A 6g passive internet of things intelligent energy scheduling method and system for positioning task

Abstract

The application discloses a 6G passive Internet of Things intelligent energy scheduling method and system for positioning tasks, and belongs to the technical field of passive Internet of Things positioning and intelligent scheduling. The method collects energy storage capacitor voltage sequences and environmental parameter sequences on the terminal side, constructs a lightweight energy risk prediction model, obtains a basic under-voltage risk probability, generates a positioning risk probability through time bias correction, and reports events in stages; on the base station side, the method fuses staged reporting, position prediction and historical dead point observation data, constructs a multipath dead point probability and a comprehensive dead point exposure probability; further, the current voltage, environmental parameters, positioning risk, staged reporting, dead point exposure probability and last round scheduling action are taken as state inputs, a deep deterministic policy gradient model is used to output a positioning sampling period and modulation information quantity, and closed-loop updating is performed in combination with task execution results, energy changes and dead point exposure feedback, so that the energy supply continuity and execution stability of passive positioning tasks are improved.

Description

Technical Field

[0001] This invention relates to the field of passive IoT positioning and intelligent scheduling technology, specifically to a 6G passive IoT intelligent energy scheduling method and system for positioning tasks. Background Technology

[0002] With the development of 6G's integrated sensing and ultra-large-scale connectivity scenarios, passive IoT, with its advantages of being battery-free, low-cost, and easy to deploy, is gradually becoming an important technological direction for building green and ubiquitous sensing networks. In application scenarios such as logistics tracking, industrial manufacturing, warehouse inspection, and equipment monitoring, the demand for location-based services is constantly increasing, placing higher requirements on the continuity, robustness, and intelligence of passive positioning systems.

[0003] One type of existing passive IoT positioning technology mainly focuses on the selection and switching of spatial positioning methods. For example, in mixed indoor and outdoor scenarios, a scene evaluation function can be used to distinguish between indoor, transitional areas, and outdoor environments, and passive IoT information can be used for fingerprint positioning, GNSS / SINS fusion positioning, or a transitional fusion positioning between the two to improve positioning continuity when switching between indoor and outdoor environments. This type of method can alleviate the problem of inconsistent reliability of observation sources in different scenarios at the spatial level, but its main focus remains on scene recognition and positioning mode switching.

[0004] However, in passive positioning systems, positioning interruptions are not always caused by improper positioning mode selection. Because passive nodes are highly dependent on external radio frequency power supply, their operating status is easily affected by factors such as propagation distance, multipath fading, obstruction changes, temperature and humidity disturbances, and spatial energy dead zones, leading to power supply intermittents, voltage drops, missing observations, and positioning performance degradation. Even if the positioning mode itself is correctly selected, if the passive tag enters an undervoltage or power-off state before the next round of positioning sampling, it will still cause positioning observation interruptions, resulting in track breakage or continuous tracking failure.

[0005] Most existing energy management and scheduling methods rely on rule-based judgment based on current voltage or fixed thresholds, representing a typical reactive approach. While these methods can achieve basic resource control in static or weakly dynamic environments, they lack forward-looking awareness of future energy evolution trends and struggle to handle sudden energy drops in complex multipath environments and rapidly changing conditions. Particularly in positioning missions, there is a significant coupling between node energy state and positioning sampling period and the amount of positioning modulation information: more frequent sampling and higher modulation load result in higher observation refresh rates and stronger information carrying capacity, but also greater energy consumption per unit time, making undervoltage interruptions more likely. Scheduling based solely on the current state often fails to achieve a reasonable balance between positioning continuity and energy security.

[0006] On the other hand, multipath dead spots in complex indoor environments exhibit distinct spatial distribution and historical statistical characteristics. Scheduling based solely on the current voltage status of a single terminal fails to reflect the potential propagation risk areas that the terminal may expose during future movement, and it is also difficult to utilize the spatial prior information implicit in historical undervoltage, power outages, or positioning anomalies of other nodes. Consequently, the scheduling strategy lacks a forward-looking spatial understanding of the potential exposure risks of future dead spots.

[0007] Furthermore, although methods such as deep reinforcement learning can be used for dynamic resource scheduling, if the scheduling state only depends on the current voltage or immediate task feedback, without simultaneously integrating terminal-side positioning risks, multipath dead point exposure probability, and energy state changes after task execution, it is difficult to form a closed-loop optimization mechanism for continuous execution of positioning tasks. Summary of the Invention

[0008] This invention provides a 6G passive IoT intelligent energy scheduling method and system for positioning tasks. It performs forward-looking perception and hierarchical reporting of energy risks on the terminal side, integrates multipath dead point exposure risks and outputs continuous scheduling actions on the base station side, and performs closed-loop correction of the scheduling strategy through task execution feedback to improve the power supply continuity and execution stability of passive positioning tasks.

[0009] A 6G passive IoT intelligent energy dispatching method for positioning tasks includes the following steps:

[0010] Step 1: At the passive tag terminal, collect the voltage time series and environmental parameter time series of the energy storage capacitor, construct a lightweight energy risk prediction model that integrates voltage evolution trend and environmental correction information, and obtain the basic undervoltage risk probability for the next scheduling cycle.

[0011] Step 2: Based on the time relationship between the positioning sampling time and the prediction time of the basic undervoltage risk probability, perform time offset correction on the basic undervoltage risk probability, generate the positioning risk probability for the positioning task, and generate hierarchical reporting events based on the dual threshold comparison mechanism composed of high risk threshold and low risk threshold.

[0012] Step 3: On the base station side, the received hierarchical reporting events, the target tag's location prediction information, and other tags' historical dead point observation data are fused to construct the multipath dead point probability, and further generate the target tag's comprehensive multipath dead point exposure probability.

[0013] Step 4: Combine the current voltage of the passive tag terminal, environmental parameters, the location risk probability, the hierarchical reporting event, the comprehensive multipath dead point exposure probability, and the previous round of scheduling actions into a state vector, input it into a preset deep reinforcement learning model, and output continuous scheduling actions containing the next round of location sampling period and location modulation information, so as to realize continuous and adjustable control of the location task intensity.

[0014] Step 5: The base station sends the continuous scheduling action to the passive tag terminal. The passive tag terminal performs the positioning task according to the continuous scheduling action and sends a feedback vector containing the task execution result and energy state change back to the base station.

[0015] Step 6: The base station uses the feedback vector to construct a corrected reward, and performs online updates and policy evolution on the deep reinforcement learning model based on the experience replay mechanism to form a closed-loop scheduling optimization.

[0016] A 6G passive IoT intelligent energy dispatching system for positioning tasks includes the following modules:

[0017] Passive tag terminal energy risk prediction and reporting module: Deployed in the passive tag terminal, it is used to collect the voltage time series of the energy storage capacitor and the environmental parameter time series, construct a lightweight energy risk prediction model that integrates voltage evolution trend and environmental correction information, obtain the basic undervoltage risk probability for the next scheduling cycle; and perform time offset correction on the basic undervoltage risk probability according to the time relationship between the positioning sampling time and the prediction time of the basic undervoltage risk probability, generate the positioning risk probability for the positioning task, and trigger hierarchical reporting events based on preset high-risk thresholds and low-risk thresholds;

[0018] Base station-side multipath dead point probability construction module: Deployed at the base station, it is used to receive the hierarchical reporting events, fuse the target label's location prediction information and other labels' historical dead point observation data, discretize the target area into multiple grid cells, calculate the historical experience dead point probability, immediate dead point evidence amount and neighborhood diffusion risk of each grid cell, obtain the multipath dead point probability of each grid cell through weighted fusion and recursive update, and generate the comprehensive multipath dead point exposure probability of the target label;

[0019] The base station-side deep reinforcement learning scheduling decision module is deployed in the base station and is used to combine the current operating voltage, environmental parameters, positioning risk probability, hierarchical reporting events, comprehensive multipath dead point exposure probability, and the previous round of scheduling actions of the passive tag terminal into a state vector, input it into a preset deep reinforcement learning model, output continuous scheduling actions containing the next round of positioning sampling period and positioning modulation information, and send them to the passive tag terminal for execution.

