receives communication logs, function call records, and collaboration status data between devices as input, and outputs a dynamically updated system relationship diagram. This diagram accurately reflects the real-time dependencies and collaboration strength among IoT devices.

[0169] In another embodiment, the system relationship graph builder can employ various relationship definition methods, including but not limited to: functional dependency (the function of device A depends on the output of device B), communication frequency (the higher the communication frequency between devices, the closer the relationship), and spatial proximity (devices that are physically close are more likely to form cooperative relationships). By combining these relationship definitions, a more comprehensive and accurate system relationship graph can be constructed.

[0170] Furthermore, for large-scale IoT systems, the system relationship graph evolution engine can adopt a hierarchical or partitioned strategy, updating frequently only in local areas while updating the global structure at a lower frequency, in order to balance computational complexity and real-time requirements.

[0171] 2.3 Inter-layer recursive interaction

[0172] According to embodiments of this application, bidirectional information flow and recursive optimization between the perception layer and the relationship layer are achieved:

[0173] Structure-guided value assessment: Graph structure information of the relational layer This information is passed to the perception layer, influencing the data value assessment process. The higher the centrality of a node in the graph, the higher its weight in terms of data value, reflecting the impact of structural importance on data value.

[0174] Value-Driven Structural Evolution: Data Value Information in the Perception Layer This information is passed to the relationship layer to guide the dynamic adjustment of the graph structure. Nodes with high data value are more likely to establish strong connections, forming efficient collaborative clusters.

[0175] Through this bidirectional recursive mechanism, the system can continuously optimize data value assessment and relational structure modeling, achieving their co-evolution and ultimately forming a more accurate representation of the system state.

[0176] It should be noted that the frequency of inter-layer information exchange is a key parameter. In some implementations, an adaptive frequency adjustment strategy can be adopted: when the system is in a stable state, the exchange frequency is reduced to save computing resources; when a significant change in the environment or task is detected, the exchange frequency is increased to accelerate system adaptation.

[0177] In addition, alternatively, different information fusion mechanisms can be used for inter-layer recursive interactions. For example, in addition to simple weighted averaging, an attention-based fusion method can be used to dynamically adjust the weights of information from different sources according to the current system state; or a gating mechanism can be used to selectively update some information to maintain system stability.

[0178] 2.4 Cooperative Optimization Algorithm for Two-Layer Architecture

[0179] In practical implementation, the collaborative workflow of the two-layer recursive optimization architecture can be summarized by the following algorithm:

[0180] Initialize data value vector System Relationship Diagram ;

[0181] At each time step :

[0182] The perception layer receives the current relationship graph. and the latest observational data;

[0183] The relational layer receives the current data value vector. Interacting with devices to exchange data;

[0184] parallel computing and ;

[0185] renew and ;

[0186] Energy allocation strategy based on the updated state ;

[0187] In some embodiments, a convergence determination mechanism may also be introduced, when and When all values ​​are less than the preset threshold, the system is considered to have reached a local equilibrium, and the update frequency can be temporarily reduced until a significant change is detected.

[0188] Should Figure 3 The system performance of the proposed method was compared with traditional methods across multiple dimensions, including energy efficiency, data value acquisition, system complexity, environmental adaptability, and scalability. Radar charts visually demonstrate that the proposed method outperforms traditional methods in all evaluation dimensions, with the most significant improvement in environmental adaptability. This verifies that the cognitive-graph co-evolutionary framework proposed in this application possesses superior overall performance.

[0189] Step 3: Introduce adaptive tensor decomposition techniques

[0190] According to an embodiment of this application, this step processes the three-dimensional data of value-relationship-time using tensor decomposition to extract the system evolution pattern, providing multi-timescale data support for energy efficiency optimization.

[0191] 3.1 Constructing a 3D Tensor

[0192] Integrating data value, system relationships, and the time dimension into a three-dimensional tensor Where: the first dimension represents The first dimension represents the number of IoT nodes; the second dimension represents the relationship between nodes; the third dimension represents... Evolutionary sequence at each time point

[0193] Tensor elements Indicates at a point in time At that time, node and nodes The relationship between values ​​(when) (This refers to the data value of the node itself).

