Electricity utilization acquisition efficiency improving method based on dual-mode and broadband carrier communication

Abstract

The invention relates to the technical field of power utilization acquisition efficiency improvement, in particular to a power utilization acquisition efficiency improvement method based on dual-mode and broadband carrier communication, which comprises the following steps of: S1, acquiring multi-dimensional characteristic parameters in power line carrier communication, and preprocessing to output a characteristic set; s2, adopting an LSTM + Transform hybrid model to predict a channel quality change trend in the future 5-15 minutes according to the feature set, and outputting a prediction result; when the method is used, channel quality changes are predicted through deep learning, prospective parameter adjustment is achieved, and communication risks caused by passive adaptation are avoided; based on multi-factor collaborative decision-making of edge intelligence, multiple influence factors are comprehensively considered, and optimal communication mode selection is facilitated; dependence on a master station is reduced through local decision making of the edge nodes, and the response speed and the real-time performance can be improved; the technical bottlenecks of untimely dual-mode switching and uneven load distribution in the prior art are solved, and the adaptability and transmission efficiency of dual-mode communication are improved.

Description

Technical Field

[0001] This invention relates to the field of power consumption data acquisition efficiency improvement technology, specifically to a method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication. Background Technology

[0002] Broadband carrier communication refers to a communication method that uses existing wired media such as power lines and coaxial cables as transmission channels, occupies a wide spectrum of 1MHz to 100MHz (or higher), and employs high-speed modulation and demodulation technologies such as Orthogonal Frequency Division Multiplexing (OFDM) and Orthogonal Frequency Division Multiple Access (OFDMA) to achieve data transmission at the Mbps to Gbps level. Dual-mode carrier communication refers to a hybrid communication technology that integrates two different communication modes (including combinations between carrier communication modes and combinations of carrier and wireless communication modes), and takes into account the communication needs of different scenarios through adaptive switching or cooperative working mechanisms. Common dual-mode combinations include: broadband carrier + narrowband carrier (such as HPLC + PRIME / PLC-NB); broadband carrier + wireless communication (such as BPLC + Wi-Fi / BLE / LoRa); and broadband carriers in different frequency bands (such as 2-10MHz low frequency band + 20-60MHz mid-high frequency band).

[0003] The patent with publication number CN118630923A states in its specification that "This invention proposes a method and system for improving the efficiency of power consumption data collection based on dual-mode and broadband carrier communication, relating to the field of power consumption data collection efficiency improvement technology. It involves constructing a heterogeneous network and a power consumption network, setting up power consumption data collection nodes, collecting power consumption information from these nodes through the heterogeneous network, obtaining a centralized channel and branch channels, calculating the power consumption data collection efficiency coefficient of each branch channel based on the collected information, determining the collection status of the branch channels, and identifying whether a branch channel adjustment command is triggered; determining the relay agent corresponding to the optimal branch channel, collecting power consumption data based on the relay agent, and transmitting data through the optimal branch channel, calculating the power consumption data collection efficiency coefficient of the optimal branch channel, and determining whether to perform a re-acquisition. This invention has the characteristics of high efficiency, stability, and intelligence, and can optimize the efficiency of power information collection and transmission."

[0004] While existing technologies offer the advantages mentioned above, they also have disadvantages: High-speed power line carrier communication (HPLC) is the mainstream technology for local communication in smart grids and has been widely used in electricity consumption information collection. In existing technologies, HPLC dual-mode communication technology combines high-speed power line carrier communication and high-speed wireless communication (HRF) to solve the "island" problem in a single communication mode through dual communication methods. However, existing HPLC communication systems mainly rely on passive parameter adjustments based on the current channel state, lacking the ability to predict changes in channel quality and thus failing to mitigate communication risks in advance. Furthermore, dual-mode communication switching is primarily based on simple threshold judgments or current channel quality, lacking multi-factor comprehensive decision-making and making it difficult to achieve optimal communication mode selection.

[0005] In conclusion, developing a method for improving power consumption data collection efficiency based on dual-mode and broadband carrier communication remains a key issue that urgently needs to be addressed in the field of power consumption data collection efficiency improvement technology. Summary of the Invention

[0006] The purpose of this invention is to address the problems in the existing technology, namely that the existing HPLC communication system mainly adopts passive parameter adjustment based on the current channel state, lacks the ability to predict changes in channel quality, and cannot avoid communication risks in advance. In addition, the dual-mode communication switching is mainly based on simple threshold judgment or current channel quality, lacks multi-factor comprehensive decision-making, and is difficult to achieve optimal communication mode selection.

[0007] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for improving the efficiency of power consumption data acquisition based on dual-mode and broadband carrier communication, comprising the following steps: S1. Collect multi-dimensional feature parameters in power line carrier communication, perform preprocessing, and output feature sets; S2. Using the LSTM+Transformer hybrid model, based on the feature set, predict the channel quality change trend in the next 5-15 minutes and output the prediction results; S3. Based on the prediction results, construct a parameter optimization objective function, use reinforcement learning algorithm to adjust multiple target parameters of HPLC communication, and establish a dynamic parameter configuration library. S4. Deploy a lightweight AI inference model on edge nodes, construct a multi-objective optimization decision model, solve the problem, and output the decision results. S5. An intelligent dual-mode switching strategy is adopted to select the optimal communication mode based on the decision results, while performing distributed data processing at the edge nodes.

