Cellular network base station energy efficiency collaborative optimization method based on multi-task optimization

Abstract

The application discloses a cell network base station energy efficiency cooperative optimization method based on multi-task optimization, which takes the ratio of total energy consumption to total capacity in a region as the target function of the task, identifies the energy consumption mode similarity of each region in the same period through time slice blocking and dynamic similarity measurement, only performs knowledge transfer on similar blocks, adaptively selects a first-order or second-order difference trend model according to local sensitivity, combines direction trust degree adjustment and variable field of view sampling, generates a conservative power adjustment strategy for a high-sensitivity region, and generates a sleep scheme for a smooth region. The application can significantly reduce the energy consumption per bit of the whole network, control the switching failure rate, shorten the optimization period, guarantee the service quality and operation reliability, and provide efficient energy efficiency optimization support for super-dense networking of mobile communication networks.

Description

Technical Field

[0001] This invention belongs to the technical field of mobile communication networks and multi-task optimization methods, specifically relating to a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization. Background Technology

[0002] With the large-scale deployment of fifth-generation (5G) and future sixth-generation (6G) mobile communication networks, the density of base stations is constantly increasing, and heterogeneous network architectures are becoming the mainstream. At the same time, network traffic exhibits high dynamism and non-uniformity in both time and space: peak traffic periods, load fluctuations, and channel interference characteristics differ significantly across different geographical areas such as commercial areas, residential areas, transportation hubs, and suburbs. This heterogeneity and dynamism lead to a sharp increase in the total energy consumption of all base stations, becoming a major component of operators' operating costs (OPEX), and posing a severe challenge to green communications under dual-carbon objectives.

[0003] In the field of base station energy efficiency optimization, existing methods are generally divided into two categories. The first category is single-region independent optimization, which establishes an independent energy consumption model for each geographical region and solves it separately using heuristic algorithms. Although this type of method can adapt to local characteristics, it ignores the similar energy-saving patterns that may exist between different regions, resulting in each region starting from scratch, with slow convergence speed, serious waste of computing resources, and difficulty in efficiently reusing existing experience in multi-region collaborative scenarios at the operator level. The second category is centralized global optimization, which models and solves all regions as a whole. Although this method can consider the coupling constraints between regions, it faces the curse of dimensionality and the contradiction of real-time performance: the decision space grows linearly with the number of regions, and the centralized uploading of privacy data between regions will cause communication overhead and data security risks, while it cannot flexibly cope with the local dynamic changes of each region.

[0004] Multi-task optimization (MTO) is a cutting-edge paradigm that has emerged in the field of evolutionary computing in recent years. Its core idea is to simultaneously solve multiple related optimization tasks, accelerating the convergence process of each task and improving the quality of solutions by mining and transferring common knowledge or complementary information between tasks. In the scenario of base station energy efficiency optimization, each geographical region can be defined as an independent task, with the goal of all tasks being to minimize the total energy consumption within that region. By identifying the similarity of energy consumption patterns in different regions during specific time periods, energy-saving strategies successfully optimized in one region can be transferred to other similar regions, thereby significantly reducing the global search overhead.

[0005] It is worth noting that the key to solving the problem of coordinated energy efficiency optimization of cellular network base stations lies in accurately identifying the structural similarities of different regions in the same time block while preserving the uniqueness of each region, and realizing safe and adaptive knowledge transfer based on local evolution trends, thereby accelerating global convergence and avoiding negative transfer. In view of this, this invention proposes a multi-task optimization method for coordinated energy efficiency of cellular network base stations based on adaptive block prediction and trust adjustment (ABPT-MTO). By generating high-quality energy-saving candidate strategies for base stations, it can significantly improve the overall energy efficiency optimization efficiency and strategy security of multi-regional base station clusters. Summary of the Invention

[0006] The purpose of this invention is to provide a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization, which solves the problem in the prior art that it is difficult to achieve efficient reuse of cross-regional common energy-saving experience while ensuring regional independence and security in the coordinated optimization of energy efficiency of cellular base stations in multiple regions.

[0007] The technical solution adopted in this invention is: a cellular network base station energy efficiency collaborative optimization method based on multi-task optimization, including initializing parameters and defining each region as an independent task; constructing a minimum per-bit energy consumption objective function containing switching power carriers; collecting data from each base station and dividing it into blocks according to time slices; extracting block description vectors and block weights through principal component analysis; then constructing overall task similarity based on inter-block distance and dynamic threshold, and selecting similar task pairs and blocks; subsequently, selecting a first-order or second-order difference model to fit the normalized trend vector according to the block's local sensitivity; then updating the direction confidence based on the improvement rate; combining the trend and zoom field of view sampling to generate a solution and migrate it to the target block; finally, adaptive differential evolution for dissimilar blocks, evaluating and archiving candidate strategies, and iterating cyclically until the maximum number of evaluations is reached.

