Electricity consumption information acquisition method and system based on dual-mode communication

Abstract

The invention relates to the technical field of smart power grids and communication, and discloses a power utilization information acquisition method and system based on dual-mode communication, and the method comprises the steps: 1, initializing a configuration network, building the neighborhood of each acquisition terminal node, carrying out the probability distribution parameterization of the channel states of a power line carrier channel and a wireless radio frequency channel, and carrying out the network configuration; according to the probability distribution parameterization, calculating a Fishi information matrix used for quantifying the information distance between different channel states; and step 2, defining a channel state geometrical relationship based on the Fishi information matrix. The probability distribution parameter space of the channel state is constructed into the information geometric manifold, the manifold Kalman filtering algorithm is adopted for state prediction, and compared with an existing channel prediction scheme adopting standard Kalman filtering or a simple time sequence model, the problem that the statistical characteristics of the channel cannot be accurately described due to model mismatch is solved, and the channel prediction accuracy is improved. Therefore, the problems of low prediction precision and incapability of effectively supporting a prospective routing decision are solved.

Description

Technical Field

[0001] This invention relates to the fields of smart grid and communication technology, specifically to a method and system for collecting electricity consumption information based on dual-mode communication. Background Technology

[0002] As a key component of smart grid construction, advanced metering architecture aims to achieve bidirectional, reliable, and efficient communication of massive amounts of electricity consumption information between data acquisition terminals and the master station system. To address the challenges of diverse electromagnetic interference and severe signal obstruction in complex residential and industrial environments, dual-mode communication technology combining power line carrier and radio frequency has become the mainstream technical solution for achieving full network coverage and high-reliability communication in current electricity consumption information collection. Communication utilizes power lines as the transmission medium, eliminating the need for additional wiring. Communication can flexibly bypass transformers, circuit breakers, etc. The natural signal blocking points complement each other, significantly improving the success rate of data acquisition.

[0003] However, in the practical application of existing dual-mode communication technology, the following technical bottlenecks still exist that restrict network performance: Traditional channel state prediction methods are usually based on Gaussian assumptions to build models. These model assumptions deviate significantly from the actual physical reality of the channel, resulting in a significant reduction in the accuracy of future channel state predictions. Consequently, upper-layer routing decisions cannot obtain reliable forward-looking information to support them, ultimately leading to a decrease in data transmission success rate and a deterioration in overall network performance.

[0004] In applications using reinforcement learning for intelligent routing decisions, the channel quality and service load of electricity information collection networks exhibit highly dynamic characteristics. If traditional policy gradient methods are used to update routing policies, the mismatch between the Euclidean geometry of the parameter space and the information geometry of the policy distribution will result in a large gradient estimation variance. This leads to frequent oscillations in the policy learning process, making it difficult to converge stably to the optimal solution. Ultimately, the network cannot quickly adapt to environmental changes, resulting in routing decisions remaining in a suboptimal state for a long time, leading to low transmission efficiency and wasted network resources.

[0005] To address these issues, this invention proposes a method and system for collecting electricity consumption information based on dual-mode communication. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for collecting electricity consumption information based on dual-mode communication, thereby solving the problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for collecting electricity consumption information based on dual-mode communication, comprising: Step 1: Initialize and configure the network, establish neighbor relationships for each acquisition terminal node, perform probability distribution parameterization on the channel states of the power line carrier channel and the radio frequency channel, and calculate the Fisher information matrix to quantify the information distance between different channel states based on the probability distribution parameterization. Step 2: Based on the channel state geometry defined by the Fisher information matrix, the manifold Kalman filter algorithm is used to predict the channel state at future times. Combined with the service quality level of the service data packets of the collected terminal nodes, local network state information containing the predicted channel state and the service quality level is generated. Step two further includes: Sub-step Channel state measurement updates; Acquiring the current time from the data acquisition terminal node Channel state measurement values Then, the update step of manifold Kalman filtering is performed to estimate the prior state from the previous time step. Make corrections; Through logarithmic mapping The measured value With the prior state estimation The difference is mapped onto the tangent space of the prior state estimation position to obtain the innovation vector. Then calculate the Kalman gain. : , in, Let be the prior state covariance matrix. For the measurement matrix, To measure the noise covariance matrix, This is the transpose operation of a matrix; Through exponential mapping The Kalman gain is used along the geodesic direction. With the new information vector The prior state estimate is updated to obtain the current time. Posterior state estimation And update the posterior state covariance matrix Obtain the updated joint channel parameter vector With the updated state covariance matrix ; Sub-step Prediction of the channel state at future moments; Based on sub-steps The updated joint channel parameter vector obtained With the updated state covariance matrix The prediction step of manifold Kalman filtering is performed; a preset state transition model is used in the updated joint channel parameter vector. In the tangent space of the position, the state is linearly extrapolated, and then the extrapolation result is projected back into the parameter space through an exponential mapping to obtain the next time step. Predicted joint channel parameter vector Simultaneously, the predicted state covariance matrix is ​​calculated. : , in, The Jacobian matrix of the state transition matrix. The process noise covariance matrix; The final output is the predicted joint channel parameter vector. ; Sub-step The generation of local network state information; The data acquisition terminal node obtains the service data packets to be transmitted and parses out the corresponding service quality level. Sub-steps The predicted joint channel parameter vector output in the middle , collect the length of the data transmission queue of the terminal node itself and the service quality level Integrate into a structured local network state information vector : , in, This is a local network state information vector. for Time Node The length of the data queue to be sent. The quality of service level for business data packets; The local network state information vector This will serve as the input for the distributed agent to make routing decisions in the subsequent step three; Step 3: Based on the local network state information, a routing strategy is generated by a distributed agent based on a graph neural network. The routing strategy determines a forwarding action that includes the next-hop node and the communication mode, and executes the forwarding action. Step four: Calculate the reward value based on the feedback information after executing the forwarding action, and update the graph neural network model parameters of the distributed agent using the natural policy gradient method to optimize the routing strategy.

