A power consumption information collection and metering box with power outage monitoring and wireless reporting.

By identifying the capacitance and internal resistance of supercapacitors online and dynamically adjusting the power outage confirmation and message transmission strategies, the problem of channel congestion under the condition of mass power outages in low-voltage distribution areas was solved, and reliable reporting and rapid response of power outage event messages were achieved.

CN123122005APending Publication Date: 2026-11-17HANGZHOU KUANCHEN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611601987.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

When a simultaneous power outage occurs at the substation or line level in a low-voltage distribution area, the existing electricity information collection and metering boxes cannot effectively coordinate and adjust the power outage confirmation waiting time and the sending and retransmission strategy of power outage event messages, resulting in channel congestion and report loss. Especially when there are no surviving upper-layer nodes available for relay, the existing processing sequence is prone to failure due to premature energy depletion or channel concurrency collisions.

Method used

By identifying the capacitance and equivalent series resistance of supercapacitors online, the residual energy budget is dynamically estimated. Combined with bus voltage detection and wireless communication channel occupancy, the power outage confirmation waiting time and message retransmission strategy are adjusted to maximize the success rate of power outage event message delivery.

Benefits of technology

In the event of a widespread power outage, it effectively reduces the probability of channel collisions, ensures the successful delivery of power outage event messages, and guarantees the power supply side's accurate reconstruction of the outage area and timely response to emergency repairs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122005A_ABST
    Figure CN123122005A_ABST
Patent Text Reader

Abstract

This application discloses a power consumption information collection and metering box with power outage monitoring and wireless reporting, including a metering box body and a collection unit, a wireless communication unit, a supercapacitor, and a bus voltage detection unit located therein; the supercapacitor supplies power to the collection unit and the wireless communication unit after an AC interruption; after the collection unit detects a power outage signal, it identifies the capacitance and equivalent series internal resistance of the supercapacitor online and converts them into a residual energy budget. Based on the power outage characteristics and channel occupancy, it infers the intensity of concurrent contention and uses the residual energy budget as the sole shared budget for power outage confirmation waiting and power outage event message transmission and retransmission to mutually compete. The decision network selects the action with the highest expected successful delivery probability to control transmission and retransmission, avoiding premature energy depletion or channel collision leading to report loss, and improving the reporting success rate under mass power outages.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of electricity information collection and remote meter reading communication technology, specifically to an electricity information collection and metering box with power outage monitoring and wireless reporting capabilities. Background Technology

[0002] The electricity information collection and metering box is installed on the user side of the low-voltage distribution area. In addition to periodically collecting and reporting metering readings, it also undertakes the function of monitoring and actively reporting power outage events, which is of great significance for power supply emergency repairs. This is known in the industry as last-minute reporting. Its typical implementation is as follows: The metering box has a built-in supercapacitor as a backup power source after the AC bus power supply is interrupted. When the AC power supply to the bus is interrupted, the collection unit relies on the non-renewable residual energy stored in the supercapacitor to maintain operation for a period of time. Before the residual energy is exhausted, it sends a frame recording the power outage event of the box to the concentrator or base station via a wireless channel. This allows the power supply side to locate the power outage area, restore the distribution area topology, and dispatch emergency repair work orders without waiting for the user to report the fault.

[0003] To distinguish between a genuine power outage and a short-term power loss during voltage dips or reclosing, existing data acquisition terminals generally wait for a preset power outage confirmation period after detecting a bus power failure. They only send a power outage event message after confirming that AC power has not been restored, thus suppressing false alarms caused by momentary power loss. The transmission phase typically uses fixed or random backoff methods for retransmission. Some solutions also rely on still-energized upper-layer relay nodes in the network to aggregate and forward power outage event messages from multiple terminals, reducing the concurrent transmission volume on the channel. In existing solutions, the power outage confirmation waiting and transmission / retransmission phases are set with independent fixed parameters, all based on a known and fixed energy budget for the backup time provided by the supercapacitor. Even if some solutions mention adjustments to the reporting based on stored energy, they do not address the residual energy drift and uncertainty caused by the capacity decrease and equivalent series resistance increase of the supercapacitor due to aging, temperature, and charging / discharging history. To address channel congestion caused by a large number of terminals reporting simultaneously, existing congestion management methods either rely on intermediate nodes that are still powered to aggregate and merge power outage event messages, or on terminals knowing their own network hierarchical structure in advance to back off transmission at different times according to the layer, or on collecting neighbor advertisements across channels for collaborative judgment. All of these require the availability of auxiliary information in the network.