[0008] The present invention is further characterized by including the following steps: Step 1: Initialize parameters, including the population for each individual task. Population size Maximum number of evaluations Number of blocks (B), initial sensitivity cache Historical Center Matrix List Largest historical window ; Step 2: Define each geographical region in the cellular network as an independent optimization task. For each task, define a decision variable vector that includes the on / off / sleep flags, transmit power, and carrier allocation ratio of each base station in the corresponding region. Construct the objective function for each task as minimizing the ratio of total energy consumption to total capacity in the region. Step 3: Determine whether the preset termination condition is met. If the preset termination condition is met, output the optimal solution for each task; otherwise, proceed to step 4. Step 4: Collect data from each base station and divide it into blocks according to time slices, then extract the block description vector and block weight; Step 5: Based on the block description vectors and block weights obtained in Step 4, determine the Euclidean distance similarity between blocks and apply a dynamic threshold to construct the overall task similarity; Step 6: Select similar task pairs based on the overall task similarity matrix obtained in Step 5, and determine the set of time segments with high similarity between similar task pairs; Step 7: Perform trend fitting on the similar task pairs obtained in Step 6 according to time blocks to obtain the normalized trend; Step 8: Transfer valuable solutions from the source task time slice to the target task block and advance and sample to generate candidate solutions based on the normalization trend at the time slice level obtained in Step 7. Step 9: Inject the final candidate solutions obtained in Step 8 into the offspring pool of the target task according to time slices. For time slices of two tasks with low similarity, they undergo independent evolution. During the independent evolution process, based on the normalized local sensitivity obtained in Step 7, the mutation rate and crossover rate of the local differential evolution algorithm are adaptively adjusted. Several generations of local evolution are carried out and an elite selection strategy is adopted to retain the best individual. Finally, the base station energy efficiency strategy is provided to the controller of each region. Step 10: Evaluate the candidate strategies generated in Step 9 on the objective function constructed in Step 2 according to the evaluation strategy, and save the evaluation results to the local archive. Repeat steps 3 to 9 until the maximum number of evaluations is reached. The optimization process is now complete.

[0009] Step 2 is as follows: Each geographic region in the cellular network is defined as an independent optimization task, totaling... One task; for the task , Define the decision variable vector This includes the on / off / sleep flags, transmit power, and carrier allocation ratios of each base station within the corresponding area:

[0010] In the formula, For the task The number of base stations in the area For base stations Switch / sleep flag, For base stations The transmission power, For base stations Carrier allocation ratio, For transpose; Build each task The objective function to be optimized To minimize the ratio of total energy consumption to total capacity in the region:

[0011] In the formula, It is a base station Static power consumption, It refers to the power amplifier efficiency. It is channel gain. It is interference power. It is noise power. It is a normal number.

[0012] Step 4 specifically includes the following steps: Step 4.1, for each task With each base station it contains Raw observation data was collected on a per-generational basis, including data from each base station. Static power consumption Transmission power Carrier allocation ratio Channel gain Interference power User throughput and total regional energy consumption; Step 4.2: Divide a 24-hour day into several time blocks. Each time slice represents a period of time within a day that has relatively stable traffic characteristics and optimization features, and the individual index corresponding to each time slice is stored in... middle; Step 4.3: Solving the solution set for each time slice Principal component analysis was performed, where Indicates the number of samples within the block. As a decision-making dimension, For the set of real numbers; before extraction Each principal component score matrix is ​​then calculated. Mean of dimensional data in each direction Standard deviation and skewness Combined into a length of 3 Description vector ; Step 4.4: Use the sum of the standard deviations of the principal components in the description vector of each time slice as the original block weights. Then, the original block weights of all time slices are normalized to obtain... .

[0013] Step 5 specifically includes the following steps: Step 5.1: Based on the description vectors of each time slice obtained in Step 4.3 Calculate any two tasks and In the Euclidean distance on a time slice :

[0014] in, and Tasks With the task In the Description vectors on each block Representing vectors Norm; Step 5.2, collect the first The set of Euclidean distances for all task pairs on a given time slice. Take set The mean scrambling action in the process is the similarity threshold for each time block. ; Step 5.3: Based on the normalized block weights obtained in step 4.4 Computational tasks With the task Overall similarity :

[0015] In the formula, This is an indicator function.

[0016] Step 7 specifically includes the following steps: Step 7.1: Based on the similar task pairs selected in Step 6 and their corresponding time slice sets, calculate the task... No. The time slice in the first The central position of the generation serves as the centroid of all individuals within the time segment. ; Step 7.2: Apply perturbation to each individual within the time slice to obtain... Then calculate the individual rate of change. And the local sensitivity at the time-block level is averaged to obtain ; Step 7.3: Normalize the local sensitivity at the time slice level to obtain... :

[0017] In the formula, and These are the minimum and maximum sensitivities among all blocks in the current generation, respectively. It is a positive number; Step 7.4: Determine the value of the normalized local sensitivity. When smoothing the region, a first-order difference model is used to calculate the trend vector. :

[0018] Conversely, the trend vector is calculated using a second-order difference model. :

[0019] Finally, the calculated trend vector is normalized to obtain... This refers to the original predicted trend direction of the current block in the current generation.

[0020] In step 7.4 when At that time, among them The trend vector is calculated using a first-order difference model to represent the sensitivity threshold; otherwise, a second-order difference model is used to calculate the trend vector.