[0008] Preferably, the channel state of the power line carrier channel adopts a symmetrical approach. The stable distribution is parameterized, and the channel state of the wireless radio frequency channel is adopted. The distribution is parameterized.

[0009] Preferably, the manifold Kalman filter algorithm performs state prediction and updating on the tangent space determined by the geometric relationship, and transforms between the tangent space and the parameter space representing the channel state through exponential and logarithmic mappings.

[0010] Preferably, the distributed intelligent agent adopts The framework, the graph neural network model includes a mechanism for generating the routing strategy. Network and the method for evaluating the value of the local network state information network.

[0011] Preferably, the reward value is calculated based on a utility function, which includes a channel capacity term, a transmission delay term, and an energy consumption term, and the weight coefficients of each term are dynamically determined by the service quality level.

[0012] Preferably, step one further includes: Sub-step Network topology discovery and initial neighbor relationship establishment; After the acquisition terminal node is powered on, it periodically broadcasts beacon frames through a preset guidance channel; each acquisition terminal node listens to the beacon frames, identifies neighboring nodes, and records the identity identifiers of the neighboring nodes and the link quality indicators of receiving the beacon frames through power line carrier and radio frequency modes, so as to establish and maintain the initial neighbor table. Sub-step Parameterization and initial estimation of the probability distribution of channel states; Based on the initial neighbor table, during the initial low-load phase, the data acquisition terminal nodes actively transmit a series of broadband probe signals to each neighbor node in power line carrier mode and radio frequency mode. Each data acquisition terminal node analyzes the received probe signals and uses a maximum likelihood estimation algorithm to parameterize the channel noise or fading characteristics of the power line carrier mode and radio frequency mode. The power line carrier channel noise is symmetrical. stable distribution Parameterization is performed, and the feature index is obtained through maximum likelihood estimation. With scale parameters : , in, These are the observed noise sample values; For characteristic index, For scale parameters, For position parameters, For integration variables, is the base of the natural logarithm. It is a cosine function. For the reason Integral operations up to infinity; The wireless radio frequency channel fading is adopted distributed Parameterization is performed, and the fading factor is obtained through maximum likelihood estimation. With average power : , in, for The probability density function of the distribution. The observed signal envelope amplitude, As the decay factor, Average power, It is a gamma function; All estimated parameters form a joint channel parameter vector. : ; Sub-step The calculation of the Fisher information matrix and the establishment of the channel state geometric relationship; Data collection terminal nodes based on sub-steps The obtained joint channel parameter vector Calculate the joint probability distribution of the power line carrier channel and the radio frequency channel. Based on the joint probability distribution, calculate the Fisher information matrix used to quantify the information distance between different channel states. ; The components of the Fisher information matrix Defined as: , in, For the expectation operator, and Joint channel parameter vector The parameter components in Let be the joint probability density function. The sign of the partial derivative. It is the natural logarithm function; The Fisher information matrix Define a Riemannian metric in the channel state parameter space to establish geometric relationships that describe the intrinsic characteristics of channel state evolution.