[0020] Closed-loop feedback and policy evolution module: Deployed at the base station, it is used to receive the closed-loop feedback vector returned by the passive tag terminal after performing the continuous scheduling action. The closed-loop feedback vector includes the location risk probability, task execution result, voltage change, and comprehensive multipath dead point exposure feedback value. Based on the closed-loop feedback vector, a modified reward is constructed, and the deep reinforcement learning model is updated online through empirical sample reweighting, weighted Critic loss function, and Actor network to perform policy evolution.

[0021] Compared with the prior art, the present invention has at least the following beneficial effects:

[0022] First, this invention collects the energy storage capacitor voltage sequence and environmental parameter sequence at the passive tag terminal side and constructs a lightweight energy risk prediction model, enabling proactive judgment of the undervoltage risk of the passive tag before the positioning task is executed. Compared with the existing technology that only relies on the current voltage or a fixed threshold for ex-post control, this invention can identify the energy decline trend and the impact of environmental disturbances on the power supply status in advance, and perform time offset correction and hierarchical reporting in conjunction with the positioning sampling time, so that the base station side obtains risk information closer to the next positioning task execution time, thereby reducing the probability of the passive tag experiencing undervoltage, power failure, or missing observations before positioning sampling, and improving the continuity and stability of positioning task execution.

[0023] Second, this invention integrates terminal hierarchical reporting, target tag location prediction information, and historical dead-point observation data from multiple tags at the base station side to construct multipath dead-point probabilities and comprehensive multipath dead-point exposure probabilities. Compared to existing technologies that schedule based solely on the current energy state of a single terminal, this invention utilizes spatial location, historical observations, and neighborhood propagation information to characterize the distribution of multipath dead points in complex environments. This enables the scheduling strategy to possess spatial forward-looking perception capabilities for future propagation risk areas, thereby reducing the likelihood of energy interruption, observation failure, and positioning failure when passive tags enter high-risk dead-point areas.

[0024] Third, this invention employs a deep deterministic strategy gradient model to continuously and jointly schedule the positioning sampling period and the amount of positioning modulation information, and updates the scheduling strategy in a closed loop based on task execution results, energy state changes, and dead-point exposure feedback. Compared to fixed-period, fixed-load, or simple rule-based scheduling methods, this invention can dynamically adjust the intensity of positioning tasks according to the terminal's energy state, positioning risk, and dead-point exposure risk. In high-risk states, it reduces the positioning load to maintain node survival, and in low-risk or stable states, it increases the positioning refresh rate and information carrying capacity. This achieves a more granular balance between energy security and positioning performance, and enhances the system's adaptability and engineering feasibility in complex dynamic environments.

[0025] In summary, this invention can improve the power supply continuity, positioning stability, and adaptability to complex environments in passive IoT positioning tasks, and is suitable for application scenarios that require low power consumption, continuous positioning, and high-reliability sensing, such as warehousing and logistics, industrial inspection, asset tracking, and smart parks. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0027] Figure 2 This is a schematic diagram of the terminal-side energy risk prediction and tiered reporting process;

[0028] Figure 3 This is a schematic diagram illustrating the multipath dead point probability construction on the base station side;

[0029] Figure 4 This is a schematic diagram of the closed-loop feedback DDPG intelligent scheduling structure;

[0030] Figure 5 This is the convergence graph of the closed-loop feedback DDPG training. Detailed Implementation

[0031] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0032] like Figure 1 The schematic diagram of the overall process of the present invention shown is as follows: A 6G passive IoT intelligent energy scheduling method for positioning tasks includes the following steps:

[0033] Step 1: At the passive tag terminal, collect the voltage time series and environmental parameter time series of the energy storage capacitor, construct a lightweight energy risk prediction model that integrates voltage evolution trend and environmental correction information, and obtain the basic undervoltage risk probability for the next scheduling cycle.

[0034] Step 2: Based on the time relationship between the positioning sampling time and the prediction time of the basic undervoltage risk probability, perform time offset correction on the basic undervoltage risk probability, generate the positioning risk probability for the positioning task, and generate hierarchical reporting events based on the dual threshold comparison mechanism composed of high risk threshold and low risk threshold.

[0035] Step 3: On the base station side, the received hierarchical reporting events, the target tag's location prediction information, and other tags' historical dead point observation data are fused to construct the multipath dead point probability, and further generate the target tag's comprehensive multipath dead point exposure probability.

[0036] Step 4: Combine the current voltage of the passive tag terminal, environmental parameters, the location risk probability, the hierarchical reporting event, the comprehensive multipath dead point exposure probability, and the previous round of scheduling actions into a state vector, input it into a preset deep reinforcement learning model, and output continuous scheduling actions containing the next round of location sampling period and location modulation information, so as to realize continuous and adjustable control of the location task intensity.

[0037] Step 5: The base station sends the continuous scheduling action to the passive tag terminal. The passive tag terminal performs the positioning task according to the continuous scheduling action and sends a feedback vector containing the task execution result and energy state change back to the base station.

[0038] Step 6: The base station uses the feedback vector to construct a corrected reward, and performs online updates and policy evolution on the deep reinforcement learning model based on the experience replay mechanism to form a closed-loop scheduling optimization.

[0039] like Figure 2 The diagram shown illustrates the terminal-side energy risk prediction and tiered reporting process: In step 1, the lightweight energy risk prediction model is constructed based on the voltage time series and environmental parameter time series of the passive tag terminal, used to obtain the basic undervoltage risk probability for the next scheduling cycle, and defines the passive tag terminal in the [missing information - likely a specific timeframe or period]. The energy storage capacitor terminal voltage collected at the end of each scheduling cycle is , For discrete scheduling periodic indexes, the environment parameter vector is:

[0040]

[0041] in, , , These represent temperature, humidity, air pressure, or equivalent environmental characteristics, respectively; the first-order voltage difference is defined as follows:

[0042]

[0043] in, Indicates the scheduling period voltage, Indicates the previous scheduling cycle The voltage;

[0044] The length of the sliding window is Then the past The average voltage change rate over each scheduling period is:

[0045]

[0046] in, Indicates the first The first-order voltage difference corresponding to each scheduling cycle The summation index within the sliding window;

[0047] The length of the sliding window is Then the past The average voltage change rate over each scheduling period is:

[0048]

[0049] Based on the average voltage change rate, a basic voltage trend prediction term is constructed:

[0050]

[0051] in, This is a smoothing factor used to adjust the degree of confidence that the lightweight energy risk prediction model has in historical voltage change trends. This is the summation index within the sliding window.

[0052] Define the change in environmental parameters between adjacent scheduling cycles as:

[0053]

[0054] Construct environmental correction terms based on the changes in the environmental parameters:

[0055]

[0056] in, This is an environmentally sensitive weight vector used to characterize the degree of influence of changes in different environmental parameters on the energy collection status of passive tag terminals;

[0057] By combining the basic voltage trend prediction term and the environmental correction term, the predicted voltage for the next scheduling cycle is obtained:

[0058]

[0059] in, This is the scaling factor for the environmental correction term, used to adjust the intensity of the impact of external environmental changes on the predicted voltage;

[0060] Then, the local fluctuation intensity can be constructed based on the discreteness of the voltage difference sequence within a short-time historical window:

[0061]

[0062] in, Used to characterize recent The degree of deviation of voltage change from the average trend within a scheduling cycle;

[0063] Let the interrupt voltage threshold of the passive tag terminal be... To prevent the denominator from being zero, a very small positive number is introduced. The normalized voltage survival margin is defined as:

[0064]

[0065] Finally, a mapping function is used to map the normalized voltage survival margin to the basic undervoltage risk probability:

[0066]

[0067] in, Indicates the first Given historical window information for each scheduling cycle, the next scheduling cycle... The passive tag terminal voltage is lower than the interrupt voltage threshold. The basic undervoltage risk probability; when the predicted voltage When the voltage is higher than the interruption voltage threshold and the normalized voltage survival margin is large, the probability of basic undervoltage risk approaches 0; when the predicted voltage is close to or lower than the interruption voltage threshold, or when the intensity of the local fluctuation increases, the probability of basic undervoltage risk increases.