[0008] Furthermore, in step S1, the method for collecting multi-dimensional feature parameters in power line carrier communication and preprocessing them to output a feature set is as follows: Multi-dimensional feature sets are constructed by real-time acquisition of multi-dimensional feature parameters in power line carrier communication scenarios using sensing units. ,in, for Timing signal-to-noise ratio, for False alarm rate at any time for Signal attenuation over time for Time-limited noise power spectral density for Time-based channel capacity for Real-time network load rate for Real-time device power consumption, sampling frequency is Continuous collection Minutes, forming the initial feature time series dataset. ,in, Represents the set of real numbers. It is the row number of the matrix. It is the number of columns in the matrix; The preprocessing uses the Grubbs criterion to detect and remove outliers. Calculate the mean and standard deviation, in At that time, among them If the value is the Grubbs threshold, it is considered an outlier and replaced with... Using Min-Max normalization Mapped to The interval is normalized. Subsequently, kernel principal component analysis was used to reduce feature dimensionality. First, a kernel matrix was constructed using kernel functions. Then, the kernel matrix was centered. Finally, eigenvalue decomposition was performed on the processed kernel matrix, and the top eigenvalues ​​were selected. Construct the dimensionality-reduced feature set from the eigenvectors corresponding to the largest eigenvalues. .

[0009] Furthermore, in step S2, an LSTM+Transformer hybrid model is used to predict the channel quality change trend over the next 5-15 minutes based on the feature set. The method for outputting the prediction result is as follows: The dimensionality-reduced feature set is divided into time windows, and the window length is set. The input sequence is constructed, and the LSTM layer captures the long-term time dependency of channel quality through a gating mechanism, expressed as: , In the formula, Indicates the current The output of the Forget Gate This represents the Sigmoid activation function, which maps the calculation result to the interval between 0 and 1. The weight matrix represents the forget gate. This indicates the hidden state at the previous moment. Indicates the current Input at any time The bias term representing the forget gate. Indicates the current The output of the input gate is always in use. This represents the weight matrix of the input gate. This represents the bias term of the input gate. Indicates the current Information that is constantly being generated and is yet to be integrated into the cell state. This represents the hyperbolic tangent activation function. The weight matrix representing the candidate cell state. Bias terms representing the candidate cell state. Indicates the current Cellular state at any given moment It represents the Hadamah accumulation. This indicates the cell state at the previous moment. Indicates the current The output of the output gate is always available. This represents the weight matrix of the output gate. This represents the bias term of the output gate. Indicates the current The hidden state at all times This is the weight matrix. This is the weight matrix. Indicates the hidden layer dimension. Indicates the input feature dimension. This indicates the shape of the weight matrix. This indicates that the bias vector is of dimension . The real-valued vector, after being processed by the LSTM layer, outputs a time feature matrix. , This indicates that the characteristic matrix is One sample × Each time step × The hidden state is a three-dimensional real matrix; the Transformer layer uses linear projection, multi-head attention computation, and a feedforward neural network to... After processing, the fused feature matrix is ​​output. , This represents the fused feature matrix obtained after processing by the Transformer layer. The shape of the fused feature matrix is Sample size × A two-dimensional real matrix with feature dimension.

[0010] Furthermore, the method in step S2 also includes: Will Input to a fully connected layer for prediction, output the future. Predicted channel quality parameters for minutes ,in, This represents the predicted value of the channel quality parameter. This represents the predicted signal-to-noise ratio. This represents the false positive rate of the prediction. Indicates the predicted signal attenuation. This represents the predicted channel capacity. The transpose notation, and the formula for the fully connected layer are: , In the formula, This represents the hyperbolic tangent activation function. This represents the weight matrix of the first layer of the fully connected layer. This represents the bias vector of the first layer of the fully connected layer. This represents the weight matrix of the second layer of the fully connected layer. This represents the bias vector of the second layer of the fully connected layer.

[0011] Furthermore, in step S3, a parameter optimization objective function is constructed based on the prediction results, and a reinforcement learning algorithm is used to adjust multiple target parameters of HPLC communication. Simultaneously, a dynamic parameter configuration library is established using the following method: The objective function for optimizing the construction parameters aims to maximize communication success rate, minimize energy consumption, minimize transmission delay, and maximize coding efficiency. Its expression is: , In the formula, It is the set of parameters to be optimized. Indicates parameters The corresponding objective function value, To improve communication success rate, For transmission energy consumption, For total delay, For coding efficiency, For the target weight, satisfying The objective function for parameter optimization is then modeled as a Markov decision process, and the state space is defined as follows: , In the formula, This represents the state space for reinforcement learning. This represents the predicted signal-to-noise ratio. This represents the predicted packet error rate. For network load rate, This refers to the remaining energy consumption of the equipment. Define the action space as follows: , In the formula, Represents the action space of reinforcement learning. Indicates the operating frequency band. It's a modulation method. It's the transmission power. It is the transmission rate; the reward function is defined as: , In the formula, This represents the reward function for reinforcement learning. This is the current state. It is the action to be performed. Adjust the penalty coefficient for the parameters. These are the parameters from the previous time step.