[0013] Preferably, step three further includes: Sub-step Graph neural aggregation of local network states; Data collection terminal node The local network state information vector generated in step two As the initial feature vector of the graph neural network ,pass Round-robin message passing, with neighboring node sets The nodes exchange and aggregate information to generate an aggregated state vector. In the In round-robin message passing, nodes Hidden state vector The update method is as follows: , in, For nodes In the The hidden state vector of the wheel, For nodes The set of neighboring nodes, Neighboring nodes In the The hidden state vector of the wheel, and For the first The learnable weight matrix of the round, It is a non-linear activation function; The final output aggregate state vector , containing nodes of Network status information within the hop neighborhood; Sub-step The generation of routing policies and the selection of forwarding actions; Sub-steps The obtained aggregation state vector Input to distributed intelligent agents The network, the The Internet is based on Functions with parameters ; The Network output and motion space Logical vectors with the same dimensions The logical vector pass Function converted to probability distribution That is, routing strategy: , , in, For action space The specific forwarding action in the middle, Choose an action given the current state. The probability, Logical vector Corresponding actions The amount, It is a natural exponential function; According to the routing strategy Perform random sampling and select specific forwarding actions. ; Sub-step The execution of the forwarding action; Based on sub-steps The forwarding action selected Data collection terminal node Execute packet forwarding operation, if node If the data sending queue is not empty, perform the following operations: First, the data packet is retrieved from the head of the sending queue; then, the destination physical address of the data packet is set as the next-hop node. The address; finally, according to the specified communication mode. Select either the power line carrier communication module or the wireless radio frequency communication module to send out the data packet.

[0014] Preferably, step four further includes: Sub-step Calculation of instant return value; Data collection terminal node The forwarding action described in step three is performed. Then, it receives confirmation feedback from the next-hop node, measures the transmission results, and bases them on the utility function. Calculate instant return value : , in, To perform the action The immediate reward value obtained afterwards To achieve through action The actual channel capacity of the selected link, To complete the end-to-end latency of this forwarding, The energy consumed to complete this forwarding , , The service quality levels of the business data packets are listed in order. Dynamically determined, when When it is a high priority, increase And decrease The value of , when For low-power services, increase The value of ; Sub-step Calculation of the dominant function value; Based on sub-steps The instant reward value calculated in and the execution of actions The next state that the network enters Calculate the calculation used to evaluate the forwarding action. Advantage function value of relative advantages and disadvantages ; First use distributed intelligent agents network Calculate the current state State value Next state State value The dominant function value The calculation method is as follows: , in, The dominant function value, for network, for Network model parameters, Discount factor; Sub-step , Network and Updating network model parameters; Based on sub-steps The value of the advantage function calculated in ,right Network and The network's model parameters are updated first by minimizing the square of the time difference error. Network model parameters , where the loss function for The update rules are as follows: , in, for The learning rate of the network For gradient operators; Update using the natural policy gradient method Network model parameters Computational strategy Fisher Information Matrix Furthermore, the gradient of the standard policy is scaled, and the update rule is as follows: , in, for The learning rate of the network; For the reason Network-defined routing policies Let f(x) be the inverse of the Fisher information matrix of the strategy. This is the gradient operator.

[0015] A power consumption information collection system based on dual-mode communication, the power consumption information collection system comprising: The initialization configuration module is used to establish the neighbor relationship of the acquisition terminal nodes, parameterize the probability distribution of the channel state, and calculate the Fisher information matrix used to determine the geometric relationship between the channel states. The state prediction module is used to predict the channel state at future times based on the geometric relationship using a manifold Kalman filter algorithm, and generate local network state information by combining the quality of service level of the service data packets. The routing decision module is used to generate and execute forwarding actions containing the next-hop node and communication mode through a distributed intelligent agent based on a graph neural network, based on the local network state information. The strategy optimization module is used to calculate the reward value based on the feedback information after the forwarding action is executed, and to update the graph neural network model parameters of the distributed agent using the natural policy gradient method.

[0016] Preferably, the routing decision module and the strategy optimization module together constitute a system employing... The framework's distributed intelligent agent, wherein the routing decision module includes a function for generating routing policies based on aggregated local network state information. network; The strategy optimization module utilizes information used to evaluate the value of the local network state information. The network calculates the advantage function, and then applies the advantage function to the network. Network and the The network's model parameters are updated.

[0017] This invention provides a method and system for collecting electricity consumption information based on dual-mode communication. It has the following beneficial effects: 1. This invention constructs the probability distribution parameter space of the channel state as an information geometric manifold and uses the manifold Kalman filter algorithm for state prediction. Compared with existing channel prediction schemes that use standard Kalman filtering or simple time series models, this invention solves the problem that the prediction accuracy is low and it cannot effectively support forward-looking routing decisions because the model mismatch cannot accurately characterize the channel statistical characteristics.