[0021] Step 8 specifically includes the following steps: Step 8.1, Task The Trust level for maintenance direction of time fragments Define the optimal objective value improvement amount for the g-th generation. Then improve the rate of change for:

[0022] Step 8.2, based on the improved rate of change Update direction trust level, if If the optimization effect of the current trend direction is further enhanced, then the direction confidence level is updated as follows:

[0023] In the formula, Used to limit trust levels Within the range, Learning rate based on trust level Limit the update range; Conversely, a fixed-amplitude decay is applied to the corresponding directional confidence level; when When the field of view is below a preset threshold, a magnification marker is triggered. Step 8.3, in the source task time slice Search among individuals within a predetermined proportion, and select the individual with the largest decrease in the target value over the last two generations as the representative solution. ; Step 8.4: Based on the ratio of the number of function evaluations used to the maximum number of evaluations, divide the algorithm process into three evolutionary stages: early, middle, and late. Step 8.5: Based on zoom imaging, use the search field of view (FOV) as a quantitative indicator of the imaging field of view, and adjust the stage adjustment factor accordingly. As the control parameter for focal length adjustment, the stage adjustment factor is determined based on the evolutionary stages divided in step 8.4. ; Step 8.6: Based on stage adjustment factors And the direction trust level updated in step 8.2 Calculate the adaptive advance distance of the target mission time slice : In the formula, Based on the scale of advancement; directional confidence level for the target task time segment; Step 8.7: Based on the normalized trend vector obtained in Step 7 The representative solution of the source task time slice obtained in step 8.3 and the adaptive propulsion distance obtained in step 8.6 Generate migration center point :

[0024] In the formula, This is the normalized trend vector for the time slice of the target task; Step 8.8: First, based on the stage adjustment factor obtained in Step 8.5... And the direction trust level updated in step 8.2 Calculate the current field of view. :

[0025] In the formula, Use this as the basic search metric; if step 8.2 confirms... If the field of view is below the preset trigger threshold, a field of view magnification operation is performed and the sampling radius is increased; otherwise, the original field of view range is maintained. The migration center point generated in step 8.7 The surrounding area is distributed in a Gaussian pattern within the current field of view. Random sampling is performed within the sample to generate candidate solutions; Step 8.9: When determining the direction confidence level in step 8.2... If the field of view is below the preset trigger threshold and the field of view magnification operation has been performed in step 8.8, the zoom repositioning mechanism is simultaneously activated, based on the optimal individual... and migration center point Reconstructing the direction to obtain a new direction :

[0026] Step 8.10: Apply the new direction obtained in Step 8.9 Normalized trend vector of the target task time slice Merge and normalize:

[0027] In the formula, The directional fusion coefficient; Then use the fused direction Replace the part in step 8.7 Then, repeat step 8.7 to generate a new migration center point, and jump to step 8.8 to sample within the current field of view, and output the sampled candidate solution as the final candidate solution.

[0028] The beneficial effects of this invention are as follows: This invention is based on a multi-task optimization method for cellular network base station energy efficiency collaborative optimization. By defining each region as an independent optimization task and using time-slice partitioning and dynamic similarity determination, it accurately identifies common energy-saving patterns across regions while ensuring regional data security and independence. Based on local sensitivity, it adaptively selects a first-order or second-order differential trend model, combined with directional confidence adjustment and zoom field-of-view sampling mechanism, to achieve safe and adaptive cross-task knowledge transfer, effectively suppressing negative transfer. Furthermore, by dynamically adjusting local differential evolution parameters based on normalization sensitivity, it significantly reduces the energy consumption per bit across the entire network and greatly shortens the optimization cycle while ensuring network coverage and service quality. At the same time, it effectively controls the handover failure rate caused by policy changes, thus providing an energy efficiency optimization scheme for multi-regional heterogeneous base station clusters that takes into account regional autonomy, data security, and global collaboration. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the energy efficiency collaborative optimization method for cellular network base stations based on multi-task optimization according to the present invention. Figure 2 This is a schematic diagram of the framework of the cellular network base station energy efficiency collaborative optimization system based on multi-task optimization of the present invention. Detailed Implementation

[0030] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0031] This invention provides a collaborative optimization method for cellular network base station energy efficiency based on multi-task optimization. Addressing the problems of slow convergence and difficulty in reusing cross-regional experience in existing cellular network base station energy efficiency optimization methods, such as the curse of dimensionality and regional privacy leakage risks associated with independent optimization in a single region, this invention aims to preserve the uniqueness of each region while fully utilizing common energy-saving patterns across regions, accelerating global convergence and effectively suppressing negative migration. By defining each geographical region as an independent task, adaptive knowledge transfer between similar regions is achieved through block partitioning and dynamic similarity determination. Combined with local sensitivity-driven trend modeling (first-order / second-order difference) and directional trust adjustment mechanisms, this method effectively reduces cross-regional computational overhead and overall network energy consumption while ensuring coverage and throughput. It provides an efficient and secure technical solution for the collaborative optimization of 5G / 6G base station sleep, power allocation, and carrier configuration.