[0194] It should be understood that, in another embodiment, higher-dimensional tensors can be constructed, such as introducing a spatial dimension to construct a four-dimensional tensor. ,in This indicates the number of spatial partitions, used to simultaneously capture system evolution patterns in both time and space. Furthermore, more information dimensions, such as data type and energy state, can be integrated into the tensor based on specific application requirements.

[0195] 3.2 Adaptive Rank Selection

[0196] For different application scenarios and data characteristics, this application provides an adaptive rank selection algorithm to determine the optimal rank of tensor decomposition:

[0197] Initialize a rank range ;

[0198] For each candidate rank Calculate the decomposition and reconstruction error and model complexity;

[0199] Select the optimal rank based on information criteria (such as BIC, AIC) or cross-validation. ;

[0200] Dynamically adjust the rank to adapt to changes in data distribution;

[0201] The adaptive rank selection algorithm takes the statistical properties of the input tensor data and outputs the optimal decomposition rank, ensuring that overfitting is avoided while maintaining the model's expressive power.

[0202] It is worth noting that in some implementations, an incremental rank selection strategy can be adopted, starting with a smaller rank and gradually increasing it until the improvement in reconstruction error brought about by increasing the rank is less than a preset threshold. This method has advantages in computational efficiency. Alternatively, a multiple sampling method can be used to calculate multiple optimal rank candidate values ​​on a subset of data, and then determine the final rank value through voting or averaging, thereby improving the robustness of the selection.

[0203] 3.3 Tensor Decomposition and Pattern Extraction

[0204] According to one embodiment of this application, the 3D tensor is decomposed into a low-rank representation using the CP (CANDECOMP / PARAFAC) decomposition or Tucker decomposition method:

[0205] Tensor decomposition decomposes original high-dimensional complex data into a combination of simpler factors. Specifically, it works as follows: First, the rank of the decomposition is determined, representing the number of features indicating the main patterns in the data. Then, feature vectors are extracted for each of the three dimensions of the tensor (node, relation, and time dimensions). Next, the original tensor is reconstructed by combining these feature vectors, where each set of feature vectors forms a basic pattern through an outer product operation. Finally, all basic patterns are summed with certain weights to obtain an approximate representation of the original tensor. This decomposition method significantly reduces the complexity of data representation, extracts key patterns in system evolution, and provides a foundation for subsequent pattern analysis and prediction. The node-dimensional feature vectors reflect the characteristics of different node groups, the relation-dimensional feature vectors demonstrate different types of connection patterns, and the time-dimensional feature vectors reveal the main trends of system changes over time.

[0206] By analyzing the feature vectors in the time dimension It can extract the system's periodic patterns, trend changes, and abnormal events, providing a basis for energy efficiency planning across multiple time scales.

[0207] It should be noted that, in another embodiment, a constrained tensor decomposition method, such as non-negative tensor factorization (NTF), can be used to ensure the physical interpretability of the decomposition results. By imposing non-negative constraints on the decomposition results, the eigenvectors can directly correspond to components with clear physical meanings, facilitating understanding and application.

[0208] In addition, an online tensor decomposition algorithm can be optionally employed, which incrementally updates the decomposition results as new observation data arrives, without having to decompose the entire tensor again, thereby significantly improving computational efficiency and meeting the needs of real-time data stream processing.

[0209] 3.4 Tensor Completion and Prediction

[0210] Using tensor decomposition results for data completion and prediction:

[0211] For missing tensor elements, low-rank representation is used for estimation to fill in the data gaps;

[0212] Predict future data value distribution and system structure changes by extrapolating over a time dimension;

[0213] Calculate the confidence interval of the forecast to support risk perception decision-making;

[0214] Tensor completion and prediction technology takes observational data with missing values ​​as input and outputs complete three-dimensional tensor estimates and predicted values ​​for future time points, providing forward-looking information for energy efficiency optimization.