[0018] 2. This invention is based on The framework is optimized and updated using the natural policy gradient method. The network routing strategy enables fast and stable convergence of routing strategies in complex network environments, solving the problems of oscillations caused by large gradient variance during policy updates, unstable learning processes, and slow convergence speed in existing reinforcement learning techniques. Attached Figure Description

[0019] Figure 1 This is a flowchart of the present invention; Figure 2 This is a system framework diagram of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to 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 some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0021] The present invention will now be described in detail with reference to the accompanying drawings: Example: Please see the appendix Figure 1 This invention provides a method for collecting electricity consumption information based on dual-mode communication, comprising: Step 1: Initialize and configure the network, establish neighbor relationships for each acquisition terminal node, perform probability distribution parameterization on the channel states of the power line carrier channel and the radio frequency channel, and calculate the Fisher information matrix to quantify the information distance between different channel states based on the probability distribution parameterization. Step 2: Based on the channel state geometry defined by the Fisher information matrix, the manifold Kalman filter algorithm is used to predict the channel state at future time. Combined with the service quality level of the service data packets collected from the terminal node, local network state information containing the predicted channel state and service quality level is generated. Step 3: Based on the local network state information, a routing strategy is generated by a distributed agent based on a graph neural network. The routing strategy determines the forwarding action, which includes the next-hop node and the communication mode, and executes the forwarding action. Step four: Calculate the reward value based on the feedback information after the forwarding action is performed, and update the parameters of the graph neural network model of the distributed agent using the natural policy gradient method to optimize the routing strategy.

[0022] Step one further includes: Sub-step Network topology discovery and initial neighbor relationship establishment; After the acquisition terminal node is powered on, it periodically broadcasts beacon frames through a preset guidance channel; each acquisition terminal node listens to the beacon frames, identifies neighboring nodes, and records the identity identifiers of neighboring nodes and the link quality indicators of receiving beacon frames through power line carrier and radio frequency modes, in order to establish and maintain the initial neighbor table. Sub-step Parameterization and initial estimation of the probability distribution of channel states; Based on the initial neighbor table, during the initial low-load phase, the data acquisition terminal nodes actively transmit a series of broadband probe signals to each neighbor node in power line carrier mode and radio frequency mode. Each data acquisition terminal node analyzes the received probe signals and uses a maximum likelihood estimation algorithm to parameterize the channel noise or fading characteristics of the power line carrier mode and radio frequency mode. Power line carrier channel noise is symmetrical stable distribution Parameterization is performed, and the feature index is obtained through maximum likelihood estimation. With scale parameters : , in, These are the observed noise sample values; For characteristic index, For scale parameters, For position parameters, For integration variables, is the base of the natural logarithm. It is a cosine function. For the reason Integral operations up to infinity; Wireless radio frequency channel fading distributed Parameterization is performed, and the fading factor is obtained through maximum likelihood estimation. With average power : , in, for The probability density function of the distribution. The observed signal envelope amplitude, As the decay factor, Average power, It is a gamma function; All estimated parameters form a joint channel parameter vector. : ; Sub-step The calculation of the Fisher information matrix and the establishment of the channel state geometric relationship; Data collection terminal nodes based on sub-steps The obtained joint channel parameter vector Calculate the joint probability distribution of the power line carrier channel and the radio frequency channel. Based on the joint probability distribution, the Fisher information matrix used to quantify the information distance between different channel states is calculated. ; Components of the Fisher Information Matrix Defined as: , in, For the expectation operator, and Joint channel parameter vector The parameter components in Let be the joint probability density function. The sign of the partial derivative. It is the natural logarithm function; Fisher Information Matrix Define a Riemannian metric in the channel state parameter space to establish geometric relationships that describe the intrinsic characteristics of channel state evolution.