[0032] The energy efficiency optimization problem of cellular network base stations is essentially a multi-task optimization problem. Each geographical region is defined as an independent task, and the function to be optimized is shown in equation (1):

[0033] For the task Its decision variable vector It includes the on / off / sleep flags, transmit power, and carrier allocation ratio of each base station in the area, as defined in equation (2):

[0034] in For the task The number of base stations in the area For base stations Switch / sleep flag, It is a base station The transmission power, Assigning a carrier ratio, The symbol is the transpose. The objective function for each task is defined as minimizing the ratio of total regional energy consumption to total capacity (i.e., the energy required per bit transmission), as shown in equation (3):

[0035] in It is a base station Static power consumption, It refers to the power amplifier efficiency. It is channel gain. It is interference power. It is noise power. It is a very small positive number, used to prevent the denominator from being zero.

[0036] Due to significant differences in base station types, user density, traffic patterns, channel conditions, and interference environments across different regions, the objective function landscape for each task exhibits high heterogeneity: for example, during peak hours in commercial areas, a trade-off between high throughput and high energy consumption is needed, while during late-night hours in suburban areas, the focus is more on reducing static power consumption under low load. Each task independently maintains its own population and achieves collaborative optimization through a multi-task knowledge transfer mechanism in subsequent steps. Figure 1 As shown, the cellular network base station multi-regional energy efficiency collaborative optimization system provided by this invention first inputs the scenario and environment: collecting base station status data in each region, dividing the region into time periods, and performing sensitivity prediction; then it enters the multi-task optimization algorithm module based on block prediction, adaptively dividing each task into blocks, constructing a task similarity matrix, and determining the block numbers of similar blocks between each pair of tasks; for blocks determined to be similar, a trend-guided migration strategy is initiated, fitting the trend through the center movement trajectory within the block, selecting representative solutions from the source task blocks to migrate to the target task block, and performing advancement and sampling to generate candidate solutions; for dissimilar blocks, the differential evolution parameters are adaptively adjusted according to the block's sensitivity for local evolution; finally, after environmental selection and population update, when the maximum number of evaluations is reached, the optimal energy-saving strategy set for each region's base station is output and executed. Figure 2 As shown, this invention first collects base station status data from various regions and divides them into blocks according to time periods. Combining this with traffic fluctuation prediction, blocks are marked as sensitive or regular areas, and control parameters are initialized. Subsequently, in the cross-task block similarity measurement stage, block features are extracted. For task pairs meeting the similarity threshold, a collaborative channel is activated to prepare for migration; otherwise, an independent optimization channel is entered for local optimization. In the trend-guided base station strategy migration stage, the target block receiving the migration strategy selects either a conservative or standard mode based on its own sensitivity to generate an optimization scheme. Finally, through sensitivity-driven local evolution, the migration strategy and independent exploration strategy are integrated, differential evolution parameters are dynamically adjusted, and iterative generation and distribution of base station sleep, power allocation, and carrier configuration schemes are achieved, realizing multi-regional energy efficiency collaborative optimization.

[0037] The cellular network base station multi-regional energy efficiency collaborative optimization system based on this invention defines base station clusters in different geographical areas such as commercial areas, residential areas, transportation hubs, and suburbs as independent optimization tasks. Utilizing a time-slice block mechanism and dynamic similarity measurement, it automatically identifies the similarity of energy consumption patterns in different regions during the same time period and initiates cross-regional knowledge transfer only for similar blocks. The system adaptively selects a first-order or second-order differential trend model based on local sensitivity, combining directional confidence adjustment and zoom field-of-view sampling to generate conservative power adjustment strategies for highly sensitive areas (such as sudden passenger flow scenarios) to avoid coverage degradation, and generates sleep schemes for stable areas (such as suburban areas at night) to minimize static power consumption. After actual deployment, this collaborative mechanism can significantly reduce the energy consumption per bit across the entire network, effectively control the handover failure rate caused by policy changes, and significantly shorten the optimization cycle compared to independent optimization in each region. While ensuring service quality and operational reliability, it provides efficient and feasible energy efficiency optimization support for 5G / 6G ultra-dense networking.

[0038] Example 1 This invention provides a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization, comprising the following steps: Each geographic region in the cellular network is defined as an independent optimization task, totaling... One task. For the task... Its decision variable vector This includes the on / off / sleep flags, transmit power, and carrier allocation ratios of each base station within the area, defined as follows:

[0039] in For the task The number of base stations in the area For base stations Switch / sleep flag, It is a base station The transmission power, Assign a proportion to each carrier, for each task The objective function is defined as minimizing the ratio of total regional energy consumption to total capacity (i.e., the energy required per bit transmission):

[0040] in It is a base station Static power consumption, It refers to the power amplifier efficiency. It is channel gain. It is interference power. It is noise power. This is a minimal positive constant used to prevent the denominator from being zero. The smaller the objective function value, the higher the energy efficiency of the region. Therefore, the function to be optimized is:

[0041] In the scenario of coordinated optimization of energy efficiency for cellular network base stations, factors such as service traffic, time-of-use electricity pricing, channel interference, and user distribution in different regions exhibit high dynamism and non-uniformity in spatiotemporal dimensions. Therefore, base station energy-saving strategies need to possess sensing and cross-regional coordinated optimization capabilities. The multi-task optimization method proposed in this invention can construct an adaptive scheduling framework for base station clusters in different geographical regions, enabling them to dynamically adjust the base station's sleep flag, transmit power, and carrier allocation ratio, thereby achieving the optimal balance between regional energy consumption and network capacity. This method not only improves overall network energy efficiency but also effectively reduces the risk of coverage blind spots and handover failures caused by policy changes, enhancing network operational stability and energy utilization efficiency.