[0004] Electricity information collection metering boxes also face a situation that is not prominent in single-point power outage scenarios in residential applications, but is quite special in low-voltage distribution areas: simultaneous power outages at the distribution area or line level. When a transformer outgoing line trips, a main line fault occurs, or a planned area-wide power outage occurs, dozens or even hundreds of metering boxes in the same distribution area lose power at the same physical instant. Their internal supercapacitors begin to discharge almost simultaneously and enter the power outage reporting process simultaneously. At this time, the upper-level nodes in the network that could originally act as aggregation relays also lose power at the same instant, leaving no one available to forward the message. All the metering boxes that are about to run out of energy compete for the same wireless channel to report to the concentrator at almost the same time window, resulting in instantaneous channel congestion and message collisions. Meanwhile, the remaining energy budget available for each metering box to confirm and send power outages is not truly fixed. The capacity and equivalent series internal resistance of the supercapacitors drift with their service life, ambient temperature, and charge / discharge history. As they age, their capacity decreases and their internal resistance increases, making the actual working time that different boxes can support at different times different and unpredictable. Furthermore, each metering box nearing energy depletion cannot sense how many neighboring units in the same area are reporting at the same time as it at the moment of power failure, nor can it exchange this information through the paralyzed network in time. The four constraints of non-renewable and unknown drift of residual energy budget, lack of live relays for aggregation, large-scale concurrent channel contention, and unobservable contention scale superimposed on each isolated metering box at the same physical instant make the processing order of first waiting for confirmation and then sending with fixed or random backoff, which is feasible in normal scenarios, tend to fail: the power outage confirmation waiting time and the timing and number of retransmissions of the power outage event message share the same dwindling and unknown residual energy budget. The longer the wait, the better it is to confirm the real power outage, but the more likely it is to run out of energy before sending and lose the report completely. If the backoff parameter is not adjusted with the actual concurrent contention scale, it is difficult to effectively reduce the collision probability under the condition of simultaneous power outages, thus causing the power supply side to lose the reconstruction of the power outage range and transformer topology. Therefore, in the case of simultaneous power outages with no surviving upper-level nodes available for relaying, and with the remaining energy budget unknown and irreversibly decreasing over time, how to coordinate the on-site adjustment of the power outage confirmation waiting time and the sending and retransmission strategy of wireless power outage event messages by the metering box remains an unsolved problem in the power outage monitoring and wireless reporting of existing electricity information collection metering boxes. Summary of the Invention

[0005] To address the issue of simultaneous, district-level or line-level power outages occurring in low-voltage distribution areas, where no surviving upper-level nodes are available for relay aggregation, a single metering box must report the power outage event on-site under the constraints of an unknown remaining supercapacitor energy budget, drift due to aging and temperature, and irreversible depletion over time. The existing processing sequence of first waiting for confirmation and then backing up transmission becomes ineffective in this situation, and is prone to report loss due to premature energy depletion or channel concurrency collisions. This invention provides a power consumption information collection metering box with power outage monitoring and wireless reporting capabilities. It includes a metering box body, and within the metering box body, a data acquisition unit, a wireless communication unit, a supercapacitor, and a bus voltage detection unit. The bus voltage detection unit is used to detect power outages on the AC bus of the metering box body. The signal indicates that the supercapacitor supplies power to the acquisition unit and the wireless communication unit after an AC power outage; the acquisition unit is configured to, upon detecting the power failure signal, identify the current capacitance and equivalent series internal resistance of the supercapacitor online and convert them into a residual energy budget, infer the concurrent contention intensity of the unit based on power failure characteristics and channel occupancy observations, use the residual energy budget as the sole shared budget, deduct the power failure confirmation waiting and the transmission and retransmission of power failure event messages from the residual energy budget according to their respective energy consumption to cause them to compete with each other, and output the expected successful delivery probability of each selectable action until the residual energy budget is exhausted through a decision network, select the action that maximizes the expected successful delivery probability to control the wireless communication unit to perform transmission and retransmission.

[0006] Specifically, this invention incorporates four constraints superimposed on each isolated metering box during a simultaneous power outage: non-renewable and unknown drift of the residual energy budget, no surviving relays available for aggregation, large-scale concurrent channel contention, and unobservable scale of concurrent contention. These constraints are integrated into a single decision-making framework. It replaces the fixed assumption of backup energy with online identification, allowing the residual energy budget to be actively estimated based on the actual state of the box's aging and temperature. It does not treat the power outage confirmation waiting time and the transmission and retransmission of power outage event messages as two separate, fixed, and unrelated processes, but rather allows them to compete and coordinate under the same dwindling residual energy budget. Even without exchanging information with neighbors or having prior knowledge of network hierarchy, it infers the intensity of concurrent contention on-site based solely on the box's power outage characteristics and local channel observations, and adjusts the dispersion of transmission timing accordingly. Thus, the contradiction of waiting longer to confirm a real power outage but also increasing the likelihood of energy depletion before transmission and loss of reporting is resolved by maximizing the probability of successful delivery of the box's power outage event message before energy depletion.

[0007] Furthermore, the acquisition unit is configured to acquire the terminal voltage sequence of the supercapacitor and the load current sequence currently sampled by the acquisition unit and the wireless communication unit at a fixed sampling period after detecting the power failure signal. Based on the fact that the terminal voltage is equal to the voltage across the capacitor minus the voltage drop across the load current on the equivalent series resistance, and that the voltage across the capacitor decreases linearly with the discharged charge, a discharge state model is established with capacitance and equivalent series resistance as the parameters to be identified. Then, using the recursive least squares method or extended Kalman filter, with the terminal voltage sequence and the load current sequence as the observations, the current capacitance and equivalent series resistance of the supercapacitor are identified online, so that the capacitance decrease and internal resistance increase caused by aging and temperature are reflected in the identification results. Specifically, the fixed sampling period can be from 1 millisecond to 10 milliseconds. At the moment of power failure, the supercapacitor bears the entire load current, and its terminal voltage exhibits an observable decreasing trend with discharge. The acquisition unit expresses the relationship that the terminal voltage equals the voltage across the capacitor minus the product of the load current and the equivalent series resistance, and that the voltage across the capacitor decreases linearly with the cumulative discharged charge according to the reciprocal of the capacitance, as a linear observation equation with the reciprocal of the capacitance and the equivalent series resistance as unknown coefficients. This equation is then recursively updated by substituting the measured terminal voltage and load current frame by frame, using recursive least squares. The method uses a forgetting factor to weight and attenuate old observations to track slowly changing parameters. When using extended Kalman filtering, the capacitance and equivalent series internal resistance are incorporated into the state vector and terminal voltage observations for recursion. Both converge to the actual current capacitance and equivalent series internal resistance of the supercapacitor in this box within a short time window of tens of milliseconds after power failure. This accurately reflects the capacitance decay and internal resistance increase of supercapacitors with the same nominal capacity due to different service years, ambient temperatures, and charge and discharge histories in the identification results, avoiding the overestimation of usable energy by using the factory nominal value.