[0023] Step two further includes: Sub-step Channel state measurement updates; Acquiring the current time from the data acquisition terminal node Channel state measurement values Then, the update step of manifold Kalman filtering is performed to estimate the prior state from the previous time step. Make corrections; Through logarithmic mapping Measurement values Compared with prior state estimation The difference is mapped onto the tangent space of the prior state estimation location to obtain the innovation vector. Then calculate the Kalman gain. : , in, Let be the prior state covariance matrix. For the measurement matrix, To measure the noise covariance matrix, This is the transpose operation of a matrix; Through exponential mapping Using Kalman gain along the geodesic direction With the new vector Update the prior state estimate to obtain the current time step. Posterior state estimation And update the posterior state covariance matrix Obtain the updated joint channel parameter vector With the updated state covariance matrix ; Sub-step Prediction of the channel state at future moments; Based on sub-steps The updated joint channel parameter vector obtained With the updated state covariance matrix The prediction step of manifold Kalman filtering is performed; a preset state transition model is used in the updated joint channel parameter vector. In the tangent space of the position, the state is linearly extrapolated, and then the extrapolation result is projected back into the parameter space through an exponential mapping to obtain the next time step. Predicted joint channel parameter vector Simultaneously, the predicted state covariance matrix is ​​calculated. : , in, The Jacobian matrix of the state transition matrix. The process noise covariance matrix; The final output is the predicted joint channel parameter vector. ; Sub-step The generation of local network state information; The data acquisition terminal node obtains the service data packets to be transmitted and parses out the corresponding service quality level. Sub-steps The predicted joint channel parameter vector output by the middle , collect the length of the data transmission queue of the terminal node itself and service quality level Integrate into a structured local network state information vector : , in, This is a local network state information vector. for Time Node The length of the data queue to be sent. The quality of service level for business data packets; Local network state information vector This will serve as the input for the distributed agent to make routing decisions in the subsequent step three.

[0024] Step three further includes: Sub-step Graph neural aggregation of local network states; Data collection terminal node The local network state information vector generated in step two As the initial feature vector of the graph neural network ,pass Round-robin message passing, with neighboring node sets The nodes exchange and aggregate information to generate an aggregated state vector. In the In round-robin message passing, nodes Hidden state vector The update method is as follows: , in, For nodes In the The hidden state vector of the wheel, For nodes The set of neighboring nodes, Neighboring nodes In the The hidden state vector of the wheel, and For the first The learnable weight matrix of the round, It is a non-linear activation function; The final output aggregate state vector , containing nodes of Network status information within the hop neighborhood; Sub-step The generation of routing policies and the selection of forwarding actions; Sub-steps The obtained aggregate state vector Input to distributed intelligent agents network, The Internet is based on Functions with parameters ; Network output and motion space Logical vectors with the same dimensions Logical vector pass Function converted to probability distribution That is, routing strategy: , , in, For action space The specific forwarding action in the middle, Choose an action given the current state. The probability, Logical vector Corresponding actions The amount, It is a natural exponential function; According to the routing policy Perform random sampling and select specific forwarding actions. ; Sub-step The execution of the forwarding action; Based on sub-steps Selected forwarding action Data collection terminal node Execute packet forwarding operation, if node If the data sending queue is not empty, perform the following operations: First, the data packet is retrieved from the head of the sending queue; then, the destination physical address of the data packet is set as the next-hop node. The address; finally, according to the specified communication mode. Select either the power line carrier communication module or the wireless radio frequency communication module to send out the data packets.

[0025] Step four further includes: Sub-step Calculation of instant return value; Data collection terminal node The forwarding action in step three is performed. Then, it receives confirmation feedback from the next-hop node, measures the transmission results, and bases them on the utility function. Calculate instant return value : , in, To perform the action The immediate reward value obtained afterwards To achieve through action The actual channel capacity of the selected link, To complete the end-to-end latency of this forwarding, The energy consumed to complete this forwarding , , The service quality levels of the business data packets are listed in order. Dynamically determined, when When it is a high priority, increase And decrease The value of , when For low-power services, increase The value of ; Sub-step Calculation of the dominant function value; Based on sub-steps Instant return value calculated in and the execution of actions The next state that the network enters Calculations used to evaluate forwarding actions Advantage function value of relative advantages and disadvantages ; First use distributed intelligent agents network Calculate the current state State value Next state State value Advantage function value The calculation method is as follows: , in, The dominant function value, for network, for Network model parameters, Discount factor; Sub-step , Network and Updating network model parameters; Based on sub-steps The dominant function value calculated in the middle ,right Network and The network's model parameters are updated first by minimizing the square of the time difference error. Network model parameters , where the loss function for The update rules are as follows: , in, for The learning rate of the network For gradient operators; Update using the natural policy gradient method Network model parameters Computational strategy Fisher Information Matrix Furthermore, the gradient of the standard policy is scaled, and the update rule is as follows: , in, for The learning rate of the network; For the reason Network-defined routing policies Let f(x) be the inverse of the Fisher information matrix of the strategy. This is the gradient operator.