[0012] Furthermore, the method in step S3 also includes: The reinforcement learning algorithm employs a proximal policy optimization algorithm to update the policy. The objective function is: , In the formula, Let the objective function of the near-end policy optimization algorithm be denoted as . It is the cutting factor. Indicates the current strategy Below, in state Execute action The probability, Indicates the old strategy before the update. Below, in state Execute action The probability, For the dominant function, Let the action value function be... The state value function; the dynamic parameter configuration library optimizes the objective function of the optimized parameters. Store in dynamic parameter configuration library , This represents the optimal combination of parameters after optimization. This represents the independent variable that yields the maximum value, while a sliding window mechanism is used to remove outdated parameters. The window size is [value missing]. The group's update formula is: ,in, This indicates the updated dynamic parameter configuration library. The difference operation represents the difference between sets. This represents the oldest set of parameters in the configuration library. This represents the union operation of sets. This represents the optimal parameter combination obtained from the new optimization.

[0013] Furthermore, in step S4, a lightweight AI inference model is deployed at the edge nodes to construct a multi-objective optimization decision model. The method for solving the model and outputting the decision results is as follows: Lightweight AI inference models are deployed at concentrators and key relay nodes. These lightweight AI inference models undergo pruning and quantization, and their inference latency is reduced after deployment. ms; Predicted value based on channel quality parameters Preset data transmission priority Network load Equipment energy consumption status , Where 1 represents the lowest and 5 represents the highest, a multi-objective decision-making model is constructed: , in, This means maximizing, that is, maximizing the following... Reach the maximum value, This represents the weight of the corresponding channel quality prediction result. This is the channel quality prediction result. This indicates the weight corresponding to the data transmission priority. Indicates the importance of the data to be transmitted. This indicates the weight corresponding to the network load. This indicates the current network traffic level. This indicates the weight of the corresponding device's energy consumption status. This indicates the percentage of remaining battery power in the device. It is the operating frequency band. It's the transmission power. It is the encoding rate. It's the modulation method, from Choose from these three options. This is the decision evaluation value.

[0014] Furthermore, the method in step S4 also includes: A non-dominated sorting genetic algorithm is used to solve the multi-objective decision model, obtaining the Pareto optimal decision set and generating... Individual decision-makers Indicate the population size, perform population initialization, based on the objective function. Calculate fitness values ​​while considering whether constraints are met. Individuals violating constraints have a fitness penalty of 0. Divide the population into different non-dominated levels. Individuals in the same level are ranked by crowding. Expression: , in, For the first The degree of crowding of individuals Indicates the first The item in the first The feature value at time 1, Indicates the first The item in the first The feature value at time 1, Indicates the first The maximum value of each component. Indicates the first The minimum value of each component. To determine the target number, a roulette wheel selection method is used, with a crossover probability set to [value missing]. Simulated binary crossover is used, and the mutation probability is set. Polynomial mutation is employed; the previous version is retained. The best individual enters the next generation of the population, with the number of iterations being... Choose from the final Pareto optimal set The largest individual outputs the decision result.

[0015] Furthermore, in step S5, an intelligent dual-mode switching strategy is adopted to select the optimal communication mode based on the decision results, while simultaneously performing distributed data processing at the edge nodes. By combining real-time channel feedback with decision results, a handover determination function is constructed, expressed as: , In the formula, To switch the value of the decision function, The decision evaluation value for HPLC communication mode. The decision evaluation value for HRF communication mode. .

[0016] Furthermore, the method in step S5 also includes: exist Select HPLC mode and execute the optimized parameter configuration in step S3. When selecting HRF mode, When dual-mode parallel transmission is enabled, the data splitting ratio is: , In the formula, Indicates the data split ratio, HPLC transfer Proportional data, HRF transmission The load balancing scheduling of proportional data and dual-mode parallel transmission adopts a load balancing algorithm based on queuing theory to schedule dual-mode transmission traffic; the distributed data processing at the edge nodes includes data filtering and data compression.