[0008] Furthermore, the acquisition unit is configured to, based on the identified capacitance, current terminal voltage, and cutoff operating voltage, perform... Calculate the current residual energy of the supercapacitor, where To identify the obtained capacitance, This is the current terminal voltage. The cutoff operating voltage is used as the cutoff voltage, and the power consumption corresponding to the current load current is combined to calculate the estimated usable operating time before the residual energy is exhausted, along with its estimation uncertainty. The acquisition unit also maintains an aging count that accumulates with the number of charge / discharge cycles and operating temperature, providing initial values ​​and priors for the identification of the capacitance and equivalent series internal resistance. Specifically, the cutoff operating voltage is taken as the minimum bus voltage required for the acquisition unit and wireless communication unit to maintain normal operation. The residual energy is obtained by multiplying the difference between the square of the current terminal voltage and the square of the cutoff voltage by half of the identified capacitance. The acquisition unit estimates the instantaneous power consumption using the product of the current load current and the terminal voltage, divides the residual energy by the instantaneous power consumption, and takes into account the power consumption change caused by the drop in terminal voltage during discharge, converting it into a budgeted usable operating time before the residual energy is exhausted. At the same time, the estimation uncertainty of this budget is given based on the covariance of the identification parameters, providing a basis for subsequent decision-making to reserve energy margin. The aging count is accumulated with corresponding weights during each supercapacitor charge-discharge cycle and high-temperature operation. The acquisition unit uses the aging count to look up a table or obtain the prior interval of capacitance and equivalent series internal resistance according to an empirical function. This is used as the initial value and prior distribution for recursive identification at the moment of power failure, shortening the time required for identification convergence and enabling identification to be fast and stable in the early stage of power failure when energy is most precious.

[0009] Furthermore, the acquisition unit is configured to read the power outage characteristics recorded by the bus voltage detection unit within a short time window before and after the power outage. The power outage characteristics include whether the three phases drop simultaneously, the rate of voltage drop and the depth of voltage drop on the bus. Simultaneous and rapid drops of the three phases are determined as simultaneous power outages at the line level or transformer substation level, while single-phase or slow drops are determined as local power outages. Based on this, a priori estimate of the concurrent contention intensity of the transformer substation is given. Specifically, the short time window is several milliseconds to tens of milliseconds before and after the power outage. Within this window, the acquisition unit retains the sampling of the three-phase phase voltage by the bus voltage detection unit and compares whether the three-phase voltage drops are synchronous, whether the drop edge is steep, and the final drop depth. When the three-phase voltage drops synchronously and rapidly to below the cutoff threshold at the same time, it is determined that the power outage is caused by a transformer outgoing line trip, main line fault, or planned full-area power outage, resulting in a simultaneous power outage at the line level or substation level. Such events mean that a large number of metering boxes in the same substation enter the power outage reporting process almost simultaneously, so a higher prior for concurrent contention intensity is given. When only a single phase drops or three phases drop slowly, it is determined to be a local power outage, and a lower prior for concurrent contention intensity is given to provide the initial belief for subsequent Bayesian fusion.

[0010] Furthermore, the acquisition unit is configured to control the wireless communication unit to perform short-time carrier sensing before transmission, measure the channel occupancy rate and received signal strength of the target channel, and count the preamble of neighbor power outage event messages superimposed on the channel. Based on the monotonic variation of channel occupancy rate and collision statistics with the number of concurrent devices when multiple devices transmit concurrently on the radio frequency channel, an observation likelihood with the number of concurrent contending devices as the latent variable is established. The prior estimate, the channel occupancy rate and the preamble count, as well as the historical power outage periods stored in this box and the historical group power outage patterns of the corresponding transformer area are fused in a Bayesian manner to infer the posterior estimate of the concurrent contention intensity and its uncertainty. Specifically, the short-time carrier sensing (SMS) involves channel monitoring for several milliseconds before formal transmission. The proportion of time the channel is occupied within the monitoring window is counted as the channel occupancy rate. The received signal strength is measured, and the characteristic preambles carried in the power outage event messages sent by other metering boxes on the channel are matched and counted. The more preamble counts and the higher the occupancy rate, the more neighbors are reporting simultaneously. The acquisition unit pre-calibrates the monotonically increasing correspondence between the number of concurrent stations, the channel occupancy rate, and the preamble count with the number of concurrent stations. Based on this, an observation likelihood function is established with the number of concurrent stations vying for power as the latent variable and the channel occupancy rate and the preamble count as the observation quantities. The prior estimate given by the power outage feature classification is then used as the prior distribution. The historical power outage period distribution recorded in the non-volatile storage of this box and the historical group power outage patterns of the corresponding transformer area are used as additional priors. These are multiplied and normalized with the channel observation likelihood in a Bayesian manner to obtain the posterior estimate of the current concurrent contention intensity and its uncertainty. This allows the prior and real-time channel observation to complement each other, providing a robust estimate even when the observation window is extremely short and information is limited.