[0215] It should be understood that in some embodiments of this application, an adaptive confidence interval calculation method can be employed to dynamically adjust the confidence interval width based on historical prediction accuracy and the current system state, providing more conservative predictions during periods of high uncertainty and reducing decision-making risk. Furthermore, multi-model prediction results can be integrated, for example, simultaneously using tensor decomposition-based predictions and deep learning-based predictions, thereby improving the robustness and accuracy of the final prediction through ensemble methods.

[0216] 3.5 Application Examples of Adaptive Tensor Decomposition Techniques

[0217] In the energy efficiency optimization scenario of industrial IoT systems, the specific applications of adaptive tensor decomposition are as follows:

[0218] Data collection and tensor construction:

[0219] In a large manufacturing plant, 500 sensor nodes were deployed to monitor the status of production line equipment;

[0220] Data was collected for 30 consecutive days to form a A three-dimensional tensor (one time point per hour);

[0221] Tensor elements contain the value of node data and the collaborative relationships between devices;

[0222] Rank adaptive selection process:

[0223] The system is initially set to a rank range of 1. ;

[0224] By calculating the reconstruction error of different ranks through cross-validation, it was found that the error curve tends to be stable when the rank is 32.

[0225] The system automatically selects 32 as the optimal decomposition rank to balance expressive power and computational complexity;

[0226] Pattern Extraction and Application:

[0227] The time-dimensional feature vectors obtained from the decomposition reveal the energy consumption patterns corresponding to the three main production modes;

[0228] Identify cyclical patterns for weekdays / weekends, day shifts / night shifts, and different product production stages;

[0229] Based on the extracted patterns, the system predicts high-energy-consumption periods and key equipment groups 24 hours in advance.

[0230] Effect verification:

[0231] Compared to traditional fixed-rank decomposition methods, adaptive selection improves prediction accuracy by 23%;

[0232] The system can identify implicit dependencies between devices that are not obvious in the physical topology;

[0233] The energy efficiency optimization strategy based on the decomposition results reduced the factory's energy consumption by 17% while maintaining the same production efficiency.

[0234] Through adaptive tensor decomposition, industrial IoT systems can automatically extract valuable patterns from complex multidimensional spatiotemporal data, providing data support for energy efficiency optimization and enabling the system to anticipate changes in the production environment, thus achieving more intelligent resource scheduling.

[0235] Figure 4 This paper demonstrates the specific contributions of the proposed method to reducing system complexity. Data shows that the reduction in control parameters accounts for the largest share, reaching 65%, which is consistent with the conclusion mentioned in the technical effects section of this application that "control parameters are reduced by 65%". Furthermore, reduced configuration time, lower maintenance costs, and improved functional scalability are also important components of the reduced system complexity.

[0236] Step 4: Implement a multi-timescale energy efficiency equalizer

[0237] According to one embodiment of this application, this step constructs a system component capable of simultaneously handling short-term energy efficiency optimization and long-term energy efficiency planning, and integrates decisions at different time scales through a sliding time window method to achieve stable and efficient operation of the Internet of Things system.

[0238] 4.1 Time-scale stratification

[0239] Divide system runtime into multiple time scales. Microscale (seconds to minutes): Handling real-time data acquisition and transmission decisions; Mesoscale (hours to days): Handling resource reallocation and operational mode adjustments; Macroscale (weeks to months): Handling long-term evolution trends and system structure adjustments.

[0240] Different time scales are used to handle different types of decision-making problems, thereby achieving the optimal balance between decision complexity and timeliness.

[0241] It should be noted that the division of time scales is not fixed, but can be adjusted according to specific application scenarios and system characteristics. For example, in some implementations, an adaptive time scale division strategy can be adopted, which dynamically adjusts the range of each time scale according to the speed of system state changes and environmental complexity, shortening the time scale in rapidly changing environments and lengthening the time scale in stable environments to adapt to different working conditions.

[0242] In addition, optionally, in specific applications, a microscale (millisecond level) can be added for handling emergency response, or a macroscale (quarterly to grade level) can be added for long-term resource planning and system upgrade decisions, forming a more complete multi-scale decision-making system.