[0017] Beneficial effects Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects: In use, this invention uses deep learning to predict channel quality changes, enabling proactive parameter adjustments and avoiding communication risks associated with passive adaptation. Based on edge intelligence and multi-factor collaborative decision-making, it comprehensively considers various influencing factors, facilitating optimal communication mode selection. Local decision-making at edge nodes reduces dependence on the master station, improving response speed and real-time performance. It also addresses the technical bottlenecks of untimely dual-mode switching and uneven load distribution, enhancing the adaptability and transmission efficiency of dual-mode communication. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] The present invention will now be described in further detail with reference to the accompanying drawings: Example: like Figure 1 As shown, this invention provides a method for improving the efficiency of power consumption data collection based on dual-mode and broadband carrier communication, comprising the following steps: S1. Collect multi-dimensional feature parameters in power line carrier communication, perform preprocessing, and output feature sets; S2. Using the LSTM+Transformer hybrid model, based on the feature set, predict the channel quality change trend in the next 5-15 minutes and output the prediction results; S3. Based on the prediction results, construct a parameter optimization objective function, use reinforcement learning algorithm to adjust multiple target parameters of HPLC communication, and establish a dynamic parameter configuration library. S4. Deploy a lightweight AI inference model on edge nodes, construct a multi-objective optimization decision model, solve the problem, and output the decision results. S5. Adopt an intelligent dual-mode switching strategy, select the optimal communication mode based on the decision results, and perform distributed data processing at the edge nodes. Furthermore, in step S1, the method for collecting multi-dimensional feature parameters in power line carrier communication and preprocessing them to output a feature set is as follows: Multi-dimensional feature sets are constructed by real-time acquisition of multi-dimensional feature parameters in power line carrier communication scenarios using sensing units. ,in, for Timing signal-to-noise ratio, for False alarm rate at any time for Signal attenuation over time for Time-limited noise power spectral density for Time-based channel capacity for Real-time network load rate for Real-time device power consumption, sampling frequency is Continuous collection Minutes, forming the initial feature time series dataset. ,in, Represents the set of real numbers. It is the row number of the matrix. It is the number of columns in the matrix; The preprocessing uses the Grubbs criterion to detect and remove outliers. Calculate the mean and standard deviation, in At that time, among them If the value is the Grubbs threshold, it is considered an outlier and replaced with... Using Min-Max normalization Mapped to The interval is normalized. Subsequently, kernel principal component analysis was used to reduce feature dimensionality. First, a kernel matrix was constructed using kernel functions. Then, the kernel matrix was centered. Finally, eigenvalue decomposition was performed on the processed kernel matrix, and the top eigenvalues ​​were selected. Construct the dimensionality-reduced feature set from the eigenvectors corresponding to the largest eigenvalues. ; In this embodiment, taking a smart electricity consumption data collection scenario in a residential community of a city as an example, the smart meters and concentrators deployed in the community collect 7-dimensional feature parameters in real time under power line carrier communication through sensing units. Data is continuously collected for 5 minutes at a sampling frequency of 10Hz, forming an initial feature time-series dataset of 7 rows and 3000 columns. First, the Grubbs criterion is used to eliminate abnormal false alarm rate data caused by voltage fluctuations during peak electricity consumption in the community. Then, Min-Max normalization is used to map all parameters to the [0,1] interval to eliminate dimensional differences. Finally, kernel principal component analysis is used... The method reduces 7-dimensional features to 3-dimensional features to simplify computation, resulting in a dimensionality-reduced feature set. Unlike existing passive adjustment schemes that rely solely on real-time channel conditions, this method combines Grubbs outlier processing with kernel principal component dimensionality reduction for feature preprocessing, spatiotemporal sequence prediction, and edge intelligent decision-making to achieve forward-looking and lightweight channel optimization. By integrating multi-dimensional feature dimensionality reduction, deep learning prediction, reinforcement learning optimization, and edge computing technologies, this method addresses the pain points of large channel fluctuations and redundant data collection during peak electricity consumption in residential areas, overcoming the limitations of traditional methods that struggle to balance acquisition efficiency and equipment energy consumption.

[0022] Furthermore, in step S2, an LSTM+Transformer hybrid model is used to predict the channel quality change trend over the next 5-15 minutes based on the feature set. The method for outputting the prediction result is as follows: The dimensionality-reduced feature set is divided into time windows, and the window length is set. The input sequence is constructed, and the LSTM layer captures the long-term time dependency of channel quality through a gating mechanism, expressed as: , In the formula, Indicates the current The output of the Forget Gate This represents the Sigmoid activation function, which maps the calculation result to the interval between 0 and 1. The weight matrix represents the forget gate. This indicates the hidden state at the previous moment. Indicates the current Input at any time The bias term representing the forget gate. Indicates the current The output of the input gate is always in use. This represents the weight matrix of the input gate. This represents the bias term of the input gate. Indicates the current Information that is constantly being generated and is yet to be integrated into the cell state. This represents the hyperbolic tangent activation function. The weight matrix representing the candidate cell state. Bias terms representing the candidate cell state. Indicates the current Cellular state at any given moment It represents the Hadamah accumulation. This indicates the cell state at the previous moment. Indicates the current The output of the output gate is always available. This represents the weight matrix of the output gate. This represents the bias term of the output gate. Indicates the current The hidden state at all times This is the weight matrix. This is the weight matrix. Indicates the hidden layer dimension. Indicates the input feature dimension. This indicates the shape of the weight matrix. This indicates that the bias vector is of dimension . The real-valued vector, after being processed by the LSTM layer, outputs a time feature matrix. , This indicates that the characteristic matrix is One sample × Each time step × The hidden state is a three-dimensional real matrix; the Transformer layer uses linear projection, multi-head attention computation, and a feedforward neural network to... After processing, the fused feature matrix is ​​output. , This represents the fused feature matrix obtained after processing by the Transformer layer. The shape of the fused feature matrix is Sample size × A two-dimensional real matrix with feature dimension; Furthermore, the method in step S2 also includes: Will Input to a fully connected layer for prediction, output the future. Predicted channel quality parameters for minutes ,in, This represents the predicted value of the channel quality parameter. This represents the predicted signal-to-noise ratio. This represents the false positive rate of the prediction. Indicates the predicted signal attenuation. This represents the predicted channel capacity. The transpose notation, and the formula for the fully connected layer are: , In the formula, This represents the hyperbolic tangent activation function. This represents the weight matrix of the first layer of the fully connected layer. This represents the bias vector of the first layer of the fully connected layer. This represents the weight matrix of the second layer of the fully connected layer. This represents the bias vector of the second layer of the fully connected layer; In this embodiment, taking the intelligent power consumption data collection scenario of an electronic industrial park as an example, to address the problem of frequent channel fluctuations caused by the start-up and shutdown of equipment such as machine tools and production lines within the park, the 3D feature set after S1 dimensionality reduction is divided into 60 time steps, corresponding to 6 seconds, with a collection frequency of 10Hz. The input sequence is divided, and firstly, the long-term time dependence of channel quality is captured through gating mechanisms such as forget gates and input gates in the LSTM layer, outputting a three-dimensional time feature matrix of sample number × time step × hidden layer dimension. Then, through linear projection, multi-head attention calculation, and feedforward neural network in the Transformer layer, the time feature matrix is ​​fused into a two-dimensional fused feature matrix of sample number × feature dimension. Finally, this matrix is ​​input into two fully connected layers to output the channel quality prediction value for the next 15 minutes, including signal-to-noise ratio, packet error rate, signal attenuation, and channel capacity. This application achieves spatiotemporal dual-dimensional prediction of channel quality by combining LSTM gated time feature capture with Transformer multi-head attention spatial fusion. Spatiotemporal feature fusion overcomes the limitation of low prediction accuracy of traditional models, which is beneficial to reducing prediction errors.