[0011] Furthermore, the acquisition unit is configured to read the three-phase phase voltage drop waveform acquired by the bus voltage detection unit, and based on the synchronicity of the drop time and the correlation of the drop slope of the metering boxes on the same feeder in the same tripping event, invert the scale and boundary of the fault subgroup to which the box belongs, so as to distinguish between the strong concurrent subgroups that trip on the same feeder and the weak concurrent subgroups that only belong to the adjacent domain, and use the scale of the fault subgroup to replace the overall scale of the distribution area to drive the inference of the concurrent contention intensity. Specifically, for metering boxes on the same feeder that are powered down by the same tripping event, the three-phase voltage drop waveforms are highly synchronized in terms of drop time and highly correlated in terms of drop slope. However, for metering boxes that do not belong to the same tripping range, the drop time and slope are discrete. The acquisition unit compares the drop waveform acquired by this box with the waveform characteristics of its neighboring area. Based on the synchronization of drop time and the correlation of drop slope, it infers the scale and boundary of the fault subgroup to which this box actually belongs. This narrows down the range of devices that are actually competing for the channel with this box to the strong concurrent subgroup that trips on the same feeder. Devices that belong to the same transformer area but are powered down by different protection actions and at different times are classified as weak concurrent subgroups and eliminated. The scale of this fault subgroup is used to replace the scale of the entire transformer area to drive the inference of concurrent competition intensity. This avoids overestimating the degree of concurrency by estimating the scale of the entire transformer area in a general way, and makes the adjustment of backoff and transmission timing more consistent with the actual competition scale of this box.

[0012] Furthermore, the acquisition unit is configured to construct a decision state vector from the residual energy budget and its estimation uncertainty, the currently consumed power outage confirmation waiting time, the AC power supply recovery flag given by the bus voltage detection unit, the posterior estimate and uncertainty of the concurrent contention intensity, the current channel occupancy rate, and the current number of retransmissions. This decision state vector is then input into the decision network with a fixed millisecond-level decision step size. The decision network is trained offline using historical or simulated power outage discharge processes, channel collision processes, and delivery results as training samples, aiming to maximize the successful delivery probability under residual energy constraints. Its inputs and outputs are physical quantities of residual energy, duration, channel occupancy rate, and concurrent contention intensity. Specifically, each component of the decision state vector is a technical quantity with definite physical meaning. The acquisition unit periodically refreshes this vector and inputs it into the decision network throughout the entire power outage discharge process with a fixed millisecond-level decision step size, allowing the decision to be dynamically updated as residual energy diminishes, waiting time accumulates, and channel occupancy and concurrent estimations change. The training of the decision network is completed offline on a computing platform outside the metering box. Part of the training samples are taken from historically reported events from the metering box, including power outage discharge processes, channel collision processes, and final delivery results. Another part is generated in batches from simulation models of power outage discharge and RF channel collision with different concurrent numbers, initial residual energy levels, and channel conditions, and augmented with data. Samples are divided into training, validation, and test sets, with proportions of 70%, 15%, and 15%, respectively. The decision network employs a multi-layer fully connected neural network or temporal network structure, built using a deep learning framework and trained on a graphics processing unit. Reinforcement learning is used to treat each successful delivery as a positive reward and any undelivered delivery before energy depletion as a zero or negative reward. The optimization objective is to maximize the expected successful delivery probability until residual energy is exhausted. An adaptive moment estimation optimizer is used, with a learning rate of [missing information]. to In conjunction with learning rate decay, batch size ranges from 64 to 256, and the number of training epochs ranges from hundreds to thousands. Weight regularization and an early stopping mechanism based on the successful delivery rate of the validation set are used to suppress overfitting. The evaluation metric is the successful delivery probability under residual energy constraints, which is compared with the baseline strategy of fixed waiting plus fixed backoff. The trained network parameters are quantized and fixed in the acquisition unit. Its input and output are all physical quantities such as residual energy, duration, channel occupancy rate and concurrent contention intensity, which facilitates fast forward inference under limited computing power during power failure.

[0013] Furthermore, the optional actions include continuing to wait for confirmation, sending immediately and retransmitting with adaptive backoff after sending, and terminating the waiting and sending in a degraded manner when the residual energy budget is expected to be insufficient; the acquisition unit is configured to widen the backoff window width of the adaptive backoff as the posterior estimate of the concurrent contention intensity increases, and when the residual energy budget is insufficient to support the complete power outage event message and its retransmission, the power outage event message is downgraded to a minimum power outage event message containing only the local box identifier and the time of power failure. Specifically, the acquisition unit uses the residual energy budget of its own unit as the sole shared budget. For operations such as continuing to wait for confirmation and sending / retransmitting, the energy consumed per unit time or per transmission is deducted from the same budget, causing them to compete with each other. When the residual energy budget is low or the concurrent contention intensity is high, the decision tends to shorten the power outage confirmation waiting time and initiate transmission earlier to avoid exhausting energy before transmission. When AC power is restored during the waiting period, it is judged as a momentary power outage and the process is terminated to suppress false alarms. When the residual energy budget is sufficient and the concurrent contention intensity is low, the decision tends to extend the power outage confirmation waiting time to improve the reliability of the actual power outage determination. The adaptive backoff widens the backoff window width as the posterior estimate of the concurrent contention intensity increases. The more concurrent nodes, the greater the dispersion of the transmission times randomly selected by each unit within the widened backoff window. Thus, without relying on any surviving upper-layer nodes, the known network layer structure, or exchanging information with neighbors, the probability of concurrent collisions is reduced solely by the local estimation of the unit. Furthermore, the acquisition unit can adaptively select the transmission power, modulation rate, and message length based on the current residual energy budget. When the residual energy budget is insufficient to support a power outage event message carrying complete fields and its retransmission, the message will be downgraded to a minimum power outage event message containing only the local box identifier and the time of power failure, so as to ensure that at least one frame of information that can be used to locate the power outage area is delivered before the energy is exhausted.