[0042] The core improvement of existing multi-task optimization methods in this invention lies in proposing a migration strategy based on block trend perception and trust adjustment. Blocks are divided into sensitive and regular regions based on local sensitivity. Sensitive regions employ a second-order difference model to capture the acceleration changes in energy consumption curves, while regular regions use a first-order difference model to avoid overfitting. Simultaneously, directional trust and zoom field-of-view mechanisms are introduced to dynamically adjust migration intensity and sampling range, and local differential evolution parameters are adaptively adjusted based on normalized local sensitivity. This method can dynamically generate personalized energy-saving strategies for each base station cluster, ensuring optimal operation with sleep combinations, power configurations, and carrier allocation schemes at different times based on service load and interference environment, thereby maximizing system energy efficiency. This algorithm can obtain optimal scheduling schemes under different service scenarios and network conditions, providing key technical support for the efficient, economical, and reliable operation and maintenance of 5G / 6G ultra-dense networks.

[0043] Example 2 This invention provides a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization, comprising the following steps: Step 1: Parameter initialization, including population for each task. Population size Maximum number of evaluations Number of blocks B, initialize cache Historical Center Matrix List Largest historical window And several other parameters.

[0044] Step 2: Define each geographic region in the cellular network as an independent optimization task, totaling... One task. For the task... Its decision variable vector This includes the on / off / sleep flags, transmit power, and carrier allocation ratios of each base station within the area, defined as follows:

[0045] in For the task The number of base stations in the area For base stations Switch / sleep flag, It is a base station The transmission power, Assign a proportion to each carrier, for each task The objective function is defined as minimizing the ratio of total regional energy consumption to total capacity (i.e., the energy required per bit transmission):

[0046] in It is a base station Static power consumption, It refers to the power amplifier efficiency. It is channel gain. It is interference power. It is noise power. This is a minimal positive constant used to prevent the denominator from being zero. The smaller the objective function value, the higher the energy efficiency of the region. Therefore, the function to be optimized is:

[0047] Each task maintains its own population independently and achieves collaborative optimization through a multi-task knowledge transfer mechanism in subsequent steps.

[0048] Step 3: Termination condition judgment: If the preset termination condition is met, output the optimal solution for each task; otherwise, proceed to step 4.

[0049] Step 4: Pre-segment and collect data from base stations in multiple regions, construct description vectors for each block, and calculate the weights of each block.

[0050] Step 5: Determine the inter-block similarity and apply a dynamic threshold to construct the overall task similarity.

[0051] Step 6: Select similar task pairs and similar block sets from the overall task similarity matrix obtained in the previous step.

[0052] Step 7: Perform trend fitting (algebraic first / second-order difference) on similar task pairs to obtain the normalized trend. .

[0053] Step 8: Transfer valuable solutions from the source block to the target block and perform a series of advancements and sampling based on block-level trends to generate candidate solutions.

[0054] Step 9: Inject candidate strategies into the offspring pool of the target task in blocks. For two blocks with low similarity, perform self-evolution and adaptively adjust the local differential evolution parameters based on local sensitivity. and Through several generations of local evolution and elite selection, the system ultimately provides base station energy efficiency strategies to controllers in each region.

[0055] Step 10: Evaluate the candidate strategies generated by multi-task optimization on the objective function of cellular network base station energy efficiency co-optimization according to the evaluation strategy, and save them to the local archive. Repeat steps 3-9 until the maximum number of evaluations is reached. Optimization complete.

[0056] Example 3 This invention provides a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization. Based on Example 2, step 4 is preferably performed using the following steps: Step 4.1: For each task (area) With each base station it contains Raw observation data was collected on a per-generational basis, including data from each base station. Static power consumption Transmission power Carrier allocation ratio Channel gain Interference power User throughput and total regional energy consumption, etc.

[0057] Step 4.2: Divide a 24-hour day into several time blocks. (00:00-6:00, 6:00-9:00, 9:00-17:00, 17:00-20:00, 20:00-24:00), each time slice represents a period of time with relatively stable traffic characteristics and optimization features within a day, and the individual index corresponding to each slice is stored in... This allows for quick location of subsequent modules.

[0058] Step 4.3: Solution set for each block ,in Indicates the number of samples within the block. As a decision-making dimension, For the set of real numbers, perform principal component analysis. Before extraction Each principal component score matrix is ​​then calculated. Mean of dimensional data in each direction Standard deviation and skewness Combined into a length of 3 Description vector This description vector is used in subsequent steps to measure the similarity between any two tasks on the same time slice.