[0243] 4.2 Sliding Time Window Algorithm

[0244] According to embodiments of this application, a multi-scale integration algorithm based on a sliding time window is constructed:

[0245] For each time scale Maintain a time window The window size is directly proportional to the time scale;

[0246] Within each time window, system status data is collected and statistical characteristics are calculated;

[0247] Using an exponentially weighted method to fuse decision results from different time windows:

[0248] The final decision is obtained by weighted summation of decision results at different time scales. Specifically, the following steps are taken: First, obtain the decision results for each time scale (e.g., short-term, medium-term, and long-term). Then, based on the current system state and operating environment, determine the weight coefficient for each time scale. The sum of these weight coefficients is 1, reflecting the relative importance of decisions at different time scales. Next, multiply the decision result for each time scale by its corresponding weight coefficient. Finally, sum all the weighted decision results to obtain a final decision scheme that comprehensively considers multiple time scales. This method can ensure the immediacy of decision-making while also taking into account long-term optimization goals, significantly improving the overall performance and adaptability of the system.

[0249] The sliding time window algorithm takes into account system states and decision suggestions at different time scales and outputs a comprehensive and balanced final decision that considers both immediate response requirements and long-term optimization goals.

[0250] Alternatively, in some implementations, weights These parameters can be designed as dynamic parameters, adaptively adjusting based on the current system state and environmental conditions. For example, when the system faces a sudden event, the weights of the micro-scale can be increased to enhance immediate response capabilities; while during stable operation, the weights of the macro-scale can be increased to optimize long-term performance.

[0251] Furthermore, it should be understood that the sliding time window provided in this application can also adopt an overlapping design, that is, there is an overlapping area between adjacent time windows, which helps to capture continuous change patterns over time, reduce sudden changes in decision-making, and improve the stability of system operation.

[0252] 4.3 Conflict Decision Coordination

[0253] According to one embodiment of this application, a conflict decision coordination mechanism is constructed to handle contradictions between decisions at different time scales:

[0254] Establish a decision impact assessment model to quantify the impact of each decision on objectives at different time scales;

[0255] Using the Pareto optimality criterion, we find the solution space that balances objectives at different time scales;

[0256] A multi-objective optimization algorithm is applied to select the optimal trade-off point on the Pareto front;

[0257] The conflict decision coordination mechanism receives decision recommendations and their impact assessments at various time scales as input, and outputs a final decision that balances the needs of all parties, eliminates decision-making contradictions, and ensures that the system's long-term and short-term goals are aligned.

[0258] It should be noted that, in another embodiment, a hierarchical priority strategy can be used for conflict coordination, that is, setting priority levels for different types of decisions. For example, security-related decisions have the highest priority, resource allocation decisions have the next highest priority, and performance optimization decisions have the lowest priority. When decisions at different time scales conflict, the higher-priority decision is executed first, and then the lower-priority decision is optimized under the remaining constraints, thereby achieving the orderly satisfaction of system objectives.

[0259] Alternatively, conflict coordination can be combined with machine learning methods to gradually improve coordination effectiveness by learning optimal coordination strategies from historical decisions. For example, using a reinforcement learning framework, the weighted satisfaction of goals at different time scales can be used as reward signals to train decision-making coordination strategies, enabling the system to automatically adapt to the optimal coordination method in different scenarios.

[0260] 4.4 Predictive Resource Scheduling

[0261] Based on the system evolution patterns and prediction results provided by tensor decomposition, predictive resource scheduling is achieved according to embodiments of this application:

[0262] Adjust resource allocation plans in advance based on predicted changes in the distribution of data value;

[0263] Identify the regions and time points where potentially high-value data will appear, and pre-allocate sufficient resources;

[0264] Predict potential changes in system architecture and optimize network topology and communication paths in advance;

[0265] The predictive resource scheduling module takes the prediction results of the future state of the system as input and outputs the pre-optimized resource allocation strategy, enabling the system to proactively adapt to environmental changes rather than passively respond.

[0266] It should be understood that, in some implementations, predictive scheduling can employ scenario analysis methods to formulate resource scheduling schemes for multiple possible future states, calculate expected returns based on the probability of each scenario, and select the scheduling scheme with the highest expected return. This method can effectively cope with future uncertainties and improve the robustness of the system.