[0014] Furthermore, the acquisition unit is configured to control the wireless communication unit to send a power outage event message according to the decided action and listen for a receipt within a set waiting window. If no receipt is received before the remaining energy budget is exhausted and the remaining energy budget still supports transmission, the unit will re-determine whether to retransmit and back off based on the updated decision state vector. If a receipt is received or the remaining energy budget is insufficient to support retransmission, the process ends. The acquisition unit also records the actual energy consumed, the predicted amount of energy consumed, and the result of whether the delivery was successful after each transmission. This information is used to correct the identification of the capacitance and equivalent series internal resistance, the inference of the concurrent contention intensity, and the parameters of the decision network online. Specifically, after each transmission, the acquisition unit listens for the receipt from the concentrator or base station within a set waiting window. Receiving a receipt indicates successful delivery of the power outage event message and the process ends. If no receipt is received, and the remaining energy budget still supports retransmission, the updated remaining energy budget, consumed waiting time, number of retransmissions, and current channel occupancy are reconstructed into a decision state vector and re-input into the decision network to decide whether to retransmit and the backoff window before retransmission. If the remaining energy budget is insufficient to support another complete transmission, the process ends to avoid unnecessary energy depletion. Simultaneously, the acquisition unit records the actual energy consumed in this transmission, the predicted energy consumption in each step, and the receipt result in non-volatile storage. These records are used to online correct the identification of supercapacitor capacitance and equivalent series internal resistance, the calibration parameters of concurrent contention observation likelihood, and the parameters of the decision network between adjacent power outage events, forming a closed-loop self-calibration between power outage events. This ensures that the remaining energy estimation and concurrent contention inference continuously adapt to the aging of the transformer and the long-term changes in the characteristics of the distribution area.

[0015] Furthermore, the metering box body is also equipped with a supercapacitor terminal voltage sampling unit, a load current sampling unit, and a temperature sampling unit. The supercapacitor terminal voltage sampling unit is used to collect the terminal voltage of the supercapacitor, the load current sampling unit is used to collect the load current currently drawn by the acquisition unit and the wireless communication unit, and the temperature sampling unit is used to collect the current temperature value. The supercapacitor is charged when the AC power supply is normal, and the bus voltage detection unit is also used to detect the phase voltage of the AC bus of the metering box body. Specifically, the supercapacitor terminal voltage sampling unit, load current sampling unit, and temperature sampling unit provide the acquisition unit with the terminal voltage sequence, load current sequence, and operating temperature required for online identification of capacitance and equivalent series internal resistance, respectively. The terminal voltage and load current constitute the observations of the discharge state model, while the temperature value participates in the accumulation of aging counts and the selection of identification priors. When AC power supply is normal, the supercapacitor is charged to its operating voltage via a charging circuit and maintained thereafter to provide backup energy for the acquisition unit and wireless communication unit after AC power supply interruption. The bus voltage detection unit not only provides the AC bus power failure signal but also detects the three-phase phase voltage of the AC bus, providing three-phase phase voltage drop waveforms for power failure feature classification and fault subgroup inversion. The sampling outputs of the supercapacitor terminal voltage sampling unit, load current sampling unit, temperature sampling unit, and bus voltage detection unit jointly support the aforementioned online identification of residual energy, concurrent contention intensity inference, and collaborative adaptive decision-making.

[0016] Using this invention, the unknown drift of the residual energy budget is actively incorporated into the decision-making process by online identification of the supercapacitor capacitance and equivalent series internal resistance. This ensures that the actual available energy caused by aging and temperature is accurately estimated instead of using fixed backup assumptions, avoiding pre-transmission exhaustion due to overestimation of energy. The power outage confirmation waiting and the transmission and retransmission of power outage event messages are placed under the same residual energy budget constraint, with each energy consumption mutually crowding out and coordinating, rather than being fixed settings. This achieves a compromise that maximizes the delivery probability in place, balancing the contradiction that longer waiting time increases the chance of confirming a real power outage but also increases the likelihood of energy exhaustion. Under the condition that there are no surviving relays available for aggregation and no prior knowledge of network layering, the intensity of concurrent contention is inferred locally from the power outage characteristics of the box, local carrier sensing, and fault subgroup inversion. The backoff window width is adaptively widened accordingly, allowing dozens or even hundreds of metering boxes that lose power almost simultaneously to spontaneously stagger their transmission times and reduce collisions without exchanging information. The decision network uses physical quantities as input and output and continuously adapts to the aging of the box and the characteristics of the distribution area through offline training and online self-calibration. Therefore, in special circumstances where there are simultaneous power outages at the distribution area or line level, the residual energy budget is unknown and drifts, and the concurrent scale is unobservable, a single metering box can maximize the probability of successfully delivering its own power outage event message within the mutually exclusive coupling constraints of energy, timeliness, and reliability. This avoids loss of reports due to premature energy depletion or channel concurrency collisions, thereby ensuring the integrity of the power supply side's power outage range definition and distribution area topology restoration, and shortening the response time for emergency repair location. Attached Figure Description

[0017] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] Figure 1 This is a structural block diagram of the electricity information collection and metering box with power outage monitoring and wireless reporting of the present invention.

[0019] Figure 2 This is a flowchart of the method for online identification of residual energy, inference of concurrent contention intensity, collaborative adaptive decision-making and transmission execution, and online self-calibration performed after detecting a bus power failure signal according to the present invention.

[0020] Figure 3 This is a power supply circuit module diagram of the supercapacitor charging and power-off switching power supply in the metering box of this invention.

[0021] Figure 4 This is a circuit module diagram of the bus voltage detection unit of the present invention. Detailed Implementation

[0022] To make the objectives and technical solutions of this invention clearer, the embodiments of this invention will be further described in detail below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of this invention and should not be construed as limiting the scope of protection of this invention.