[0023] Furthermore, in step S3, a parameter optimization objective function is constructed based on the prediction results, and a reinforcement learning algorithm is used to adjust multiple target parameters of HPLC communication. Simultaneously, a dynamic parameter configuration library is established using the following method: The objective function for optimizing the construction parameters aims to maximize communication success rate, minimize energy consumption, minimize transmission delay, and maximize coding efficiency. Its expression is: , In the formula, It is the set of parameters to be optimized. Indicates parameters The corresponding objective function value, To improve communication success rate, For transmission energy consumption, For total delay, For coding efficiency, For the target weight, satisfying The objective function for parameter optimization is then modeled as a Markov decision process, and the state space is defined as follows: , In the formula, This represents the state space for reinforcement learning. This represents the predicted signal-to-noise ratio. This represents the predicted packet error rate. For network load rate, This refers to the remaining energy consumption of the equipment. Define the action space as follows: , In the formula, Represents the action space of reinforcement learning. Indicates the operating frequency band. It's a modulation method. It's the transmission power. It is the transmission rate; the reward function is defined as: , In the formula, This represents the reward function for reinforcement learning. This is the current state. It is the action to be performed. Adjust the penalty coefficient for the parameters. These are the parameters from the previous time step; Furthermore, the method in step S3 also includes: The reinforcement learning algorithm employs a proximal policy optimization algorithm to update the policy. The objective function is: , In the formula, Let the objective function of the near-end policy optimization algorithm be denoted as . It is the cutting factor. Indicates the current strategy Below, in state Execute action The probability, Indicates the old strategy before the update. Below, in state Execute action The probability, For the dominant function, Let the action value function be... The state value function; the dynamic parameter configuration library optimizes the objective function of the optimized parameters. Store in dynamic parameter configuration library , This represents the optimal combination of parameters after optimization. This represents the independent variable that yields the maximum value, while a sliding window mechanism is used to remove outdated parameters. The window size is [value missing]. The group's update formula is: ,in, This indicates the updated dynamic parameter configuration library. The difference operation represents the difference between sets. This represents the oldest set of parameters in the configuration library. This represents the union operation of sets. This represents the optimal parameter combination obtained through the new optimization. In this embodiment, taking the intelligent power consumption data collection scenario of a large commercial complex as an example, to address the channel congestion problem caused by the simultaneous operation of air conditioning, lighting, and shop equipment during peak hours of the complex, such as weekend peak customer traffic, a weighted function is first constructed with the objectives of maximizing communication success rate, minimizing energy consumption, minimizing transmission delay, and maximizing coding efficiency. The objective weights are set as follows: The optimization problem is then modeled as a Markov decision process, using parameters such as the operating frequency band and modulation scheme of HPLC communication as the optimization object. The state space consists of the predicted signal-to-noise ratio, packet error rate, network load, and remaining equipment power consumption. The action space consists of the 0.7-12MHz frequency band, QPSK / 16QAM / 64QAM modulation scheme, 10-30dBm transmit power, and 1-12Mbps transmission rate. A penalty coefficient for parameter adjustment is also introduced. A reward function is constructed to avoid communication fluctuations caused by parameter mutations; subsequently, a near-end policy optimization algorithm is adopted, which adjusts the pruning coefficient. The policy update range is limited to stabilize the optimization process. Finally, a dynamic parameter configuration library with a sliding window size of 50 groups is established. After obtaining the optimal parameters each time, the oldest parameters in the library are removed and new parameters are stored to achieve dynamic updates of the configuration. This application differs from existing single-objective parameter optimization or schemes without parameter adjustment constraints. By combining a multi-objective weighted objective function, a reward function with penalties, a pruning strategy of the near-end policy optimization algorithm, and sliding window configuration, it achieves multi-objective balance and dynamic adaptation of parameter optimization. Addressing the pain points of complex channels and large load fluctuations in commercial complexes, it breaks through the limitations of traditional parameter optimization that suffers from one aspect at the expense of another, which is conducive to improving the adaptation efficiency of parameter adjustment.