[0059] Step 4.4: Use the sum of the standard deviations of the principal components in each block description vector as the original block weights. Then, the weights of all blocks are normalized to obtain .

[0060] Example 4 This invention provides a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization. Based on Example 2, step 5 preferably includes the following steps: Step 5.1, for the task With the task In the The distances on each block are calculated using Euclidean distance:

[0061] in, and Tasks With the task In the Description vectors on each block Representing vectors Norm.

[0062] Step 5.2, collect the first The set of distances between all task pairs in each block Take the mean and add perturbation to set the similarity threshold. .

[0063] Step 5.3, obtained from step 4.4 Afterwards, the task With the task The overall similarity is:

[0064] in This is an indicator function.

[0065] Example 5 This invention provides a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization. Based on Example 2, step 7 is preferably performed using the following steps: Step 7.1, Calculation Task No. The block in the The central position of the generation serves as the centroid of all individuals within the block. .

[0066] Step 7.2: Apply a small perturbation to each individual within the block to obtain... Then calculate the individual rate of change. The block-level local sensitivity is averaged to obtain .

[0067] Step 7.3, normalization yields... :

[0068] in and It is divided into minimum and maximum sensitivity among all blocks in the current generation.

[0069] Step 7.4, when At lower levels (regional smoothing), i.e. (in When representing the sensitivity threshold (used to distinguish between regions of gradual change and regions of drastic change), a first-order difference model is used:

[0070] when When the value is high (due to drastic regional changes), a second-order difference model is introduced:

[0071] Finally, the trend vector is normalized to obtain... This vector represents the original predicted trend direction of the current block in the current generation, and is used for direction fusion and migration guidance in subsequent steps.

[0072] Example 6 This invention provides a method for coordinated optimization of energy efficiency of cellular network base stations based on multi-task optimization. Based on Example 2, step 8 preferably includes the following steps: Step 8.1, Task The Trust in the direction of block maintenance Initially 0.5. Define the optimal objective value improvement amount for the g-th generation. That is, the optimal objective function value of the (g-1)th generation minus the optimal objective function value of the gth generation, then the improvement rate of change is defined as:

[0073] in, This indicates that the optimization effect has been enhanced, while Indicates a reduction; Step 8.2, if the improvement rate If the optimization effect of the current trend direction is further enhanced, then the direction confidence level is updated as follows:

[0074] in, The learning rate for trust level is set to 0.1. Limit the update range. Used to limit trust levels Within the range.

[0075] Conversely, the corresponding directional confidence level is decayed with a fixed amplitude step size of 0.1. When This triggers the field-of-view magnification marker.

[0076] Step 8.3, Solution Set Zhongyuan Task Block The representative solution The individual with the largest decrease in target value in the last two generations (search only in the top 30% of individuals).

[0077] Step 8.4: Evaluate the proportion based on the function. Divide into three evolutionary stages: Initial stage: ( <0.3), Mid-term: 0.3≤ <0.7, later stage: ≥0.7.

[0078] Step 8.5: Drawing inspiration from the zoom imaging principle in photographic optics, which continuously adjusts the field of view by changing the focal length, the search field of view (FOV) is used as a quantitative indicator of the imaging field of view, and the stage adjustment factor is... As a control parameter for focal length adjustment, the stage adjustment factor is calculated based on the evolutionary stage determined in step 8.4. The correspondence between its value and the evolutionary stage is as follows: in the early stages of evolution, it takes... This corresponds to a large field of view to enhance global exploration capabilities; in the middle stage of evolution, it takes... Maintain a balance between exploration and development; take advantage of later stages of evolution. It corresponds to a small field of view to focus on fine-grained local searches.

[0079] Step 8.6: Calculate the adaptive advance distance of the target task block. :

[0080] Among them, the basic advancement scale , and They represent the first Upper and lower bounds of dimensional variables; The degree of confidence in the direction of the target task time segment.

[0081] Step 8.7: Generate migration center points based on trend direction. :

[0082] In the formula, This is the normalized trend vector for the time slice of the target task.

[0083] Step 8.8: First, calculate the current field of view:

[0084] in The basic search metric is calculated using the same method and steps as in step 8.6. Same. This field of view is the coverage radius of subsequent Gaussian sampling, equivalent to the field of view angle in optical imaging. If step 8.2 confirms... If the threshold is lower than the preset trigger threshold, a field-of-view magnification operation is performed and the sampling radius is increased; otherwise, the original field of view is maintained. Then, the migration center point generated in step 8.7... Random samples are taken from nearby locations within the current field of view using a Gaussian distribution to generate candidate solutions.