[0026] A power consumption information collection system based on dual-mode communication, the power consumption information collection system includes: The initialization configuration module is used to establish the neighbor relationship of the acquisition terminal nodes, parameterize the probability distribution of the channel state, and calculate the Fisher information matrix used to determine the geometric relationship between the channel states. The state prediction module is used to predict the channel state at future moments based on geometric relationships and using the manifold Kalman filter algorithm. Combined with the quality of service level of service data packets, it generates local network state information. The routing decision module is used to generate and execute forwarding actions containing the next-hop node and communication mode through a distributed intelligent agent based on graph neural network, based on local network state information. The strategy optimization module is used to calculate the reward value based on the feedback information after the forwarding action is executed, and to update the graph neural network model parameters of the distributed agent using the natural policy gradient method.

[0027] By parameterizing the dual-mode channel with a non-Gaussian probability distribution that conforms to its physical characteristics, and constructing a Fisher information matrix in the channel state parameter space, the advantage lies in establishing a high-fidelity mathematical model for complex and dynamic communication channels. This step can capture the impulse noise of power line channels and the fading characteristics of wireless channels, and innovatively endows the abstract channel state parameters with information geometry, providing a unified, intrinsic, and robust geometric framework for subsequent prediction and analysis. This fundamentally solves the problem that traditional models cannot accurately describe the channel due to mismatch.

[0028] Employing a manifold Kalman filter algorithm to predict channel states offers the advantage of predicting future states of non-Gaussian channels, thus providing forward-looking routing decisions. This step fully utilizes the channel geometry established in step one, filtering through a mapping between the tangent space and the parametric manifold. This perfectly adapts to the nonlinear characteristics of the channel parameter space, achieving prediction accuracy far exceeding that of traditional filters operating in Euclidean space.

[0029] The advantage of generating routing strategies using distributed agents based on graph neural networks lies in enabling nodes to make routing decisions with a holistic view of the local network and contextual awareness. Unlike greedy routing that relies on single-hop metrics, graph neural networks aggregate state information from multi-hop neighborhoods, allowing each node to understand the surrounding network topology, channel quality distribution, and congestion status, thus making intelligent, global considerations. This decision-making model effectively avoids potential congestion bottlenecks, discovers non-intuitive high-quality paths, and significantly improves the rationality of routing selection and the efficiency of network resource utilization.

[0030] The advantage of using the natural policy gradient method to update the model parameters of a distributed agent is that it ensures the learning process of the intelligent routing policy is efficient and stable in complex and dynamic environments. The natural policy gradient utilizes the Fisher information matrix of the policy distribution to correct the update direction, ensuring that parameter updates always point to the path with the fastest performance improvement in the policy space, effectively suppressing learning oscillations caused by large gradient variance in traditional policy gradient methods. This mechanism enables the agent to converge quickly and smoothly to a high-performance routing policy, ensuring that the network can continuously adaptively optimize and maintain efficient and reliable communication performance.

[0031] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for collecting electricity consumption information based on dual-mode communication, characterized in that, include: Step 1: Initialize and configure the network, establish neighbor relationships for each acquisition terminal node, perform probability distribution parameterization on the channel states of the power line carrier channel and the radio frequency channel, and calculate the Fisher information matrix to quantify the information distance between different channel states based on the probability distribution parameterization. Step 2: Based on the channel state geometry defined by the Fisher information matrix, the manifold Kalman filter algorithm is used to predict the channel state at future times. Combined with the service quality level of the service data packets of the collected terminal nodes, local network state information containing the predicted channel state and the service quality level is generated. Step two further includes: Sub-step Channel state measurement updates; Acquiring the current time from the data acquisition terminal node Channel state measurement values Then, the update step of manifold Kalman filtering is performed to estimate the prior state from the previous time step. Make corrections; Through logarithmic mapping The measured value With the prior state estimation The difference is mapped onto the tangent space of the prior state estimation position to obtain the innovation vector. Then calculate the Kalman gain. : ， in, Let be the prior state covariance matrix. For the measurement matrix, To measure the noise covariance matrix, This is the transpose operation of a matrix; Through exponential mapping The Kalman gain is used along the geodesic direction. With the new information vector The prior state estimate is updated to obtain the current time. Posterior state estimation And update the posterior state covariance matrix Obtain the updated joint channel parameter vector With the updated state covariance matrix ; Sub-step Prediction of the channel state at future moments; Based on sub-steps The updated joint channel parameter vector obtained With the updated state covariance matrix The prediction step of manifold Kalman filtering is performed; a preset state transition model is used in the updated joint channel parameter vector. In the tangent space of the position, the state is linearly extrapolated, and then the extrapolation result is projected back into the parameter space through an exponential mapping to obtain the next time step. Predicted joint channel parameter vector Simultaneously, the predicted state covariance matrix is ​​calculated. : ， in, The Jacobian matrix of the state transition matrix. The process noise covariance matrix; The final output is the predicted joint channel parameter vector. ; Sub-step The generation of local network state information; The data acquisition terminal node obtains the service data packets to be transmitted and parses out the corresponding service quality level. Sub-steps The predicted joint channel parameter vector output in the middle , collect the length of the data transmission queue of the terminal node itself and the service quality level Integrate into a structured local network state information vector : ， in, This is a local network state information vector. for Time Node The length of the data queue to be sent. The quality of service level for business data packets; The local network state information vector This will serve as the input for the distributed agent to make routing decisions in the subsequent step three; Step 3: Based on the local network state information, a routing strategy is generated by a distributed agent based on a graph neural network. The routing strategy determines a forwarding action that includes the next-hop node and the communication mode, and executes the forwarding action. Step four: Calculate the reward value based on the feedback information after executing the forwarding action, and update the graph neural network model parameters of the distributed agent using the natural policy gradient method to optimize the routing strategy.