[0024] Furthermore, in step S4, a lightweight AI inference model is deployed at the edge nodes to construct a multi-objective optimization decision model. The method for solving the model and outputting the decision results is as follows: Lightweight AI inference models are deployed at concentrators and key relay nodes. These lightweight AI inference models undergo pruning and quantization, and their inference latency is reduced after deployment. ms; Predicted value based on channel quality parameters Preset data transmission priority Network load Equipment energy consumption status , Where 1 represents the lowest and 5 represents the highest, a multi-objective decision-making model is constructed: , in, This means maximizing, that is, maximizing the following... Reach the maximum value, This represents the weight of the corresponding channel quality prediction result. This is the channel quality prediction result. This indicates the weight corresponding to the data transmission priority. Indicates the importance of the data to be transmitted. This indicates the weight corresponding to the network load. This indicates the current network traffic level. This indicates the weight of the corresponding device's energy consumption status. This indicates the percentage of remaining battery power in the device. It is the operating frequency band. It's the transmission power. It is the encoding rate. It's the modulation method, from Choose from these three. This is the decision evaluation value; Furthermore, the method in step S4 also includes: A non-dominated sorting genetic algorithm is used to solve the multi-objective decision model, obtaining the Pareto optimal decision set and generating... Individual decision-makers Indicate the population size, perform population initialization, based on the objective function. Calculate fitness values ​​while considering whether constraints are met. Individuals violating constraints have a fitness penalty of 0. Divide the population into different non-dominated levels. Individuals in the same level are ranked by crowding. Expression: , in, For the first The degree of crowding of individuals Indicates the first The item in the first The feature value at time 1, Indicates the first The item in the first The feature value at time 1, Indicates the first The maximum value of each component. Indicates the first The minimum value of each component. To determine the target number, a roulette wheel selection method is used, with a crossover probability set to [value missing]. Simulated binary crossover is used, and the mutation probability is set. Polynomial mutation is employed; the previous version is retained. The best individual enters the next generation of the population, with the number of iterations being... Choose from the final Pareto optimal set The largest individual outputs the decision result; In this embodiment, taking a smart community's power consumption data collection scenario encompassing residential building meters, elevator safety monitoring, and public lighting as an example, a pruned and quantized lightweight AI inference model is deployed at the community's concentrator and key relay nodes. After deployment, the inference latency is controlled within 40ms, meeting the real-time decision-making requirements of the edge side. Combining the previously output channel quality prediction results, preset data transmission priorities (e.g., elevator safety power consumption data priority set to 5, general lighting data priority set to 2), real-time network load, and device remaining energy consumption status, a multi-objective decision-making model is constructed with the goal of maximizing the decision evaluation value B. The objective weights are configured as follows: Simultaneously, the operating frequency band of HPLC communication was limited to 0.7-12MHz, and the transmission power to 10-27dBm, among other parameter constraints. Subsequently, a non-dominated sorting genetic algorithm was used to solve the multi-objective decision-making model: first, 200 decision-makers were initialized, each corresponding to a set of communication parameters and mode schemes. Fitness values ​​were calculated based on the objective function B, with individuals violating parameter constraints receiving a fitness penalty of 0. The population was then divided into different non-dominated levels, and finally, the crowding formula was used... Sort individuals at the same level, combine roulette wheel selection and simulated binary crossover, and calculate the crossover probability. =0.8, polynomial mutation, mutation probability =0.05, perform genetic operations, retain the top 20% of the best individuals for 50 iterations, and finally select the individual with the largest decision evaluation value B from the Pareto optimal set, outputting the decision result of the optimal communication parameters and mode; This application differs from existing centralized decision-making or single-objective decision-making schemes that rely on the master station. By combining the edge lightweight inference model with the multi-objective non-dominated sorting genetic algorithm, it realizes low-latency multi-dimensional intelligent decision-making on the edge side. It addresses the pain points of data priority differences among multiple devices and channel load fluctuations in smart communities, and breaks through the limitations of slow traditional decision response and single objective, which is conducive to improving the decision response speed.

[0025] Furthermore, in step S5, an intelligent dual-mode switching strategy is adopted to select the optimal communication mode based on the decision results, while simultaneously performing distributed data processing at the edge nodes. By combining real-time channel feedback with decision results, a handover determination function is constructed, expressed as: , In the formula, To switch the value of the decision function, The decision evaluation value for HPLC communication mode. The decision evaluation value for HRF communication mode. ; Furthermore, the method in step S5 also includes: exist Select HPLC mode and execute the optimized parameter configuration in step S3. When selecting HRF mode, When dual-mode parallel transmission is enabled, the data splitting ratio is: , In the formula, Indicates the data split ratio, HPLC transfer Proportional data, HRF transmission The load balancing scheduling of proportional data and dual-mode parallel transmission adopts a queuing theory-based load balancing algorithm to schedule dual-mode transmission traffic; the distributed data processing at the edge nodes includes data filtering and data compression. In this embodiment, taking the smart electricity consumption data collection scenario in a suburban area as an example, this scenario has problems such as HPLC channel fluctuations caused by aging rural power grid lines and uneven load caused by mixed connection of residential and agricultural electrical equipment. After the concentrator obtains real-time channel feedback data and the decision result of step S4, this application switches the judgment function. Calculate the percentage difference in decision evaluation values ​​between HPLC and HRF modes; under stable residential electricity consumption and excellent HPLC channel quality on weekdays, Select HPLC mode and execute the optimized parameter configuration in step S3; when agricultural machinery starts and stops during busy farming seasons, causing severe interference in the HPLC channel, Switching to HRF mode ensures transmission; during peak electricity consumption periods in the morning and evening, when the performance of the two modes is similar, Enable dual-mode parallel transmission, according to the splitting ratio Data traffic is allocated, with HPLC handling 60% and HRF handling 40%, and dual-mode traffic is scheduled using a queuing theory-based load balancing algorithm to avoid congestion on a single link. Simultaneously, distributed data processing is performed at edge nodes, filtering out invalid interference data generated by the start-up and shutdown of agricultural machinery, and then compressing the valid collected data to reduce transmission redundancy. This application differs from existing dual-mode handover schemes based on fixed thresholds by constructing a handover determination function based on the proportion of decision evaluation values. Combined with dynamic traffic splitting ratios and queuing theory load balancing, dual-mode collaboration is achieved, overcoming the limitations and passivity of traditional handover strategies. Addressing the pain points of complex channels and large load fluctuations in urban-rural fringe areas, this solution overcomes the technical bottlenecks of untimely dual-mode handover and uneven load distribution, thus improving the adaptability and transmission efficiency of dual-mode communication.