[0267] In addition, alternatively, a predictive scheduling and real-time feedback adjustment mechanism can be combined to continuously monitor the deviation between the actual system state and the prediction while executing the predictive scheduling plan. When the deviation exceeds the threshold, the resource allocation can be adjusted in a timely manner to achieve closed-loop control and ensure that the system can maintain good performance even when the prediction is inaccurate.

[0268] 4.5 Application Examples of Multi-Time-Scale Energy Efficiency Equalizers

[0269] The specific applications of multi-timescale energy efficiency equalizers in environmental monitoring IoT systems are as follows:

[0270] System deployment status:

[0271] 300 environmental monitoring nodes were deployed in a certain ecological protection area to monitor parameters such as temperature, humidity, air pressure, and air quality;

[0272] The nodes mainly rely on solar power, and energy acquisition is significantly affected by weather conditions;

[0273] The system needs to maintain monitoring coverage while responding to energy fluctuations and seasonal changes;

[0274] Time-scale layered implementation:

[0275] Microscale (5-minute level): Adjusting sensor sampling frequency and data transmission power to respond to instantaneous energy changes;

[0276] Mesoscale (12-hour level): Adjust node operating modes (active / dormant) to adapt to diurnal energy changes;

[0277] Macro-scale (7-day scale): Planning network topology adjustments and task reallocation to respond to changes in weather systems and seasonal transitions;

[0278] Sliding time window processing:

[0279] Microscale window size: 30 minutes, including 6 time points;

[0280] Mesoscale window size: 48 hours, including 4 time points;

[0281] Macro-scale window size: 21 days, including 3 time points;

[0282] The decision fusion adopts an index-weighted average with weights of 0.6, 0.3, and 0.1.

[0283] Conflict Coordination Case Study:

[0284] Emergency: Possible wildfire signs detected, micro-level decision-making requires increasing sampling frequency;

[0285] Conflict: Mid-level decision-making suggests reducing the number of active nodes based on energy conditions;

[0286] Coordination results: The system dynamically reallocated resources, temporarily shut down non-critical area nodes, and concentrated energy on the fire monitoring area;

[0287] Effect verification:

[0288] Compared to single-timescale decision-making, the system's energy efficiency has improved by 41%;

[0289] Even under extreme weather conditions (7 consecutive days of overcast and rainy weather), the system can still maintain 85% monitoring coverage of key areas;

[0290] Detection rates for critical events (such as wildfires, floods, and animal migrations) increased by 32%, while overall energy consumption decreased by 21%.

[0291] By employing a multi-timescale energy efficiency equalizer, the environmental monitoring network can achieve stable and sustained operation under energy-constrained conditions. It can respond quickly to immediate events and conduct strategic planning based on long-term trends, significantly improving the overall system performance and stability. This technology is particularly suitable for scenarios with significant fluctuations in energy conditions and dynamically changing importance of monitoring tasks.

[0292] Figure 5This paper presents a comparison of the energy efficiency stability of a traditional system and the performance of the multi-timescale energy efficiency equalizer of this application during long-term operation. It can be seen that the energy efficiency stability of the traditional system exhibits significant fluctuations, while the multi-timescale energy efficiency equalizer of this application maintains a stable upward trend, ultimately achieving a high level of stability. This verifies the conclusion mentioned in the technical effects of this application that "system energy efficiency stability is improved by 78%, and the temporal uniformity of energy utilization is improved by 65%".

[0293] Step 5: Construct a hybrid optimization objective based on information entropy and graph centrality

[0294] According to the embodiments of this application, this step constructs a hybrid optimization objective that comprehensively considers the value of data acquisition and system energy efficiency, providing a unified evaluation standard for energy allocation decisions.

[0295] 5.1 Information Entropy Calculation

[0296] Based on data value vector and energy allocation strategy The information entropy of the information acquired by the calculation system:

[0297] Vector of data value Normalization to probability distribution

[0298] Calculate the effective working time vector under the current energy allocation. ;

[0299] Calculate the weighted information entropy by combining the two:

[0300] The formula for calculating the weighted information entropy of the information acquired by the system is as follows:

[0301]

[0302] in, It is a node The value of normalized data (probability distribution form). It is a node Effective working time under the current energy allocation.