[0085] Step 8.9: When determining the direction confidence level in step 8.2... If the field of view is below the preset trigger threshold and the field of view magnification operation has been performed in step 8.8, the zoom repositioning mechanism is simultaneously activated, based on the optimal individual... and migration center point Reconstructing the direction to obtain a new direction :

[0086] Step 8.10: The new direction is fused with the original predicted direction and normalized; the direction fusion coefficient is calculated. =0.3:

[0087] Finally, use the fused direction. Replace the direction in step 8.7 Then, repeat step 8.7 to generate a new migration center point, and jump to step 8.8 to sample within the current field of view, and output the sampled candidate solution as the final candidate solution.

Claims

1. A method for coordinated optimization of energy efficiency of base stations in a cellular network based on multi-task optimization, characterized in that, The process includes initializing parameters and defining each region as an independent task. After constructing a minimum per-bit energy consumption objective function with switching power carriers, data from each base station is collected and divided into blocks by time slices. Principal component analysis is used to extract block description vectors and block weights. Then, based on inter-block distance and dynamic thresholds, the overall task similarity is constructed, and similar task pairs and blocks are selected. Subsequently, a first-order or second-order difference model is selected to fit the normalized trend vector according to the local sensitivity of the block. Then, the direction confidence is updated based on the improvement rate. The solution is generated by combining the trend and zoom field of view sampling and transferred to the target block. Finally, adaptive differential evolution is performed on dissimilar blocks, and candidate strategies are evaluated, archived, and iterated until the maximum number of evaluations is reached.

2. The method for multi-task optimization based coordinated optimization of energy efficiency of base stations of cellular networks as claimed in claim 1, wherein, Includes the following steps: Step 1, initialize parameters, including population for each independent task , population size , maximum number of evaluations , number of blocks B, initialize sensitivity cache , list of historical center matrices , maximum history window ; Step 2: Define each geographical region in the cellular network as an independent optimization task. For each task, define a decision variable vector that includes the on / off / sleep flags, transmit power, and carrier allocation ratio of each base station in the corresponding region. Construct the objective function for each task as minimizing the ratio of total energy consumption to total capacity in the region. Step 3: Determine whether the preset termination condition is met. If the preset termination condition is met, output the optimal solution for each task; otherwise, proceed to step 4. Step 4: Collect data from each base station and divide it into blocks according to time slices, then extract the block description vector and block weight; Step 5: Based on the block description vectors and block weights obtained in Step 4, determine the Euclidean distance similarity between blocks and apply a dynamic threshold to construct the overall task similarity; Step 6: Select similar task pairs based on the overall task similarity matrix obtained in Step 5, and determine the set of time segments with high similarity between similar task pairs; Step 7: Perform trend fitting on the similar task pairs obtained in Step 6 according to time blocks to obtain the normalized trend; Step 8: Transfer valuable solutions from the source task time slice to the target task block and advance and sample to generate candidate solutions based on the normalization trend at the time slice level obtained in Step 7. Step 9: Inject the final candidate solutions obtained in Step 8 into the offspring pool of the target task according to time slices. For time slices of two tasks with low similarity, they undergo independent evolution. During the independent evolution process, based on the normalized local sensitivity obtained in Step 7, the mutation rate and crossover rate of the local differential evolution algorithm are adaptively adjusted. Several generations of local evolution are carried out and an elite selection strategy is adopted to retain the best individual. Finally, the base station energy efficiency strategy is provided to the controller of each region. Step 10, evaluate the candidate strategy generated in step 9 according to the evaluation strategy on the objective function constructed in step 2, and store the evaluation result in the local file, and execute steps 3-9 in a loop until the maximum evaluation times are reached , the optimization process ends.

3. The method for multi-task optimization based coordinated optimization of energy efficiency of cellular network base stations, according to claim 2, wherein, Step 2 specifically involves: Each geographical area in the cellular network is defined as an independent optimization task, and there are total tasks; for each task , , a decision variable vector is defined, which contains the on / off flag, the transmit power and the carrier allocation ratio of each base station in the corresponding area: wherein is the task the number of base stations in the area, is the base station the on / off flag of the base station, is the transmission power of the base station is the carrier allocation ratio of the base station is the carrier allocation ratio of the base station is the carrier allocation ratio of the base station is the transpose; Constructing each task of the objective function to be optimized to minimize the ratio of the total energy consumption to the total capacity of the area: wherein is the static power consumption of the base station , is the power amplifier efficiency, is the channel gain, is the interference power, is the noise power, is a normal number.

4. The method for coordinated energy efficiency optimization of cellular network base stations based on multi-task optimization as described in claim 2, characterized in that, Step 4 specifically includes the following steps: Step 4.1, for each task With each base station it contains Raw observation data was collected on a per-generational basis, including data from each base station. Static power consumption Transmission power Carrier allocation ratio Channel gain Interference power User throughput and total regional energy consumption; Step 4.2: Divide a 24-hour day into several time blocks. Each time slice represents a period of time within a day that has relatively stable traffic characteristics and optimization features, and the individual index corresponding to each time slice is stored in... middle; Step 4.3: Solving the solution set for each time slice Principal component analysis was performed, where Indicates the number of samples within the block. As a decision-making dimension, For the set of real numbers; before extraction Each principal component score matrix is ​​then calculated. Mean of dimensional data in each direction Standard deviation and skewness Combined into a length of 3 Description vector ; Step 4.4: Use the sum of the standard deviations of the principal components in the description vector of each time slice as the original block weights. Then, the original block weights of all time slices are normalized to obtain... .