[0068] Basic undervoltage risk probability Time offset correction is performed based on the positioning sampling time to generate a positioning risk probability for the positioning task. And generate hierarchical reporting events based on a dual threshold mechanism;

[0069] Define the current scheduling time as The scheduling cycle is The basic undervoltage risk probability is then... The corresponding predicted time is:

[0070]

[0071] Let the next positioning sampling time be Define the prediction time. With the next positioning sampling time The time deviation between them is:

[0072]

[0073] in, The symbol is used to characterize the relative order between the prediction time corresponding to the basic undervoltage risk probability and the next positioning sampling time;

[0074] When the predicted time is earlier than or equal to the next positioning sampling time, that is:

[0075]

[0076] Explain the probability of the basic undervoltage risk. This corresponds to an energy state judgment that is earlier than or equal to the execution time of the positioning task. In this case, as the interval between the prediction time and the positioning sampling time increases, the representativeness of the basic risk probability to the positioning sampling time decreases, requiring risk amplification compensation. The intermediate positioning risk in this scenario is defined as:

[0077]

[0078] in, This is the time deviation risk amplification factor, used to control the risk compensation intensity when the predicted time is earlier than the positioning sampling time; when hour, ,when When it increases, the intermediate value of the positioning risk It increases with increasing time deviation;

[0079] When the predicted time is later than the next positioning sampling time, that is:

[0080]

[0081] Explain the probability of the basic undervoltage risk. This corresponds to a future energy state judgment that occurs later than the location task execution time. In this case, as the prediction time lags further behind the location sampling time, the energy state at the time of the location task is further away from the predicted risk range, requiring risk attenuation correction. The intermediate location risk in this scenario is defined as:

[0082]

[0083] in, This is the time deviation risk attenuation coefficient; when hour, ,when When it increases, the intermediate value of the positioning risk It decays as the time deviation increases;

[0084] Therefore, the probability of positioning risk for positioning tasks is defined as follows:

[0085]

[0086] in, This indicates the probability of the passive tag terminal performing the next positioning task at the current scheduling moment, representing the risk to its security.

[0087] After obtaining the location risk probability Then, set a low-risk threshold. and high risk threshold ,satisfy:

[0088]

[0089] Based on the low-risk threshold and the high-risk threshold The terminal-side hierarchical reporting rules are defined as follows:

[0090] when When the passive tag terminal is determined to be in a high-risk state, a high-risk reporting event is triggered. The high-risk reporting event is used to instruct the base station-side scheduling model to increase the positioning sampling period or reduce the positioning modulation information in subsequent scheduling, so as to reduce the energy consumption per unit time and maintain node survival.

[0091] when When the passive tag terminal is determined to be in a low-risk state, a low-risk reporting event is triggered. The low-risk reporting event is used to instruct the base station-side scheduling model to reduce the positioning sampling period or increase the amount of positioning modulation information in subsequent scheduling, so as to improve the positioning refresh rate and the reporting information carrying capacity.

[0092] when When the passive tag terminal is in a balanced operating state, it is determined that it does not trigger a high-risk reporting event or a low-risk reporting event, or it triggers a stable state reporting event, so as to maintain the continuity and stability of the current scheduling state.

[0093] like Figure 3 The diagram shown illustrates the construction of multipath dead point probability on the base station side: In step 3, the base station side constructs the multipath dead point probability and generates the target label's comprehensive multipath dead point exposure probability;

[0094] The area to be located is discretized into... Let there be a reference grid cell, and the grid set be denoted as:

[0095]

[0096] in, It remains a discrete scheduling cycle index; for any grid cell Define a binary random variable:

[0097]

[0098] in, Indicates the first Within a scheduling cycle, the grid When in a multipath dead point or deep energy decay state, there are propagation conditions that lead to a significant decrease in the energy collection of passive tags; This indicates that the grid is not in a multipath dead point or deep energy fading state; the base station side is at the... Each scheduling cycle for the grid The dead point probability estimate is defined as:

[0099]

[0100] This yields the spatial dead point probability distribution vector for the current scheduling period:

[0101]

[0102] Target terminal Due to the uncertainty of the future location, a location prediction probability distribution is introduced; let the target terminal... Fall into the grid in the next scheduling cycle The probability is:

[0103]

[0104] in, Indicates the target terminal The predicted position in the next scheduling cycle, satisfying:

[0105]

[0106] Based on the terminal-side lightweight risk prediction and anomaly reporting results, the target terminal is obtained. In the Location risk probability per scheduling cycle Abnormal signs and charging recovery time And normalize the charging recovery time to Define the target terminal In the The normalized energy anomaly intensity for each scheduling cycle is:

[0107]

[0108] in, For non-negative weighting coefficients, satisfying:

[0109]

[0110] It is a monotonically increasing mapping function used to map comprehensive risk to... interval; Used to reflect the target terminal Risk of energy interruption for the positioning task corresponding to the current scheduling cycle; Used to characterize sudden blockage, power drop, or undervoltage events; Used to characterize the target terminal Normalized result of the time required to recover from a low voltage state to a state capable of supporting positioning tasks;

[0111] The target terminal is defined by combining the predicted location probability with the energy anomaly intensity. For the grid The amount of evidence for immediate dead points is:

[0112]

[0113] and target terminal The resulting immediate prior dead point probability is denoted as:

[0114]

[0115] Let the length be Within the sliding time window, historical observation samples from multiple terminals are displayed in the grid. Internally generated observation samples The number of samples that were judged as significant dead-point events was [number]. Then the grid The historical experience dead point probability is defined as:

[0116]

[0117] in, This is a smoothing constant used to avoid the denominator being zero;

[0118] To characterize the spatial neighborhood correlation of multipath dead points, let the grid be denoted as . The neighborhood set is The risk propagation volume in its neighborhood is defined as:

[0119]

[0120] in, Neighborhood weight coefficients Represents a grid neighborhood set The neighborhood grid in For neighborhood grid indexing, Represents the neighborhood grid In the Historical experience dead point probability for each scheduling cycle;

[0121] And satisfy:

[0122]

[0123] The neighborhood weight coefficient is set using Gaussian decay or inverse distance weighting based on the distance between the grid centers, so that the closer the neighboring grid is, the greater its influence on the current grid.

[0124] After obtaining immediate evidence from the target terminal, historical experience of the group, and diffusion information in the neighborhood space, a weighted fusion method is used to obtain the grid. Comprehensive dead point probability estimation:

[0125]

[0126] in, To merge weights, and satisfy:

[0127]

[0128] To balance the long-term stability of spatial dead-point distribution with the ability to respond to short-term sudden disturbances, the base station side, according to the first The comprehensive dead-point probability estimate obtained within a scheduling cycle is used to recursively update the grid dead-point probability for the next scheduling cycle:

[0129]

[0130] in, To update the coefficients, the update results are truncated to ensure probabilistic meaning.

[0131]

[0132] Based on the above grid dead point probability estimation, the target terminal is defined. The overall multipath dead point exposure probability in the next scheduling cycle is:

[0133]

[0134] in, Indicates the target terminal The expected overlap between the future location probability distribution and the spatial dead-point probability map; when the target terminal When there is a high probability of entering a high-risk grid in the future, Increase; when the target terminal When the future is mainly in areas with relatively stable transmission conditions, Decrease.