[0303] The specific calculation process is as follows: First, the data value of each node in the IoT system is normalized, transforming it into a probability distribution so that the sum of the normalized data values ​​of all nodes equals 1. Then, based on the current energy allocation strategy, the effective working time of each node under given energy conditions is calculated. Next, for each node, its normalized data value is multiplied by its effective working time to obtain the node's weighted contribution. Then, the logarithm of each node's weighted contribution is calculated and multiplied by the weighted contribution itself. Finally, the negative value of this product of all nodes is taken and summed to obtain the system's weighted information entropy. This weighted information entropy index comprehensively reflects the diversity and balance of information acquisition in the system. The higher the index value, the richer and more balanced the information acquired by the system, and the higher the energy utilization efficiency.

[0304] The information entropy calculation module takes data value vectors and energy allocation strategies as inputs and outputs an information entropy index that measures the diversity and balance of information acquisition in the system. The higher the index, the richer and more balanced the information acquired by the system.

[0305] Optionally, in some implementations, advanced information theory metrics such as conditional information entropy or mutual information can be used to more accurately characterize the quality of information acquired by the system. For example, conditional information entropy can be introduced. This is used to measure the uncertainty of data value given a known system relationship graph, thereby better guiding the optimization of energy allocation strategies.

[0306] Furthermore, it should be understood that the information entropy calculation in this application can also integrate the time dimension to construct a time-series information entropy model, evaluate the system's information acquisition performance over a period of time, and not be limited to static evaluation at a single point in time, thus providing a more comprehensive optimization objective.

[0307] 5.2 Graph Centrality Indicators

[0308] Based on system relationship diagram Calculate various graph centrality indices:

[0309] Degree centrality: measures the number of direct connections between nodes;

[0310] Proximity centrality: measures the average distance from a node to all other nodes;

[0311] Betweenness centrality: measures how frequently a node acts as a relay for the shortest path between other nodes;

[0312] Eigenvector centrality: A centrality measure that takes into account the importance of the connected nodes;

[0313] These centrality indicators are weighted and fused to form a comprehensive centrality score for the node. .

[0314] The graph centrality calculation module takes the system relationship graph as input and outputs the comprehensive centrality score of each node, reflecting the importance and influence of the node in the system structure.

[0315] 5.3 Hybrid Optimization Objective Function

[0316] Construct a hybrid optimization objective function that simultaneously considers information entropy and graph centrality:

[0317] The mathematical expression for the hybrid optimization objective function is:

[0318]

[0319] in, It is a balance parameter. It is the weighted information entropy defined earlier. It is a node The overall centrality score, It is a normalized energy allocation. It is the value of normalized data. It is a node Energy efficiency performance indicators.

[0320] The specific calculation process is as follows: First, the weighted information entropy of the system under the current data value distribution and energy allocation strategy is calculated. This information entropy represents the diversity and balance of information acquired by the system. Then, based on the system relationship graph, the comprehensive centrality score of each node is calculated. This score reflects the importance of the node in the network structure. Next, for each node, its centrality score, normalized data value, and energy efficiency performance index are multiplied together, and this product is summed for all nodes to obtain the system's structural energy efficiency score. Finally, a balance parameter is introduced, with a value between 0 and 1. The information entropy score is multiplied by the balance parameter, and the structural energy efficiency score is multiplied by one minus the balance parameter. The two parts are then added together to obtain the final hybrid optimization objective function value. This function balances the diversity of information acquisition with the stability and energy efficiency of the network structure, providing a comprehensive evaluation standard for the system's energy allocation decisions.

[0321] The hybrid optimization objective function module takes the complete state of the system (data value, system structure, energy allocation) as input and outputs a comprehensive score that considers both the diversity of information acquisition and the stability and energy efficiency of the system structure.