[0023] Example 1 This embodiment provides a complete method for an electricity information collection and metering box with power outage monitoring and wireless reporting, such as... Figure 1 As shown, the metering box body is equipped with a data acquisition unit, a wireless communication unit, a supercapacitor, a bus voltage detection unit, a supercapacitor terminal voltage sampling unit, a load current sampling unit, and a temperature sampling unit. The bus voltage detection unit detects the three-phase phase voltage of the AC bus of the metering box body and its power failure signal, and its circuit configuration is as follows. Figure 4 As shown: After the three-phase phase voltage is stepped down and isolated by the three-phase voltage divider isolation circuit, one path is sent to the power-down comparison circuit to be compared with the reference voltage obtained by resistor voltage division of the DC power supply to generate a power-down signal. The other path is sent to the waveform sampling circuit to record the drop waveform. Both the power-down signal and the drop waveform are sent to the acquisition unit. The supercapacitor is charged to the working voltage through the charging circuit and maintained when the AC power supply is normal. After the AC power supply is interrupted, it supplies power to the acquisition unit and the wireless communication unit; its power supply circuit is as follows. Figure 3 As shown: The AC bus outputs DC power through a rectifier-step-down circuit. One path charges the supercapacitor via a charging circuit, while the other path, connected to the positive terminal of the supercapacitor, is combined with two diodes in a power switching circuit to supply power to the acquisition unit and the wireless communication unit. When the AC power supply is interrupted, the diode on one side of the supercapacitor naturally conducts and seamlessly takes over the power supply. The supercapacitor terminal voltage sampling unit, load current sampling unit, and temperature sampling unit provide the acquisition unit with the terminal voltage sequence, load current sequence, and operating temperature value, respectively. In this embodiment, the nominal capacity of the supercapacitor is taken as 10 farads to 50 farads, the charging and holding operating voltage is taken as 5.0 volts, and the minimum bus voltage required for the acquisition unit and the wireless communication unit to maintain normal operation is the cutoff operating voltage. The voltage is set to 2.7 volts. When the bus voltage detection unit detects a bus power failure signal, the acquisition unit sequentially executes steps one through four, and the overall process is as follows: Figure 2 As shown.

[0024] Step 1: Residual Energy Online Identification. Upon detecting a bus power failure signal, the acquisition unit immediately initiates the residual energy estimation process. In sub-step 1, the acquisition unit acquires the terminal voltage sequence of the supercapacitor at a fixed sampling period. The load current sequence currently acquired by the acquisition unit and the wireless communication unit And simultaneously collect the current temperature value, among which This is the sampling sequence number. The fixed sampling period is preferably between 1 millisecond and 10 milliseconds, and in this embodiment, it is 2 milliseconds. At the moment of power failure, the supercapacitor bears all the load current, and its terminal voltage shows an observable decreasing trend as it discharges. The shorter the sampling period, the more observable samples are available in the short time window and the faster the identification convergence. If the value is too short, it will increase the power consumption of the acquisition unit itself. Therefore, a compromise of 2 milliseconds is made to achieve a balance between identification accuracy and self-consumption.

[0025] In sub-step 2, the acquisition unit establishes a discharge state model based on the charge conservation and Ohmic laws governing the supercapacitor discharge process. Specifically, the terminal voltage equals the voltage across the capacitor minus the voltage drop across the equivalent series resistance of the load current, and the voltage across the capacitor decreases linearly with the cumulative discharged charge according to the reciprocal of the capacitance. ,in The voltage across the capacitor at the initial moment of power failure. For the capacitance to be identified, The equivalent series internal resistance to be identified is... The sampling period is As of the date The cumulative charge released by each beat. Rewriting the above formula as the reciprocal of capacitance... With equivalent series internal resistance Linear observation equations with unknown coefficients ,in To accumulate the released charge.

[0026] The acquisition unit substitutes the measured terminal voltage and load current frame by frame, and uses recursive least squares method or extended Kalman filter to identify the parameter vector online. When using the recursive least squares method, its recursive update is as follows: Gain Covariance recursion ,in For the regression vector, The forgetting factor is preferably set between 0.95 and 0.99; in this embodiment, it is set to 0.98 to weight and attenuate old observations, thereby tracking slowly changing parameters. When using extended Kalman filtering, the capacitance and equivalent series internal resistance are incorporated into the state vector and recursively updated along with the terminal voltage observation equation. The process noise covariance is set to a small value according to the slow-changing characteristics of aging. Both methods converge within a short time window of tens of milliseconds after power failure. In this embodiment, the typical convergence time is 20 to 40 milliseconds. This ensures that the capacitance decay and internal resistance increase of supercapacitors with the same nominal capacity due to differences in service life, ambient temperature, and charge / discharge history are accurately reflected in the identification results, avoiding overestimation of usable energy by using the factory nominal value.

[0027] In sub-step 3, the acquisition unit uses the identified capacitance... Current terminal voltage With cutoff operating voltage ,according to The residual energy of the supercapacitor is calculated, where the residual energy is the difference between the square of the current terminal voltage and the square of the cutoff voltage multiplied by half to identify the capacitance. The acquisition unit uses the product of the current load current and the terminal voltage. Estimate the instantaneous power consumption, divide the remaining energy by the instantaneous power consumption, and take into account the power consumption change caused by the drop in terminal voltage during discharge. This is then converted into a budget for the usable operating time that can be maintained before the remaining energy is exhausted. Simultaneously, based on the covariance matrix of the identification parameters The estimation uncertainty of this budget is given by error propagation. This information is used to inform subsequent decision-making regarding energy margin. Preferably, the data acquisition unit maintains an aging count that accumulates with the number of charge-discharge cycles and operating temperature. The value is incremented by one for each supercapacitor charge-discharge cycle, and additionally accumulated based on temperature weighting during high-temperature operation; the acquisition unit obtains the prior interval of capacitance and equivalent series internal resistance by looking up a table or by using an empirical function based on aging counts, which serves as the initial value for recursive identification at the moment of power failure. with initial covariance This shortens the recognition convergence time, enabling rapid and stable recognition during the initial power outage phase when energy is most precious. In this embodiment, the actual capacitance of a supercapacitor with a nominal capacity of 20 farads can be reduced to 15 farads after aging, and the equivalent series internal resistance can be increased to 1.4 times the nominal value. Based on this, the estimated available working time is shortened by about 25% compared to using the nominal value, thereby avoiding overestimation of energy and depletion before transmission.