[0135] The significant dead-point events are used to statistically analyze the probability of dead points based on historical experience. The specific determination process is as follows:

[0136] set up Let be the set of terminals used for historical dead point statistics; for the The terminal in the first Within a scheduling cycle, located in the grid For a single observation sample, the indicator variable for the dead-point salient event is defined as:

[0137]

[0138] in, For the first The terminal in the first The probability of location risk per scheduling cycle For the first The normalized result of the charging recovery time required for a terminal to recover from a low voltage state to a state capable of supporting positioning tasks. This is an exception flag variable. To locate the risk threshold, This is the charging delay threshold;

[0139] when When this occurs, it indicates that the terminal faces a high risk of energy interruption for the positioning task within the current grid; when When this occurs, it indicates that the time required for the terminal to recover from a low-voltage state to a state capable of performing positioning tasks within the current grid is too long; when When a terminal experiences a sudden occlusion, power drop, or undervoltage event within the current grid, the observed sample is identified as a significant dead point event if any of the above conditions are met.

[0140] In length of Within the sliding time window, the grid Number of significant events at the mid-dead point The following results were obtained from the statistics of the indicator variables:

[0141]

[0142] in, For indicator functions, Indicates the first The terminal in the first Position within a scheduling cycle;

[0143] Correspondingly, grid Total number of observed samples for:

[0144]

[0145] This leads to the probability of dead points based on historical experience:

[0146]

[0147] This determination method distinguishes between transient voltage drops caused by instantaneous fluctuations and persistent energy supply shortages caused by spatial propagation anomalies by jointly judging the location risk probability, normalized charging recovery time, and abnormal marker variables. This provides a group historical observation basis for updating the multipath dead point probability on the base station side.

[0148] like Figure 4 The diagram shows a closed-loop feedback DDPG intelligent scheduling structure: The base station side uses a deep deterministic policy gradient algorithm to construct a deep reinforcement learning scheduling model, which maps the energy state, environmental state, positioning risk state, and multipath dead point exposure risk of the passive tag terminal into continuous scheduling actions. The deep reinforcement learning scheduling model includes an Actor network and a Critic network.

[0149] Define the base station-side reinforcement learning agent in the first... The state vector for each scheduling cycle is:

[0150]

[0151] in, Indicates that the passive tag terminal is in the first The current operating voltage for each scheduling cycle is used to reflect the current remaining energy level of the terminal. This represents the current multidimensional environmental parameter vector, used to describe the external environmental conditions that affect energy harvesting efficiency and propagation conditions; Indicates the target terminal In the The current location risk probability for each scheduling cycle, satisfying:

[0152]

[0153] in, For target terminal The probability of positioning risk;

[0154] This indicates the current triggered tiered reporting type, including high-risk reporting, low-risk reporting, stable status reporting, or no reporting. Indicates that the base station side is in the first The target terminal's comprehensive multipath dead point exposure prediction value for the next scheduling cycle can be obtained when making decisions in each scheduling cycle, and satisfies:

[0155]

[0156] in, For target terminal The comprehensive multipath dead point exposure probability in the next scheduling cycle is used to characterize the possibility that the terminal will enter a high-risk deep fading area during the next round of positioning task execution.

[0157] Indicates the previous round of scheduling actions:

[0158]

[0159] This indicates the previous positioning sampling period. This represents the amount of positioning modulation information from the previous round, used to explicitly characterize the impact of motion inertia on subsequent energy evolution;

[0160] Define the base station-side agent in the first... The continuous scheduling actions output in each scheduling cycle are:

[0161]

[0162] in, Indicates the next round of positioning sampling period. This indicates the amount of positioning modulation information for the next round; the range of values ​​for the action is:

[0163]

[0164] in, This indicates the minimum allowed positioning sampling period of the system. This indicates the maximum waiting time set in the project; when A smaller value indicates that the terminal performs positioning tasks at a higher frequency, thereby improving the positioning refresh rate and time resolution; when the value is smaller... As the value gradually increases, it indicates that the terminal extends the interval between two adjacent positioning tasks, thereby reducing the energy consumption per unit time. and These represent the minimum and maximum values ​​of the location modulation information, respectively;

[0165] To ensure that the agent balances terminal energy safety, localization performance, multipath dead point avoidance, and action smoothness during the learning process, the following weighted instantaneous reward function is constructed:

[0166]

[0167] in, , , , These are non-negative reward weighting coefficients, used to adjust the importance of energy safety reward, positioning performance reward, multipath dead point penalty, and resource consumption and motion smoothness in the overall reward, respectively.

[0168] The energy safety reward item is defined as follows:

[0169]

[0170] in, The current location risk probability; when the terminal's energy status is good and the risk of short-term interruption is low, The value is relatively small. The corresponding increase; when the terminal approaches the undervoltage boundary or there is a significant risk of power outage in the next scheduling cycle. As the value increases, The corresponding reduction guides the intelligent agent to prioritize avoiding dangerous actions that could cause terminal malfunction.

[0171] The positioning performance bonus item is defined as follows:

[0172]

[0173] in, and These are internal weighting coefficients used to adjust the relative contributions of the positioning sampling period and the amount of positioning modulation information to the gains in positioning performance. This is a smoothing constant used to avoid the denominator being zero; this reward term enables the agent to obtain higher positioning performance gains when outputting shorter positioning sampling periods and higher positioning modulation information, thereby avoiding the model from adopting an overly conservative low-load strategy in the long term in pursuit of absolute energy safety.

[0174] The multipath dead point penalty term is defined as follows:

[0175]

[0176] in, and The action load penalty coefficient under multipath dead point risk is used to adjust the weights of the positioning sampling period and the positioning modulation information in the dead point risk penalty. This term is used to characterize the coupling relationship between multipath dead point exposure risk and the aggressiveness of scheduling actions: when the terminal is in or about to enter a region with a high multipath dead point exposure probability, if a shorter positioning sampling period and a higher positioning modulation information are still selected, the multipath dead point penalty term increases; when the terminal is in a low dead point risk region, the penalty corresponding to the same scheduling action decreases, thereby enabling the agent to learn the scheduling rule of reducing positioning load in high-risk regions and improving positioning performance in stable regions.

[0177] The resource consumption and motion smoothing term is defined as follows:

[0178]

[0179] in, and The penalty coefficient for smoothing action changes is used to constrain the variation amplitude of the positioning sampling period and the positioning modulation information between adjacent scheduling periods. This term is used to constrain the variation amplitude of the positioning sampling period and the positioning modulation information between adjacent scheduling periods, suppressing the strategy from severe oscillations under slight environmental disturbances, thereby improving the continuity and engineering executability of base station scheduling commands.

[0180] In the DDPG algorithm, the Actor network uses the state vector As input, output continuous scheduling actions The Critic network uses the aforementioned state vector With the continuous scheduling action The concatenated vector is used as input, and the output is a state-action value estimate. It is used to evaluate the long-term benefits of performing corresponding continuous scheduling actions in the current state and to provide the policy gradient update direction for the Actor network. Through the above state construction, continuous action output and reward function constraints, the base station can form a continuous scheduling strategy that takes into account energy security, positioning performance, multipath dead point avoidance and scheduling smoothness in a complex dynamic environment.

[0181] After the continuous scheduling action is completed, the base station receives the task execution information and energy status information returned by the passive tag terminal, and forms a closed-loop feedback vector by combining the multipath dead point probability model. Based on the closed-loop feedback vector, the instant reward is corrected so that the reward function reflects both the prior rationality before the action is executed and the actual result after the action is executed.

[0182] Assume the base station side is at the 1st Each scheduling cycle is based on the current state. Output continuous scheduling actions:

[0183]

[0184] in, For the next round of positioning sampling cycle, Modulate the amount of information for the next round of positioning;

[0185] The continuous scheduling action After the task is sent to the passive tag terminal for execution, the passive tag terminal completes the positioning task within the corresponding scheduling period and returns the task execution result and energy state change information to the base station. The base station combines the task execution result, energy state change information, and multipath dead point probability model to determine the location within the specified time period. Each scheduling cycle forms a closed-loop feedback vector:

[0186]

[0187] in, Indicates the target terminal In the The location risk probability is obtained in each scheduling cycle, and satisfies:

[0188]

[0189] The result of the location task execution satisfies:

[0190]

[0191] Its value is used to comprehensively characterize whether the positioning task was successfully completed, whether the positioning refresh rate met the requirements, and whether the effective modulation information was successfully carried out within the current cycle. The larger the value, the better the positioning task execution effect.