[0322] 5.4 Adaptive Parameter Adjustment

[0323] Construct an adaptive parameter adjustment algorithm to dynamically adjust the balance parameters in the hybrid optimization objective based on the system's operating state. :

[0324] Historical data on the energy efficiency performance and information acquisition quality of the monitoring system;

[0325] Identify the different stages and modes of system operation;

[0326] Using reinforcement learning methods, adjust based on historical performance value;

[0327] The adaptive parameter adjustment algorithm takes historical system operating data as input and outputs the optimal balance parameter settings, enabling the system to automatically adjust the optimization focus according to operating conditions and environmental changes, and find the best balance between data value acquisition and energy efficiency optimization.

[0328] Figure 6 The paper demonstrates a comparison of energy consumption between the proposed method and traditional methods in different application scenarios, assuming the acquisition of the same data value. Data shows that the proposed method achieves significant energy consumption reduction in all test scenarios, averaging approximately 42%, which verifies the conclusion mentioned in the technical effects section of this application that "energy consumption is reduced by 42% under the same data value requirements."

[0329] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An energy efficiency synergistic optimization method based on the Internet of Things, characterized in that, Includes the following steps: Step 1: Construct a value-structure-energy efficiency ternary coupling model, which includes a data value vector, a system relationship diagram, and an energy allocation strategy vector; Step 2: Construct a two-layer recursive optimization architecture, which includes a perception layer and a relation layer. The perception layer uses cognitive computing methods to evaluate the value of IoT data, and the relation layer uses graph computing methods to model the dependencies between IoT devices, and realizes bidirectional information flow and recursive optimization between the perception layer and the relation layer. The perception layer uses cognitive computing methods to evaluate the value of IoT data, and specifically includes the following components: Value assessment unit: Calculates the value of raw data using a multi-dimensional scoring model, considering factors such as data completeness, timeliness, signal-to-noise ratio, and application relevance. The scoring model employs a weighted summation method. ,in It is the data value score of the i-th IoT node; d represents the total number of feature dimensions used in data value assessment; It is the first Dimension weight coefficients, It is a normalization function that normalizes the original characteristics of different dimensions. Mapped to a uniform scoring range [0,1]; Context-aware module: Analyzes the spatiotemporal environment of data collection and adjusts the weights for value assessment; Historical Value Modeler: Based on historical data analysis of value patterns, it constructs time-series prediction models to support predictive resource scheduling; The perception layer receives raw data and metadata from IoT devices as input, and through the collaborative processing of the aforementioned components, outputs a real-time updated data value vector. Each element in this vector corresponds to a value score of the data generated by an IoT node; The relationship layer uses graph computing methods to model the dependencies between IoT devices, and specifically includes the following components: System Relationship Graph Builder: Constructs an initial system relationship graph based on IoT topology and functional dependencies. Nodes represent devices, and edges represent dependencies or collaborations between devices; System Relationship Graph Evolution Engine: Dynamically updates the system relationship graph structure based on real-time data flow and collaboration patterns, adjusting edge weights or adding / deleting edges; System graph analyzer: calculates various graph centrality indicators, identifies key nodes and paths, and provides input for energy allocation decisions; The relationship layer receives communication logs, function call records, and collaboration status data between devices as input, and outputs a dynamically updated system relationship diagram. This diagram accurately reflects the real-time dependencies and collaboration strength among IoT devices; The following steps are taken to achieve bidirectional information flow and recursive optimization between the perception layer and the relation layer: Structure-guided value assessment: Graph structure information of the relational layer This information is transmitted to the perception layer, influencing the data value assessment process. The higher the centrality of a node in the graph, the higher its data value weight, reflecting the impact of structural importance on data value. Value-Driven Structural Evolution: Data Value Information in the Perception Layer This information is passed to the relation layer to guide the dynamic adjustment of the graph structure; nodes with high data value are more likely to establish strong connections, forming efficient collaborative clusters. Step 3: Introduce adaptive tensor decomposition to integrate data value, system relationships, and time dimension into a three-dimensional tensor to extract system evolution patterns; Step 4: Implement a multi-timescale energy efficiency equalizer by integrating decisions from different timescales using a sliding time window method; Step 5: Construct a hybrid optimization objective based on information entropy and graph centrality to provide a unified evaluation standard for energy allocation decisions; In the value-structure-energy efficiency ternary coupling model, the data value vector and system relationship diagram are dynamically updated through a value-structure mutual feedback mechanism.