[0028] Step 2, Concurrency Contention Intensity Inference. The acquisition unit estimates the concurrent contention scale of channels within the same area when the local box loses power. In sub-step 1, the acquisition unit reads the power failure characteristics recorded by the bus voltage detection unit within a short time window of several milliseconds to tens of milliseconds before and after the power failure. In this embodiment, the short time window is 10 milliseconds before and after the power failure. The power failure characteristics include whether the three phases drop simultaneously, the bus voltage drop rate, and the drop depth. The acquisition unit compares whether the three-phase voltage drops are synchronized, whether the drop edges are steep, and the final drop depth. When the three-phase voltages drop synchronously and rapidly to below the cutoff threshold at the same time, i.e., the three-phase drop time difference is less than 5 milliseconds and the drop rate exceeds the set threshold, it is determined that the power outage is caused by a transformer outgoing line trip, a main line fault, or a planned full-area power outage, resulting in a simultaneous power outage at the line level or substation level. Such events mean that a large number of metering boxes in the same substation enter the power outage reporting process almost simultaneously, so a higher prior for concurrent contention intensity is given. When only a single-phase drop or a three-phase slow drop occurs, it is determined to be a local power outage, and a lower prior for concurrent contention intensity is given to provide initial beliefs for subsequent Bayesian fusion.

[0029] In sub-step 2, the acquisition unit controls the wireless communication unit to perform a short carrier listening period before formal transmission. In this embodiment, the listening window is 5 milliseconds. The acquisition unit calculates the proportion of time the channel is occupied within the listening window as the channel occupancy rate. The received signal strength is measured, and the characteristic preamble carried in the power outage event messages sent by other metering boxes on the channel is matched and counted to obtain the preamble count. A higher leading count and higher occupancy rate indicate that more neighbors are reporting simultaneously.

[0030] In sub-step 3, the acquisition unit performs Bayesian fusion based on the physical law that the occupancy rate and collision statistics of multiple devices transmitting concurrently on the radio frequency channel change monotonically with the number of concurrent devices. The acquisition unit pre-calibrates the correspondence between the number of concurrent devices and the channel occupancy rate and preamble count, which increase monotonically with the number of concurrent devices, and establishes a system based on the number of concurrent contention devices. As a latent variable, with channel occupancy rate With leader count The observation likelihood function of the observed quantity The prior estimate given by the power-down feature classification in sub-step 1 is used as the prior distribution. The distribution of historical power outage periods recorded in the non-volatile storage of this box and the historical patterns of clustered power outages in the corresponding transformer area are used as additional priors. Multiplying the likelihood by the channel observations in a Bayesian manner and normalizing it, i.e. The posterior estimate and uncertainty of the current concurrent contention intensity are obtained. In this embodiment, the expectation of the posterior distribution is used as the point estimate of the concurrent contention intensity, and its standard deviation is used as the uncertainty. Furthermore, the acquisition unit also reads the three-phase phase voltage drop waveforms acquired by the bus voltage detection unit. Based on the synchronicity of the drop time and the correlation of the drop slope of the metering boxes on the same feeder in the same tripping event, the drop waveform of this box is compared with the characteristics of its neighboring waveforms. The scale and boundary of the fault subgroup to which this box actually belongs are inverted. The range of devices that are truly competing for the channel with this box is converged to the strong concurrent subgroup that trips at the same time on the same feeder. Devices that belong to the same transformer area but are de-energized by different protection actions at different times are classified as weak concurrent subgroups and eliminated. The scale of this fault subgroup is used to replace the overall scale of the transformer area to drive the inference of the concurrent contention intensity, avoiding an overestimation of the concurrency level due to a general estimate of the entire transformer area. Thus, prior and real-time channel observations complement each other, providing robust estimates even with extremely short observation windows and limited information.

[0031] Step 3: Coordinated adaptive decision-making between power outage confirmation waiting time and power outage event message sending strategy. The acquisition unit will use the residual energy budget obtained in Step 1. and its uncertainty Current power outage confirmation waiting time The bus voltage detection unit provides an indicator of whether the AC power supply has been restored. The posterior estimate of the concurrent contention intensity obtained in step two and its uncertainty Current channel occupancy rate and the current number of resentments Together they constitute the decision state vector Each component is a technical quantity with a definite physical meaning. The acquisition unit periodically refreshes this vector and inputs it into the decision network throughout the entire power-down discharge process with a fixed millisecond-level decision step size. In this embodiment, the decision step size is 10 milliseconds, so that the decision is dynamically updated as the residual energy dissipates, the waiting time accumulates, and the channel occupancy and concurrency estimation change.