5. The method for coordinated energy efficiency optimization of cellular network base stations based on multi-task optimization as described in claim 4, characterized in that, Step 5 specifically includes the following steps: Step 5.1: Based on the description vectors of each time slice obtained in Step 4.3 Calculate any two tasks and In the Euclidean distance on a time slice : in, and Tasks With the task In the Description vectors on each block Representing vectors Norm; Step 5.2, collect the first The set of Euclidean distances for all task pairs on a given time slice. Take set The mean scrambling action in the process is the similarity threshold for each time block. ; Step 5.3: Based on the normalized block weights obtained in step 4.4 Computational tasks With the task Overall similarity : In the formula, This is an indicator function.

6. The method for coordinated energy efficiency optimization of cellular network base stations based on multi-task optimization as described in claim 5, characterized in that, Step 7 specifically includes the following steps: Step 7.1: Based on the similar task pairs selected in Step 6 and their corresponding time slice sets, calculate the task... No. The time slice in the first The central position of the generation serves as the centroid of all individuals within the time segment. ; Step 7.2: Apply perturbation to each individual within the time slice to obtain... Then calculate the individual rate of change. And the local sensitivity at the time slice level is averaged to obtain ; Step 7.3: Normalize the local sensitivity at the time slice level to obtain... : In the formula, and These are the minimum and maximum sensitivities among all blocks in the current generation, respectively. It is a positive number; Step 7.4: Determine the value of the normalized local sensitivity. When smoothing the region, a first-order difference model is used to calculate the trend vector. : Conversely, the trend vector is calculated using a second-order difference model. : Finally, the calculated trend vector is normalized to obtain... This refers to the original predicted trend direction of the current block in the current generation.

7. The method for coordinated energy efficiency optimization of cellular network base stations based on multi-task optimization as described in claim 6, characterized in that, In step 7.4, when At that time, among them The trend vector is calculated using a first-order difference model to represent the sensitivity threshold; otherwise, a second-order difference model is used to calculate the trend vector.

8. The method for coordinated energy efficiency optimization of cellular network base stations based on multi-task optimization as described in claim 6, characterized in that, Step 8 specifically includes the following steps: Step 8.1, Task The Trust level for maintenance direction of time fragments Define the optimal objective value improvement amount for the g-th generation. Then improve the rate of change for: Step 8.2, based on the improved rate of change Update direction trust level, if If the optimization effect of the current trend direction is further enhanced, then the direction confidence level is updated as follows: In the formula, Used to limit trust levels Within the range, For trust learning rate, Limit the update range; Conversely, a fixed-amplitude decay is applied to the corresponding directional confidence level; when When the field of view is below a preset threshold, a magnification marker is triggered. Step 8.3, in the source task time slice Search among individuals within a predetermined proportion, and select the individual with the largest decrease in the target value over the last two generations as the representative solution. ; Step 8.4: Based on the ratio of the number of function evaluations used to the maximum number of evaluations, divide the algorithm process into three evolutionary stages: early, middle, and late. Step 8.5: Based on zoom imaging, use the search field of view (FOV) as a quantitative indicator of the imaging field of view, and adjust the stage adjustment factor accordingly. As the control parameter for focal length adjustment, the stage adjustment factor is determined based on the evolutionary stages divided in step 8.

4. ; Step 8.6: Based on stage adjustment factors And the direction trust level updated in step 8.2 Calculate the adaptive advance distance of the target mission time slice : In the formula, Based on the scale of advancement; directional confidence level for the target task time segment; Step 8.7: Based on the normalized trend vector obtained in Step 7 The representative solution of the source task time slice obtained in step 8.3 and the adaptive propulsion distance obtained in step 8.6 Generate migration center point : In the formula, This is the normalized trend vector for the time slice of the target task; Step 8.8: First, based on the stage adjustment factor obtained in Step 8.5... And the direction trust level updated in step 8.2 Calculate the current field of view. : In the formula, Use this as the basic search metric; if step 8.2 confirms... If the field of view is below the preset trigger threshold, a field of view magnification operation is performed and the sampling radius is increased; otherwise, the original field of view range is maintained. The migration center point generated in step 8.7 The surrounding area is distributed in a Gaussian pattern within the current field of view. Random sampling is performed within the sample to generate candidate solutions; Step 8.9: When determining the direction confidence level in step 8.2... If the field of view is below the preset trigger threshold and the field of view magnification operation has been performed in step 8.8, the zoom repositioning mechanism is simultaneously activated, based on the optimal individual... and migration center point Reconstructing the direction to obtain a new direction : Step 8.10: Apply the new direction obtained in Step 8.9 Normalized trend vector of the target task time slice Merge and normalize: In the formula, The directional fusion coefficient; Then use the fused direction. Replace the part in step 8.7 Then, repeat step 8.7 to generate a new migration center point, and jump to step 8.8 to sample within the current field of view, and output the sampled candidate solution as the final candidate solution.

Images