2. The energy efficiency synergistic optimization method based on the Internet of Things according to claim 1, characterized in that, The value-structure feedback mechanism includes: The mapping function based on the system structure updates the data value utilizes a graph neural network to perform message passing on the system relationship graph, calculates the importance of each node in the network, and weights and fuses the node importance with the current data value. The mapping function updates the system structure based on data value. The mapping function calculates the collaboration potential between nodes and updates the weights of edges in the system relationship graph.

3. The energy efficiency synergistic optimization method based on the Internet of Things according to claim 1, characterized in that, The adaptive tensor decomposition method includes: Construct a three-dimensional tensor, where the first dimension of the three-dimensional tensor represents the Internet of Things node, the second dimension represents the relationship between nodes, and the third dimension represents the evolution sequence at time points; An adaptive rank selection algorithm is used to determine the optimal rank of the tensor decomposition. The three-dimensional tensor is decomposed into a low-rank representation by applying CP decomposition or Tucker decomposition methods; Use tensor decomposition results for data completion and prediction.

4. The energy efficiency synergistic optimization method based on the Internet of Things according to claim 1, characterized in that, The multi-time-scale energy efficiency equalizer includes: The system runtime is divided into time scales of micro, meso, and macro scales. A time window is maintained for each time scale, and the size of the time window is proportional to the time scale; The decision results from different time windows are combined using an exponential weighting method. Establish a conflict decision-making coordination mechanism to address contradictions between decisions made at different time scales.

5. The energy efficiency synergistic optimization method based on the Internet of Things according to claim 1, characterized in that, The hybrid optimization objective based on information entropy and graph centrality includes: Based on data value vectors and energy allocation strategies, calculate the information entropy of the information acquired by the system. Based on the system relationship graph, multiple graph centrality indices are calculated, and these centrality indices are weighted and fused to form a comprehensive centrality score for the nodes; Construct a hybrid optimization objective function that simultaneously considers information entropy and graph centrality; An adaptive parameter adjustment algorithm is adopted to dynamically adjust the balance parameters in the hybrid optimization objective according to the system operating status.

6. The energy efficiency synergistic optimization method based on the Internet of Things according to claim 2, characterized in that, The calculation process of the mapping function based on the data value update system structure includes: Calculate the similarity of data value between nodes; Calculate the complementarity index between nodes; Apply spatial distance attenuation factor; Update the edge weights in the adjacency matrix; Sparsification is applied to ensure that the relationship graph has a good sparse structure.

7. The energy efficiency synergistic optimization method based on the Internet of Things according to claim 4, characterized in that, The multi-timescale energy efficiency equalizer also includes predictive resource scheduling, which, based on the system evolution model and prediction results provided by tensor decomposition, achieves the following functions: Adjust resource allocation plans in advance based on predicted changes in the distribution of data value; Identify the regions and time points where potentially high-value data will appear, and pre-allocate sufficient resources; Predict potential changes in system architecture and optimize network topology and communication paths in advance.

8. The energy efficiency synergistic optimization method based on the Internet of Things according to any one of claims 1 to 7, characterized in that, The method is applicable to at least one of the following scenarios: Smart city sensing network; Industrial Internet of Things (IoT) systems; Large-scale distributed Internet of Things (IoT) infrastructure; Internet of Things (IoT) systems in dynamic task environments.

9. An energy efficiency collaborative optimization system based on the Internet of Things, characterized in that, include: The device layer, including IoT nodes, is responsible for data acquisition, preprocessing, and executing energy efficiency control commands; Edge layer: Computing nodes deployed at the edge of the network are responsible for data processing, value assessment, and relationship modeling in local areas; The central layer, the core computing unit deployed on cloud servers or data centers, is responsible for global optimization, tensor decomposition, multi-timescale integration, and hybrid optimization target computation. The system employs the IoT-based energy efficiency synergistic optimization method as described in any one of claims 1 to 8, thereby achieving the synergistic evolution of data value and system structure, and solving the challenges of resource allocation and overall coordination in IoT energy efficiency optimization.

Images