[0192] The voltage change before and after the action is performed is defined as:

[0193]

[0194] in, Before the action is executed Terminal operating voltage for each scheduling cycle The first time after the action is executed Terminal operating voltage for each scheduling cycle;

[0195] This indicates that after the action is executed, the target terminal is updated by the base station based on the multipath dead point probability model. In the The comprehensive multipath dead point exposure feedback value for each scheduling cycle is used to reflect the changes in the risk of the propagation environment after the terminal performs the current action.

[0196] Therefore, the original state Continuous scheduling actions With closed-loop feedback vector Together, they form a complete strategy closed loop, enabling the base station to interpret and correct the scheduling decision results from three dimensions: energy status, positioning task completion status, and propagation risk exposure.

[0197] The revised definition of instant reward is:

[0198]

[0199] in, Basic rewards; , , The weighting coefficients are corrected for non-negative feedback. , and These represent the task completion feedback item, the energy evolution feedback item, and the risk deviation penalty item, respectively.

[0200] The task completion feedback item is defined as:

[0201]

[0202] in, This is the scaling factor for the task completion feedback item; it is used when the positioning task in this cycle is successfully executed, the positioning refresh rate meets the standard, and the effective modulation information amount meets the requirements. A larger value is selected to increase the task completion feedback item; when the positioning task fails, the refresh rate does not meet the requirements, or the effective information is insufficient, reduce;

[0203] The energy evolution feedback term is defined as:

[0204]

[0205] in, The scaling factor for the energy evolution feedback term. and These are the upper and lower bounds of the terminal operating voltage, respectively, and satisfy the following conditions: When the terminal voltage recovers significantly or drops only slightly after the action is performed, it indicates that the current action is well matched with the environmental energy supply level, and the energy evolution feedback term is a positive benefit; when the action causes the terminal voltage to drop rapidly, the energy evolution feedback term decreases accordingly.

[0206] The risk deviation penalty item is defined as follows:

[0207]

[0208] in, Indicates that the base station side is in the first Each scheduling cycle is based on state. and continuous scheduling actions Prior prediction results of location risks after action execution; This indicates the actual location risk probability obtained from the information fed back by the terminal after the action is executed; The penalty coefficient for multipath dead point exposure risk;

[0209] The actual positioning risk after the action is performed Significantly deviates from the prior prediction results on the base station side Or the target terminal after the action is executed. Comprehensive multipath dead point exposure feedback value As the risk deviation penalty term increases, the base station-side agent will be more cautious in assessing the risks of energy outages and propagation dead spots during subsequent scheduling.

[0210] like Figure 5 The closed-loop feedback DDPG training convergence diagram shown: The base station side constructs an empirical sample reweighting mechanism based on the closed-loop feedback results and updates the deep reinforcement learning model online;

[0211] For the Given a set of interactive samples, its empirical sample is defined as:

[0212]

[0213] in, For the first The state vector corresponding to each empirical sample For the corresponding continuous scheduling actions, To adjust the rewards, The next state vector is reconstructed based on the action execution result and the closed-loop feedback vector;

[0214] Based on the closed-loop feedback results after the action is executed, the experience sample is... Assign importance weights to feedback :

[0215]

[0216] in, Indicates the first The probability of location risk corresponding to each empirical sample; This is a task failure indicator variable. It takes the value of 1 when the positioning task fails to execute within the corresponding scheduling period, the positioning refresh rate does not meet the requirements, or the amount of effective modulation information is insufficient; otherwise, it takes the value of 0. Indicates the first The comprehensive multipath dead point exposure feedback value is obtained after the action is executed for each experience sample. The normalized voltage change is defined as:

[0217]

[0218] in, Indicates the first The voltage change before and after the action is performed in an empirical sample. and These are the upper and lower bounds of the terminal operating voltage, respectively, and satisfy the following conditions: ; These are non-negative adjustment coefficients, used to control the contribution of positioning risk, mission failure, dead-point exposure feedback, and voltage changes to the importance of empirical samples, respectively.

[0219] The importance weight of the feedback Satisfy the following: When the terminal positioning risk is higher, the task failure is more obvious, the comprehensive multipath dead point exposure feedback value is higher, or the voltage change after the action is executed is more drastic, the feedback importance weight corresponding to the empirical sample is greater, so that the deep reinforcement learning model on the base station side pays more attention to the key state transition samples that lead to energy reduction, positioning interruption or increased propagation risk in the subsequent policy update process.

[0220] When updating the Critic network, the base station constructs a time-difference target value based on the target Critic network and the target Actor network:

[0221]

[0222] in, For the first The temporal difference target value corresponding to each empirical sample; To enhance the learning discount factor; For the target Critic network; For the target Actor network; and These are the target Critic network parameters and the target Actor network parameters, respectively.

[0223] After considering the feedback importance weights, the loss function of the Critic network is defined as the weighted mean square time difference error:

[0224]

[0225] in, This represents the number of empirical samples in a small batch. For the current Critic network; The current Critic network parameters are determined by the feedback importance weights. By introducing loss terms for each empirical sample, the Critic network can be given higher learning intensity when updating parameters for key samples with high positioning risk, high dead point exposure feedback values, task failure, or rapid voltage drop.

[0226] The Actor network updates its parameters guided by the value gradient provided by the Critic network, and its policy gradient is expressed as:

[0227]

[0228] in, For the current Actor network, These are the current Actor network parameters; This represents the policy objective function of the current Actor network. Represents the gradient operator. This indicates that the gradient of the action variable is calculated. This represents calculating the gradient of the current Actor network parameters. The Actor network is updated using the gradient of the above strategy, causing it to tend to output consecutive scheduling actions with higher long-term benefits in subsequent scheduling.

[0229] It should be noted that the above description of the embodiments is only for the purpose of helping to understand the method and core idea of ​​this application. For those skilled in the art, several improvements and modifications can be made to this application without departing from the principle of this application, and these improvements and modifications are also within the protection scope of the claims of this application.

Claims

1. A 6G passive IoT intelligent energy scheduling method for positioning tasks, characterized in that, The method includes the following steps: Step 1: At the passive tag terminal, collect the voltage time series and environmental parameter time series of the energy storage capacitor, construct a lightweight energy risk prediction model that integrates voltage evolution trend and environmental correction information, and obtain the basic undervoltage risk probability for the next scheduling cycle. Step 2: Based on the time relationship between the positioning sampling time and the prediction time of the basic undervoltage risk probability, perform time offset correction on the basic undervoltage risk probability, generate the positioning risk probability for the positioning task, and generate hierarchical reporting events based on the dual threshold comparison mechanism composed of high risk threshold and low risk threshold. Step 3: On the base station side, the received hierarchical reporting events, the target tag's location prediction information, and other tags' historical dead point observation data are fused to construct the multipath dead point probability, and further generate the target tag's comprehensive multipath dead point exposure probability. Step 4: Combine the current voltage of the passive tag terminal, environmental parameters, the location risk probability, the hierarchical reporting event, the comprehensive multipath dead point exposure probability, and the previous round of scheduling actions into a state vector, input it into a preset deep reinforcement learning model, and output continuous scheduling actions containing the next round of location sampling period and location modulation information, so as to realize continuous and adjustable control of the location task intensity. Step 5: The base station sends the continuous scheduling action to the passive tag terminal. The passive tag terminal performs the positioning task according to the continuous scheduling action and sends a feedback vector containing the task execution result and energy state change back to the base station. Step 6: The base station uses the feedback vector to construct a corrected reward, and performs online updates and policy evolution on the deep reinforcement learning model based on the experience replay mechanism to form a closed-loop scheduling optimization.

2. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 1, characterized in that, In step 1, the lightweight energy risk prediction model is constructed based on the voltage time series and environmental parameter time series of the passive tag terminal to obtain the basic undervoltage risk probability for the next scheduling cycle. The passive tag terminal is defined in the [missing information - likely a specific time series]. The energy storage capacitor terminal voltage collected at the end of each scheduling cycle is , For discrete scheduling cycle indexes, environmental parameter vectors for: ； in, , , These represent temperature, humidity, air pressure, or equivalent environmental characteristics, respectively; the first-order voltage difference is defined as follows: ； in, Indicates the scheduling period voltage, Indicates the previous scheduling cycle The voltage; The length of the sliding window is Then the past The average voltage change rate over each scheduling period is: ； in, Indicates the first The first-order voltage difference corresponding to each scheduling cycle The summation index within the sliding window; Based on the average voltage change rate, a basic voltage trend prediction term is constructed: ； in, This is a smoothing factor used to adjust the degree of confidence that the lightweight energy risk prediction model has in historical voltage change trends. The summation index within the sliding window; Define the change in environmental parameters between adjacent scheduling cycles as: ； Construct environmental correction terms based on the changes in the environmental parameters: ； in, This is an environmentally sensitive weight vector used to characterize the degree of influence of changes in different environmental parameters on the energy collection status of passive tag terminals; By combining the basic voltage trend prediction term and the environmental correction term, the predicted voltage for the next scheduling cycle is obtained: ； in, This is the scaling factor for the environmental correction term, used to adjust the intensity of the impact of external environmental changes on the predicted voltage; The local fluctuation intensity is constructed based on the discreteness of the voltage difference sequence within a short-time historical window: ； in, Used to characterize recent The degree of deviation of voltage change from the average trend within a scheduling cycle; Let the interrupt voltage threshold of the passive tag terminal be... To prevent the denominator from being zero, a very small positive number is introduced. The normalized voltage survival margin is defined as: ； in, Smoothing constant Finally, a mapping function is used to map the normalized voltage survival margin to the basic undervoltage risk probability: ； in, Indicates the first Given historical window information for each scheduling cycle, the next scheduling cycle... The passive tag terminal voltage is lower than the interrupt voltage threshold. The basic undervoltage risk probability; when the predicted voltage When the voltage is higher than the interruption voltage threshold and the normalized voltage survival margin is large, the probability of basic undervoltage risk approaches 0; when the predicted voltage is close to or lower than the interruption voltage threshold, or when the intensity of the local fluctuation increases, the probability of basic undervoltage risk increases.

3. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 2, characterized in that, In step 2, the basic undervoltage risk probability is... Time offset correction is performed based on the positioning sampling time to generate a positioning risk probability for the positioning task. And generate hierarchical reporting events based on a dual threshold mechanism; Define the current scheduling time as The next scheduling cycle is The basic undervoltage risk probability is then... The corresponding prediction time is: ； Let the next positioning sampling time be Define the prediction time. With the next positioning sampling time Time deviation between for: ； in, This is used to characterize the relative order between the prediction time corresponding to the basic undervoltage risk probability and the next location sampling time; When the predicted time is earlier than or equal to the next positioning sampling time, that is: ； Explain the probability of the basic undervoltage risk. This corresponds to an energy state judgment that is earlier than or equal to the location task execution time. In this case, as the interval between the prediction time and the location sampling time increases, the representativeness of the basic risk probability to the location sampling time decreases, requiring risk amplification compensation. Therefore, an intermediate location risk quantity is defined for this scenario. for: ； in, This is the time deviation risk amplification factor, used to control the risk compensation intensity when the predicted time is earlier than the positioning sampling time; when hour, ,when When it increases, the intermediate value of the positioning risk It increases with increasing time deviation; When the predicted time is later than the next positioning sampling time, that is: ； Explain the probability of the basic undervoltage risk. This corresponds to a future energy state judgment that occurs later than the location task execution time. In this case, as the prediction time lags further behind the location sampling time, the energy state at the time of the location task is further away from the predicted risk range, requiring risk attenuation correction. Therefore, an intermediate location risk quantity is defined for this scenario. for: ； in, This is the time deviation risk attenuation coefficient; when hour, ,when When it increases, the intermediate value of the positioning risk It decreases as the time deviation increases; Therefore, the probability of positioning risk for positioning tasks is defined as follows: ； in, This indicates the probability of the passive tag terminal performing the next positioning task at the current scheduling moment, representing the risk to its security. After obtaining the location risk probability Then, set a low-risk threshold. and high risk threshold ,satisfy: ； Based on the low-risk threshold and the high-risk threshold The terminal-side hierarchical reporting rules are defined as follows: when When the passive tag terminal is determined to be in a high-risk state, a high-risk reporting event is triggered. The high-risk reporting event is used to instruct the base station-side scheduling model to increase the positioning sampling period or reduce the positioning modulation information in subsequent scheduling, so as to reduce the energy consumption per unit time and maintain node survival. when When the passive tag terminal is determined to be in a low-risk state, a low-risk reporting event is triggered. The low-risk reporting event is used to instruct the base station-side scheduling model to reduce the positioning sampling period or increase the amount of positioning modulation information in subsequent scheduling, so as to improve the positioning refresh rate and the reporting information carrying capacity. when When the passive tag terminal is in a balanced operating state, it is determined that it does not trigger a high-risk reporting event or a low-risk reporting event, or it triggers a stable state reporting event, so as to maintain the continuity and stability of the current scheduling state.

4. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 3, characterized in that, In step 3, the base station side constructs the multipath dead point probability and generates the target label's comprehensive multipath dead point exposure probability; The area to be located is discretized into... Let there be a reference grid cell, and the grid set be denoted as: ； in, It remains a discrete scheduling cycle index; for any grid cell Define a binary random variable: ； in, Indicates the first Within a scheduling cycle, the grid When in a multipath dead point or deep energy decay state, there are propagation conditions that lead to a significant decrease in the energy collection of passive tags; This indicates that the grid is not in a multipath dead point or deep energy fading state; the base station side is at the... Each scheduling cycle for the grid The dead point probability estimate is defined as: ； This yields the spatial dead point probability distribution vector for the current scheduling period: ； Target terminal Due to the uncertainty of the future location, a location prediction probability distribution is introduced; let the target terminal... Fall into the grid in the next scheduling cycle The probability is: ； in, Indicates the target terminal The predicted position in the next scheduling cycle, satisfying: ； Based on the terminal-side lightweight risk prediction and anomaly reporting results, the target terminal is obtained. In the Location risk probability per scheduling cycle Abnormal signs and charging recovery time And normalize the charging recovery time to Define the target terminal In the The normalized energy anomaly intensity for each scheduling cycle is: ； in, For non-negative weighting coefficients, satisfying: ； It is a monotonically increasing mapping function used to map comprehensive risk to... interval; Used to reflect the target terminal Risk of energy interruption for the positioning task corresponding to the current scheduling cycle; Used to characterize sudden blockage, power drop, or undervoltage events; Used to characterize the target terminal Normalized result of the time required to recover from a low voltage state to a state capable of supporting positioning tasks; The target terminal is defined by combining the predicted location probability with the energy anomaly intensity. For the grid The amount of evidence for immediate dead points is: ； and target terminal The resulting immediate prior dead point probability is denoted as: ； Let the length be Within the sliding time window, historical observation samples from multiple terminals are displayed in the grid. Internally generated observation samples The number of samples that were judged as significant dead-point events was [number]. Then the grid The historical experience dead point probability is defined as: ； in, This is a smoothing constant used to avoid the denominator being zero; To characterize the spatial neighborhood correlation of multipath dead points, let the grid be denoted as . The neighborhood set is The risk propagation volume in its neighborhood is defined as: ； in, Neighborhood weight coefficients Represents a grid neighborhood set The neighborhood grid in For neighborhood grid index, Represents the neighborhood grid In the Historical experience dead point probability for each scheduling cycle; And satisfy: ； The neighborhood weight coefficient is set using Gaussian decay or inverse distance weighting based on the distance between the grid centers, so that the closer the neighboring grid is, the greater its influence on the current grid. After obtaining immediate evidence from the target terminal, historical experience of the group, and diffusion information in the neighborhood space, a weighted fusion method is used to obtain the grid. Comprehensive dead point probability estimation: ； in, To merge weights, and satisfy: ； To balance the long-term stability of spatial dead-point distribution with the ability to respond to short-term sudden disturbances, the base station side, according to the first The comprehensive dead-point probability estimate obtained within a scheduling cycle is used to recursively update the grid dead-point probability for the next scheduling cycle: ； in, To update the coefficients, the update results are truncated to ensure probabilistic meaning. ； Based on the above grid dead point probability estimation, the target terminal is defined. The overall multipath dead point exposure probability in the next scheduling cycle is: ； in, Indicates the target terminal The expected overlap between the future location probability distribution and the spatial dead-point probability map; when the target terminal When there is a high probability of entering a high-risk grid in the future, Increase; when the target terminal When the future is mainly in areas with relatively stable transmission conditions, Decrease.

5. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 4, characterized in that, The significant dead-point events are used to statistically analyze the probability of dead points based on historical experience. The specific determination process is as follows: set up Let be the set of terminals used for historical dead point statistics; for the The terminal in the first Within a scheduling cycle, located in the grid For a single observation sample, the indicator variable for the dead-point salient event is defined as: ； in, For the first The terminal in the first The probability of location risk per scheduling cycle. For the first The normalized result of the charging recovery time required for a terminal to recover from a low voltage state to a state capable of supporting positioning tasks. This is an exception flag variable. To locate the risk threshold, This is the charging delay threshold; when When this occurs, it indicates that the terminal faces a high risk of energy interruption for the positioning task within the current grid; when When this occurs, it indicates that the time required for the terminal to recover from a low-voltage state to a state capable of performing positioning tasks within the current grid is too long; when When a terminal experiences a sudden occlusion, power drop, or undervoltage event within the current grid, the observed sample is identified as a significant dead point event if any of the above conditions are met. In length Within the sliding time window, the grid Number of significant events at the mid-dead point The following results were obtained from the statistics of the indicator variables: ； in, For indicator functions, Indicates the first The terminal in the first Position within a scheduling cycle; Correspondingly, grid Total number of observed samples for: ； This leads to the probability of dead points based on historical experience: ； This determination method distinguishes between transient voltage drops caused by instantaneous fluctuations and persistent energy supply shortages caused by spatial propagation anomalies by jointly judging the location risk probability, normalized charging recovery time, and abnormal marker variables. This provides a group historical observation basis for updating the multipath dead point probability on the base station side.

6. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 5, characterized in that, In step 4, the base station side uses a deep deterministic policy gradient algorithm to construct a deep reinforcement learning scheduling model, which maps the energy state, environmental state, positioning risk state, and multipath dead point exposure risk of the passive tag terminal into continuous scheduling actions. The deep reinforcement learning scheduling model includes an Actor network and a Critic network. Define the base station-side reinforcement learning agent in the first... The state vector for each scheduling cycle is: ； in, Indicates that the passive tag terminal is in the first The current operating voltage for each scheduling cycle, and compared with the voltage Correspondingly, it is used to reflect the current remaining energy level of the terminal; Represents the current multidimensional environment parameter vector, and is related to the environment parameter vector. Correspondingly, it is used to describe the external environmental conditions that affect energy harvesting efficiency and propagation conditions; Indicates the target terminal In the The current location risk probability for each scheduling cycle, satisfying: ； in, For target terminal The probability of positioning risk; This indicates the current triggered tiered reporting type, including high-risk reporting, low-risk reporting, stable status reporting, or no reporting. Indicates that the base station side is at the The target terminal's comprehensive multipath dead point exposure prediction value for the next scheduling cycle can be obtained when making decisions in each scheduling cycle, and satisfies: ； in, For target terminal The comprehensive multipath dead point exposure probability in the next scheduling cycle is used to characterize the possibility that the terminal will enter a high-risk deep fading area during the next round of positioning task execution. Indicates the previous round of scheduling actions: ； This indicates the previous positioning sampling period. This represents the amount of positioning modulation information from the previous round, used to explicitly characterize the impact of motion inertia on subsequent energy evolution; Define the base station-side agent in the first... The continuous scheduling actions output in each scheduling cycle are: ； in, Indicates the next round of positioning sampling period. This indicates the amount of positioning modulation information for the next round; the range of values ​​for the action is: ； in, This indicates the minimum allowed positioning sampling period. This indicates the maximum waiting time set in the project; when A smaller value indicates that the terminal performs positioning tasks at a higher frequency, thereby improving the positioning refresh rate and time resolution; when the value is smaller... As the value gradually increases, it indicates that the terminal extends the interval between two adjacent positioning tasks, thereby reducing the energy consumption per unit time. and These represent the minimum and maximum values ​​of the location modulation information, respectively; To ensure that the agent balances terminal energy safety, localization performance, multipath dead point avoidance, and action smoothness during the learning process, the following weighted instantaneous reward function is constructed: ； in, , , , These are non-negative reward weighting coefficients, used to adjust the energy safety reward items. Positioning performance bonus items Multipath dead point penalty item Resource consumption and motion smoothing items Its importance in the overall reward; The energy safety reward item is defined as follows: ； in, The current location risk probability; when the terminal's energy status is good and the risk of short-term interruption is low, The value is relatively small. The corresponding increase; when the terminal approaches the undervoltage boundary or there is a significant risk of power outage in the next scheduling cycle. As the value increases, The corresponding reduction guides the intelligent agent to prioritize avoiding dangerous actions that could cause terminal malfunction. The positioning performance bonus item is defined as follows: ； in, and These are internal weighting coefficients used to adjust the relative contributions of the positioning sampling period and the amount of positioning modulation information to the gains in positioning performance. This is a smoothing constant used to avoid the denominator being zero; this reward term enables the agent to obtain higher positioning performance gains when outputting shorter positioning sampling periods and higher positioning modulation information, thereby avoiding the model from adopting an overly conservative low-load strategy in the long term in pursuit of absolute energy safety. The multipath dead point penalty term is defined as follows: ； in, and The action load penalty coefficient under multipath dead point risk is used to adjust the weights of the positioning sampling period and the positioning modulation information in the dead point risk penalty. This term is used to characterize the coupling relationship between multipath dead point exposure risk and the aggressiveness of scheduling actions: when the terminal is in or about to enter a region with a high multipath dead point exposure probability, if a shorter positioning sampling period and a higher positioning modulation information are still selected, the multipath dead point penalty term increases; when the terminal is in a low dead point risk region, the penalty corresponding to the same scheduling action decreases, thereby enabling the agent to learn the scheduling rule of reducing positioning load in high-risk regions and improving positioning performance in stable regions. The resource consumption and motion smoothing term is defined as follows: ； in, and The penalty coefficient for smoothing action changes is used to constrain the variation amplitude of the positioning sampling period and the positioning modulation information between adjacent scheduling periods. This term is used to constrain the variation amplitude of the positioning sampling period and the positioning modulation information between adjacent scheduling periods, suppressing the strategy from causing violent oscillations under slight environmental disturbances, thereby improving the continuity and engineering feasibility of base station scheduling commands.

7. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 6, characterized in that, In steps 5 and 6, after the continuous scheduling action is completed, the base station receives the task execution information and energy status information returned by the passive tag terminal, and forms a closed-loop feedback vector by combining the multipath dead point probability model. Based on the closed-loop feedback vector, the instant reward is corrected so that the reward function reflects both the prior rationality before the action is executed and the actual result after the action is executed. Assume the base station side is at the 1st Each scheduling cycle is based on the current state. Output continuous scheduling actions: ； in, For the next round of positioning sampling cycle, Modulate the amount of information for the next round of positioning; The continuous scheduling action After the task is sent to the passive tag terminal for execution, the passive tag terminal completes the positioning task within the corresponding scheduling period and returns the task execution result and energy state change information to the base station. The base station combines the task execution result, energy state change information, and multipath dead point probability model to determine the location within the specified time period. Each scheduling cycle forms a closed-loop feedback vector: ； in, Indicates the target terminal In the The location risk probability is obtained in each scheduling cycle, and satisfies: ； The result of the location task execution satisfies: ； Its value is used to comprehensively characterize whether the positioning task was successfully completed, whether the positioning refresh rate met the requirements, and whether the effective modulation information was successfully carried out within the current cycle. The larger the value, the better the positioning task execution effect. The voltage change before and after the action is performed is defined as: ； in, Before the action is executed Terminal operating voltage for each scheduling cycle The first time after the action is executed Terminal operating voltage for each scheduling cycle; This indicates that after the action is executed, the target terminal is updated by the base station based on the multipath dead point probability model. In the The comprehensive multipath dead point exposure feedback value for each scheduling cycle is used to reflect the changes in the risk of the propagation environment after the terminal performs the current action. Therefore, the original state Continuous scheduling actions With closed-loop feedback vector Together, they form a complete strategy closed loop, enabling the base station to interpret and correct the scheduling decision results from three dimensions: energy status, positioning task completion status, and propagation risk exposure. The revised definition of instant reward is: ； in, Basic rewards; , , The weighting coefficients are corrected for non-negative feedback. , and These represent the task completion feedback item, the energy evolution feedback item, and the risk deviation penalty item, respectively. The task completion feedback item is defined as follows: ； in, This is the scaling factor for the task completion feedback item; it is used when the positioning task in this cycle is successfully executed, the positioning refresh rate meets the standard, and the effective modulation information amount meets the requirements. Take a larger value to increase the task completion feedback item; when the positioning task fails, the refresh rate does not meet the requirements, or the effective information is insufficient, reduce; The energy evolution feedback term is defined as: ； in, The scaling factor for the energy evolution feedback term. and These are the upper and lower bounds of the terminal operating voltage, respectively, and satisfy the following conditions: When the terminal voltage recovers significantly or drops only slightly after the action is performed, it indicates that the current action is well matched with the environmental energy supply level, and the energy evolution feedback term is a positive benefit; when the action causes the terminal voltage to drop rapidly, the energy evolution feedback term decreases accordingly. The risk deviation penalty item is defined as follows: ； in, Indicates that the base station side is at the Each scheduling cycle is based on state. and continuous scheduling actions Prior prediction results of location risks after action execution; This indicates the actual location risk probability obtained from the information fed back by the terminal after the action is executed; The penalty coefficient for multipath dead point exposure risk; The actual positioning risk after the action is performed Significantly deviates from the prior prediction results on the base station side Or the target terminal after the action is executed. Comprehensive multipath dead point exposure feedback value As the risk deviation penalty term increases, the base station-side agent will be more cautious in assessing the risks of energy outages and propagation dead spots during subsequent scheduling.

8. The 6G passive IoT intelligent energy dispatching method for positioning tasks according to claim 7, characterized in that, In step 6, the base station constructs an empirical sample reweighting mechanism based on the closed-loop feedback results and updates the deep reinforcement learning model online. For the Given a set of interactive samples, its empirical sample is defined as follows: ； in, For the first The state vector corresponding to each empirical sample For the corresponding continuous scheduling actions, To adjust the rewards, The next state vector is reconstructed based on the action execution result and the closed-loop feedback vector; Based on the closed-loop feedback results after the action is executed, the experience sample is... Assign importance weights to feedback : ； in, Indicates the first The probability of location risk corresponding to each empirical sample. This is a task failure indicator variable. It takes the value of 1 when the positioning task fails to execute within the corresponding scheduling period, the positioning refresh rate does not meet the requirements, or the amount of effective modulation information is insufficient; otherwise, it takes the value of 0. Indicates the first The comprehensive multipath dead point exposure feedback value after the action is executed for each experience sample; The normalized voltage change is defined as: ； in, Indicates the first The voltage change before and after the action is performed in an empirical sample. and These are the upper and lower bounds of the terminal operating voltage, respectively, and satisfy the following conditions: ; These are non-negative adjustment coefficients, used to control the contribution of positioning risk, mission failure, dead-point exposure feedback, and voltage changes to the importance of empirical samples, respectively. The importance weight of the feedback Satisfy the following: When the terminal positioning risk is higher, the task failure is more obvious, the comprehensive multipath dead point exposure feedback value is higher, or the voltage change after the action is executed is more drastic, the feedback importance weight corresponding to the empirical sample is greater, so that the deep reinforcement learning model on the base station side pays more attention to the key state transition samples that lead to energy reduction, positioning interruption or increased propagation risk in the subsequent policy update process. When updating the Critic network, the base station constructs a time-difference target value based on the target Critic network and the target Actor network: ； in, For the first The temporal difference target value corresponding to each empirical sample; To enhance the learning discount factor; For the target Critic network; For the target Actor network; and These are the target Critic network parameters and the target Actor network parameters, respectively. After considering the feedback importance weights, the loss function of the Critic network is defined as the weighted mean square time difference error: ； in, This represents the number of empirical samples in a small batch. For the current Critic network; The current Critic network parameters are determined by the feedback importance weights. By introducing loss terms for each empirical sample, the Critic network can be given higher learning intensity when updating parameters for key samples with high positioning risk, high dead point exposure feedback value, task failure or rapid voltage drop. The Actor network updates its parameters guided by the value gradient provided by the Critic network, and its policy gradient is expressed as: ； in, For the current Actor network, These are the current Actor network parameters; This represents the policy objective function of the current Actor network. Represents the gradient operator, This indicates that the gradient of the action variable is calculated. This indicates that the gradient of the current Actor network parameters is calculated; the Actor network is updated through the gradient of the above strategy so that it tends to output continuous scheduling actions with higher long-term benefits in subsequent scheduling.

9. A 6G passive IoT intelligent energy dispatching system for positioning tasks, characterized in that, Includes the following modules: Passive tag terminal energy risk prediction and reporting module: Deployed in the passive tag terminal, it is used to collect the voltage time series and environmental parameter time series of the energy storage capacitor, build a lightweight energy risk prediction model that integrates voltage evolution trend and environmental correction information, and obtain the basic undervoltage risk probability for the next scheduling cycle; Based on the time relationship between the location sampling time and the prediction time of the basic undervoltage risk probability, the basic undervoltage risk probability is time-biased and corrected to generate a location risk probability for the location task. Based on the preset high-risk threshold and low-risk threshold, a hierarchical reporting event is triggered. Base station-side multipath dead point probability construction module: Deployed at the base station, it is used to receive the hierarchical reporting events, fuse the target label's location prediction information and other labels' historical dead point observation data, discretize the target area into multiple grid cells, calculate the historical experience dead point probability, immediate dead point evidence amount and neighborhood diffusion risk of each grid cell, obtain the multipath dead point probability of each grid cell through weighted fusion and recursive update, and generate the comprehensive multipath dead point exposure probability of the target label; The base station-side deep reinforcement learning scheduling decision module is deployed in the base station and is used to combine the current operating voltage, environmental parameters, positioning risk probability, hierarchical reporting events, comprehensive multipath dead point exposure probability, and the previous round of scheduling actions of the passive tag terminal into a state vector, input it into a preset deep reinforcement learning model, output continuous scheduling actions containing the next round of positioning sampling period and positioning modulation information, and send them to the passive tag terminal for execution. Closed-loop feedback and policy evolution module: Deployed at the base station, it is used to receive the closed-loop feedback vector returned by the passive tag terminal after performing the continuous scheduling action. The closed-loop feedback vector includes the location risk probability, task execution result, voltage change, and comprehensive multipath dead point exposure feedback value. Based on the closed-loop feedback vector, a modified reward is constructed, and the deep reinforcement learning model is updated online through empirical sample reweighting, weighted Critic loss function, and Actor network to perform policy evolution.

Images