2. The method for collecting electricity consumption information based on dual-mode communication according to claim 1, characterized in that, The channel state of the power line carrier channel adopts a symmetrical approach. The stable distribution is parameterized, and the channel state of the wireless radio frequency channel is adopted. The distribution is parameterized.

3. The method for collecting electricity consumption information based on dual-mode communication according to claim 1, characterized in that, The manifold Kalman filter algorithm performs state prediction and updates in the tangent space determined by the geometric relationship, and transforms between the tangent space and the parameter space representing the channel state through exponential and logarithmic mappings.

4. The method for collecting electricity consumption information based on dual-mode communication according to claim 1, characterized in that, The distributed intelligent agent adopts The framework, the graph neural network model includes a mechanism for generating the routing strategy. Network and the method for evaluating the value of the local network state information network.

5. The method for collecting electricity consumption information based on dual-mode communication according to claim 1, characterized in that, The reward value is calculated based on a utility function, which includes a channel capacity term, a transmission delay term, and an energy consumption term. The weight coefficients of each term are dynamically determined by the quality of service level.

6. The method for collecting electricity consumption information based on dual-mode communication according to claim 1 is characterized in that, Step one further includes: Sub-step Network topology discovery and initial neighbor relationship establishment; After the acquisition terminal node is powered on, it periodically broadcasts beacon frames through a preset guidance channel; each acquisition terminal node listens to the beacon frames, identifies neighboring nodes, and records the identity identifiers of the neighboring nodes and the link quality indicators of receiving the beacon frames through power line carrier and radio frequency modes, so as to establish and maintain the initial neighbor table. Sub-step Parameterization and initial estimation of the probability distribution of channel states; Based on the initial neighbor table, during the initial low-load phase, the data acquisition terminal nodes actively transmit a series of broadband probe signals to each neighbor node in power line carrier mode and radio frequency mode. Each data acquisition terminal node analyzes the received probe signals and uses a maximum likelihood estimation algorithm to parameterize the channel noise or fading characteristics of the power line carrier mode and radio frequency mode. The power line carrier channel noise is symmetrical. stable distribution Parameterization is performed, and the feature index is obtained through maximum likelihood estimation. With scale parameters : ， in, These are the observed noise sample values; For characteristic index, For scale parameters, For position parameters, For integration variables, is the base of the natural logarithm. It is a cosine function. For the reason Integral operations up to infinity; The wireless radio frequency channel fading is adopted distributed Parameterization is performed, and the fading factor is obtained through maximum likelihood estimation. With average power : ， in, for The probability density function of the distribution. The observed signal envelope amplitude, As the decay factor, Average power, It is a gamma function; All estimated parameters form a joint channel parameter vector. : ； Sub-step The calculation of the Fisher information matrix and the establishment of the channel state geometric relationship; Data collection terminal nodes based on sub-steps The obtained joint channel parameter vector Calculate the joint probability distribution of the power line carrier channel and the radio frequency channel. Based on the joint probability distribution, calculate the Fisher information matrix used to quantify the information distance between different channel states. ; The components of the Fisher information matrix Defined as: ， in, For the expectation operator, and Joint channel parameter vector The parameter components in Let be the joint probability density function. The sign of the partial derivative. It is the natural logarithm function; The Fisher information matrix Define a Riemannian metric in the channel state parameter space to establish geometric relationships that describe the intrinsic characteristics of channel state evolution.