[0026] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for improving the efficiency of power consumption data acquisition based on dual-mode and broadband carrier communication, characterized in that, Includes the following steps: S1. Collect multi-dimensional feature parameters in power line carrier communication, perform preprocessing, and output feature sets; S2. Using the LSTM+Transformer hybrid model, based on the feature set, predict the channel quality change trend in the next 5-15 minutes and output the prediction results; S3. Based on the prediction results, construct a parameter optimization objective function, use reinforcement learning algorithm to adjust multiple target parameters of HPLC communication, and establish a dynamic parameter configuration library. S4. Deploy a lightweight AI inference model on edge nodes, construct a multi-objective optimization decision model, solve the problem, and output the decision results. S5. An intelligent dual-mode switching strategy is adopted to select the optimal communication mode based on the decision results, while performing distributed data processing at the edge nodes.

2. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 1, characterized in that, In step S1, the method for collecting multi-dimensional feature parameters in power line carrier communication and preprocessing them to output a feature set is as follows: Multi-dimensional feature parameters in power line carrier communication scenarios are collected in real time by sensing units to construct a multi-dimensional feature set. ,in, for Timing signal-to-noise ratio, for False alarm rate at any time for Signal attenuation over time for Time-limited noise power spectral density for Time-based channel capacity for Real-time network load rate for Real-time device power consumption, sampling frequency is Continuous collection Minutes, forming the initial feature time series dataset. ,in, Represents the set of real numbers. It is the row number of the matrix. It is the number of columns in the matrix; The preprocessing uses the Grubbs criterion to detect and remove outliers. Calculate the mean and standard deviation, in At that time, among them If the value is the Grubbs threshold, it is considered an outlier and replaced with... Using Min-Max normalization Mapped to The interval is normalized. Subsequently, kernel principal component analysis was used to reduce feature dimensionality. First, a kernel matrix was constructed using kernel functions. Then, the kernel matrix was centered. Finally, eigenvalue decomposition was performed on the processed kernel matrix, and the top eigenvalues ​​were selected. The eigenvectors corresponding to the largest eigenvalues ​​are used to construct the dimensionality-reduced feature set. .

3. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 2, characterized in that, In step S2, an LSTM+Transformer hybrid model is used to predict the channel quality change trend over the next 5-15 minutes based on the feature set. The method for outputting the prediction results is as follows: The dimensionality-reduced feature set is divided into time windows, and the window length is set. The input sequence is constructed, and the LSTM layer captures the long-term time dependency of channel quality through a gating mechanism, expressed as: ， In the formula, Indicates the current The output of the Forget Gate This represents the Sigmoid activation function, which maps the calculation result to the interval between 0 and 1. The weight matrix represents the forget gate. This indicates the hidden state at the previous moment. Indicates the current Input at any time The bias term representing the forget gate. Indicates the current The output of the input gate is always in use. This represents the weight matrix of the input gate. This represents the bias term of the input gate. Indicates the current Information that is constantly being generated and is yet to be integrated into the cell state. This represents the hyperbolic tangent activation function. The weight matrix representing the candidate cell state. Bias terms representing the candidate cell state. Indicates the current Cellular state at any given moment It represents the Hadamah accumulation. This indicates the cell state at the previous moment. Indicates the current The output of the output gate is always available. This represents the weight matrix of the output gate. This represents the bias term of the output gate. Indicates the current The hidden state at all times This is the weight matrix. This is the weight matrix. Indicates the hidden layer dimension. Indicates the input feature dimension. This indicates the shape of the weight matrix. This indicates that the bias vector is of dimension . The real-valued vector, after being processed by the LSTM layer, outputs a time feature matrix. , This indicates that the characteristic matrix is One sample × Each time step × The hidden state is a three-dimensional real matrix; the Transformer layer uses linear projection, multi-head attention computation, and a feedforward neural network to... After processing, the fused feature matrix is ​​output. , This represents the fused feature matrix obtained after processing by the Transformer layer. The shape of the fused feature matrix is Sample size × A two-dimensional real matrix with feature dimension.

4. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 3, characterized in that, The method in step S2 also includes: Will Input to a fully connected layer for prediction, output the future. Predicted channel quality parameters for minutes ,in, This represents the predicted value of the channel quality parameters. This represents the predicted signal-to-noise ratio. This represents the false positive rate of the prediction. Indicates the predicted signal attenuation. This represents the predicted channel capacity. The transpose notation, and the formula for the fully connected layer are: ， In the formula, This represents the hyperbolic tangent activation function. This represents the weight matrix of the first layer of the fully connected layer. This represents the bias vector of the first layer of the fully connected layer. This represents the weight matrix of the second layer of the fully connected layer. This represents the bias vector of the second layer of the fully connected layer.

5. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 4, characterized in that, In step S3, a parameter optimization objective function is constructed based on the prediction results. A reinforcement learning algorithm is used to adjust multiple target parameters of the HPLC communication. Simultaneously, a dynamic parameter configuration library is established using the following method: The objective function for optimizing the construction parameters aims to maximize communication success rate, minimize energy consumption, minimize transmission delay, and maximize coding efficiency. Its expression is: ， In the formula, It is the set of parameters to be optimized. Indicates parameters The corresponding objective function value, To improve communication success rate, For transmission energy consumption, For total delay, For coding efficiency, For the target weight, satisfying The objective function for parameter optimization is then modeled as a Markov decision process, and the state space is defined as follows: ， In the formula, This represents the state space for reinforcement learning. This represents the predicted signal-to-noise ratio. This represents the predicted packet error rate. For network load rate, This refers to the remaining energy consumption of the equipment. Define the action space as follows: ， In the formula, Represents the action space of reinforcement learning. Indicates the operating frequency band. It's a modulation method. It's the transmission power. It is the transmission rate; the reward function is defined as: ， In the formula, This represents the reward function for reinforcement learning. This is the current state. It is the action to be performed. Adjust the penalty coefficient for the parameters. These are the parameters from the previous time step.

6. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 5, characterized in that, The method in step S3 also includes: The reinforcement learning algorithm employs a proximal policy optimization algorithm to update the policy. The objective function is: ， In the formula, Let the objective function of the near-end policy optimization algorithm be denoted as . It is the cutting factor. Indicates the current strategy Below, in state Execute action The probability, Indicates the old strategy before the update. Below, in state Execute action The probability, For the dominant function, Let the action value function be... The state value function; the dynamic parameter configuration library optimizes the objective function of the optimized parameters. Store in dynamic parameter configuration library , This represents the optimal combination of parameters after optimization. This represents the independent variable that yields the maximum value, while a sliding window mechanism is used to remove outdated parameters. The window size is [value missing]. The group's update formula is: ,in, This indicates the updated dynamic parameter configuration library. The difference operation represents the difference between sets. This represents the oldest set of parameters in the configuration library. This represents the union operation of sets. This represents the optimal parameter combination obtained from the new optimization.

7. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 6, characterized in that, In step S4, a lightweight AI inference model is deployed at the edge nodes to construct a multi-objective optimization decision model. The method for solving the model and outputting the decision results is as follows: Lightweight AI inference models are deployed at concentrators and key relay nodes. These lightweight AI inference models undergo pruning and quantization, and their inference latency is reduced after deployment. ms; Predicted value based on channel quality parameters Preset data transmission priority Network load Equipment energy consumption status , Where 1 represents the lowest and 5 represents the highest, a multi-objective decision-making model is constructed: ， in, This means maximizing, that is, maximizing the following... Reach the maximum value, This represents the weight of the corresponding channel quality prediction result. This is the channel quality prediction result. This indicates the weight corresponding to the data transmission priority. Indicates the importance of the data to be transmitted. This indicates the weight corresponding to the network load. This indicates the current level of network activity. This indicates the weight of the corresponding device's energy consumption status. This indicates the percentage of remaining battery power in the device. It is the operating frequency band. It's the transmission power. It is the encoding rate. It's the modulation method, from Choose from these three options. This is the decision evaluation value.

8. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 7, characterized in that, The method in step S4 also includes: A non-dominated sorting genetic algorithm is used to solve the multi-objective decision-making model, obtaining the Pareto optimal decision set and generating... Individual decision-makers Indicate the population size, perform population initialization, based on the objective function. Calculate fitness values ​​while considering whether constraints are met. Individuals violating constraints have a fitness penalty of 0. Divide the population into different non-dominated levels. Individuals in the same level are ranked by crowding. Expression: ， in, For the first The degree of crowding of individuals Indicates the first The item in the first The feature value at time 1, Indicates the first The item in the first The feature value at time 1, Indicates the first The maximum value of each component. Indicates the first The minimum value of each component. To determine the target number, a roulette wheel selection method is used, with a crossover probability set to [value missing]. Simulated binary crossover is used, and the mutation probability is set. Polynomial mutation is employed; the previous version is retained. The best individual enters the next generation of the population, with the number of iterations being... Choose from the final Pareto optimal set The largest individual outputs the decision results.

9. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 8, characterized in that, In step S5, an intelligent dual-mode switching strategy is adopted to select the optimal communication mode based on the decision results. Simultaneously, distributed data processing is performed at the edge nodes as follows: By combining real-time channel feedback with decision results, a handover determination function is constructed, expressed as: ， In the formula, To switch the value of the decision function, The decision evaluation value for HPLC communication mode. The decision evaluation value for HRF communication mode. .

10. The method for improving power consumption data acquisition efficiency based on dual-mode and broadband carrier communication according to claim 8, characterized in that, The method in step S5 also includes: exist Select HPLC mode and execute the optimized parameter configuration in step S3. When selecting HRF mode, When dual-mode parallel transmission is enabled, the data splitting ratio is: ， In the formula, Indicates the data split ratio, HPLC transfer Proportional data, HRF transmission The load balancing scheduling of proportional data and dual-mode parallel transmission adopts a load balancing algorithm based on queuing theory to schedule dual-mode transmission traffic; the distributed data processing at the edge nodes includes data filtering and data compression.

Images