7. The method for collecting electricity consumption information based on dual-mode communication according to claim 1, characterized in that, Step three further includes: Sub-step Graph neural aggregation of local network states; Data collection terminal node The local network state information vector generated in step two As the initial feature vector of the graph neural network ,pass Round-robin message passing, with neighboring node sets The nodes exchange and aggregate information to generate an aggregated state vector. In the In round-robin message passing, nodes Hidden state vector The update method is as follows: ， in, For nodes In the The hidden state vector of the wheel, For nodes The set of neighboring nodes, Neighboring nodes In the The hidden state vector of the wheel, and For the first The learnable weight matrix of the round, It is a non-linear activation function; The final output aggregate state vector , containing nodes of Network status information within the hop neighborhood; Sub-step The generation of routing policies and the selection of forwarding actions; Sub-steps The obtained aggregation state vector Input to distributed intelligent agents The network, the The Internet is based on Functions with parameters ; The Network output and motion space Logical vectors with the same dimensions The logical vector pass Function converted to probability distribution That is, routing strategy: ， ， in, For action space The specific forwarding action in the middle, Choose an action given the current state. The probability, Logical vector Corresponding actions The amount, It is a natural exponential function; According to the routing strategy Perform random sampling and select specific forwarding actions. ; Sub-step The execution of the forwarding action; Based on sub-steps The forwarding action selected Data collection terminal node Execute packet forwarding operation, if node If the data sending queue is not empty, perform the following operations: First, the data packet is retrieved from the head of the sending queue; then, the destination physical address of the data packet is set as the next-hop node. The address; finally, according to the specified communication mode. Select either the power line carrier communication module or the wireless radio frequency communication module to send out the data packet.

8. The method for collecting electricity consumption information based on dual-mode communication according to claim 1, characterized in that, Step four further includes: Sub-step Calculation of instant return value; Data collection terminal node The forwarding action described in step three is performed. Then, it receives confirmation feedback from the next-hop node, measures the transmission results, and bases them on the utility function. Calculate instant return value : ， in, To perform the action The immediate reward value obtained afterwards To achieve through action The actual channel capacity of the selected link, To complete the end-to-end latency of this forwarding, The energy consumed to complete this forwarding , , The quality of service levels of the service data packets are ranked in order. Dynamically determined, when When it is a high priority, increase And decrease The value of , when For low-power services, increase The value of ; Sub-step Calculation of the dominant function value; Based on sub-steps The instant reward value calculated in and the execution of actions The next state that the network enters Calculate the calculation used to evaluate the forwarding action. Advantage function value of relative advantages and disadvantages ; First use distributed intelligent agents network Calculate the current state State value Next state State value The dominant function value The calculation method is as follows: ， in, The dominant function value, for network, for Network model parameters, Discount factor; Sub-step , Network and Updating network model parameters; Based on sub-steps The value of the advantage function calculated in ,right Network and The network's model parameters are updated first by minimizing the square of the time difference error. Network model parameters , where the loss function for The update rules are as follows: ， in, for The learning rate of the network For gradient operators; Update using the natural policy gradient method Network model parameters Computational strategy Fisher Information Matrix Furthermore, the gradient of the standard policy is scaled, and the update rule is as follows: ， in, for The learning rate of the network; For the reason Network-defined routing policies Let f(x) be the inverse of the Fisher information matrix of the strategy. This is the gradient operator.

9. A power consumption information acquisition system based on dual-mode communication, comprising a power consumption information acquisition method based on dual-mode communication according to any one of claims 1 to 9, characterized in that, The electricity consumption information collection system includes: The initialization configuration module is used to establish the neighbor relationship of the acquisition terminal nodes, parameterize the probability distribution of the channel state, and calculate the Fisher information matrix used to determine the geometric relationship between the channel states. The state prediction module is used to predict the channel state at future times based on the geometric relationship using a manifold Kalman filter algorithm, and generate local network state information by combining the quality of service level of the service data packets. The routing decision module is used to generate and execute forwarding actions containing the next-hop node and communication mode through a distributed intelligent agent based on a graph neural network, based on the local network state information. The strategy optimization module is used to calculate the reward value based on the feedback information after the forwarding action is executed, and to update the graph neural network model parameters of the distributed agent using the natural policy gradient method.

10. A power consumption information acquisition system based on dual-mode communication according to claim 9, characterized in that, The routing decision module and the strategy optimization module together constitute a system employing... The framework's distributed intelligent agent, wherein the routing decision module includes a function for generating routing policies based on aggregated local network state information. network; The strategy optimization module utilizes information used to evaluate the value of the local network state information. The network calculates the advantage function, and then applies the advantage function to the network. Network and the The network's model parameters are updated.

Images