[0032] During the inference phase, the acquisition unit employs a pre-trained decision network, using the decision state vector... The input is the energy source, and the output is the expected successful delivery probability of the power outage event message for this box until the residual energy is exhausted, corresponding to each optional action in the current state. In sub-step 1, the acquisition unit uses the residual energy budget of this box as the only shared budget. For the two types of operations, continuing to wait for confirmation and sending and retransmitting, the energy consumed by each operation per unit time or per transmission is deducted from the same residual energy budget, so that the two compete for each other under the same budget constraint. In sub-step 2, the decision network that has been trained and quantized and solidified in the acquisition unit performs forward inference on the current state vector and outputs the expected successful delivery probability of each action. In sub-step 3, the set of optional actions includes continuing to wait for confirmation, sending immediately and then retransmitting with adaptive backoff after sending, and terminating the waiting and sending in a degraded manner when the residual energy is expected to be insufficient; the acquisition unit selects the action that maximizes the expected successful delivery probability to execute. Specifically, when the remaining energy budget is low or the intensity of concurrent contention is high, the decision tends to shorten the power outage confirmation waiting time and initiate transmission earlier to avoid exhausting energy before transmission; when AC power is restored during the waiting period, it indicates... When the flag is set, a momentary power outage is identified and the process terminates to suppress false alarms. When the residual energy budget is sufficient and the intensity of concurrent contention is low, the decision tends to extend the power outage confirmation waiting time to improve the reliability of the actual power outage determination. Furthermore, the adaptive backoff increases the backoff window width. Posterior estimation of concurrent contention intensity The width increases and expands, and in this embodiment, it is taken as... ,in Based on the retreat window, The expansion factor represents the number of concurrent units. The greater the dispersion of transmission times randomly selected by each box within the expanded backoff window, the lower the probability of concurrent collisions can be achieved solely through local estimation, without relying on any surviving upper-layer nodes, a known network hierarchy, or information exchange with neighbors. Preferably, the acquisition unit also adaptively selects the transmission power, modulation rate, and message length based on the current residual energy budget: when the residual energy is insufficient to support a power outage event message carrying complete fields and its retransmission, the message is downgraded to a minimum power outage event message containing only the box identifier and the time of power failure, ensuring that at least one frame of information usable for locating the power outage area is delivered before energy depletion.

[0033] The training of the decision network is completed offline on a computing platform outside the metering boxes. Part of the training dataset is taken from historical reports of power outage discharge processes, channel collision processes, and final delivery results from the field metering boxes. Another part is generated in batches by simulation models of power outage discharge and RF channel collisions, with different concurrent metering numbers, different initial residual energies, and different channel conditions. The simulation covers various operating conditions with concurrent metering numbers ranging from a few to hundreds and initial residual energies from ample to critical, generating samples on the order of millions. In the data preprocessing stage, the acquired raw sequences are cleaned to remove sampling anomalies and missing data. The physical quantities of the state vector are normalized according to their dimensions. For rare high-concurrency extreme operating conditions, data augmentation is performed using resampling and noise addition to balance the sample distribution. The samples are divided into training, validation, and test sets, with proportions of 70%, 15%, and 15%, respectively. The decision network adopts a multi-layer fully connected neural network or temporal network structure. This embodiment uses an input layer, three fully connected hidden layers (each with 128 neurons), and an output layer, built using a deep learning framework and trained on a graphics processing unit. Training is conducted using reinforcement learning, where each successful delivery is counted as a positive reward, and any undelivered delivery before energy depletion is counted as a zero or negative reward. The optimization objective is to maximize the expected successful delivery probability until the remaining energy is exhausted. The loss function is the mean square form of temporal difference error, ensuring that the network's output expected successful delivery probability approximates the target value obtained from reward backtracking. The optimizer employs an adaptive moment estimation optimizer, with an initial learning rate of [value missing]. In conjunction with an exponential decay strategy triggered by the validation loss platform, a batch size of 128 was used. The number of training epochs was typically 800 to 2000 epochs, with the convergence criterion being that the successful delivery rate on the validation set no longer increased. L2 regularization of the weights and an early stopping mechanism based on the successful delivery rate on the validation set were employed to suppress overfitting. The evaluation metric was the successful delivery probability under residual energy constraints, which was compared with a baseline strategy of fixed waiting plus fixed backoff to verify the delivery probability gain of the decision network under the condition of simultaneous power outages. The trained network parameters were quantized and fixed in the acquisition unit, with all inputs and outputs being physical quantities such as residual energy, duration, channel occupancy rate, and concurrent contention intensity, facilitating rapid forward inference under limited computing power during power outages.

[0034] Step four: Sending execution and online self-calibration. The acquisition unit, following the action decided in step three, controls the wireless communication unit to send a power outage event message and listens for the receipt from the concentrator or base station within a set waiting window. In this embodiment, the receipt waiting window is 50 milliseconds. Receiving a receipt is considered a successful delivery of the power outage event message and the process ends. If no receipt is received before the remaining energy is exhausted and the remaining energy budget still supports retransmission, the acquisition unit reconstructs the decision state vector with the updated remaining energy budget, consumed waiting time, number of retransmissions, and current channel occupancy, and returns to step three to re-input the decision network to decide whether to retransmit and the backoff window before retransmission. If a receipt is received or the remaining energy budget is insufficient to support a complete retransmission, the process ends to avoid unnecessary energy depletion. Simultaneously, the acquisition unit records the actual energy consumed in this transmission, the predicted amount consumed in step one, and the result of whether a receipt was received in non-volatile storage. These records are used to online correct the identification of supercapacitor capacitance and equivalent series internal resistance, the calibration parameters of concurrent contention observation likelihood, and the parameters of the decision network between adjacent power outage events, forming a closed-loop self-calibration between power outage events. Thus, the residual energy estimation and concurrent contention inference continuously adapt to the aging of the metering box and the long-term changes in the characteristics of the corresponding transformer area. This allows the invention to maximize the probability of successful delivery of power outage event messages for a single metering box within the mutually exclusive coupling constraints of energy, timeliness, and reliability, avoiding reporting loss due to premature energy depletion or channel concurrency collisions. This ensures the integrity of the power outage range definition and transformer area topology restoration on the power supply side and shortens the response time for emergency repair location.

[0035] Example 2 In a preferred embodiment of the present invention, after detecting a power failure signal, the acquisition unit models two previously unrelated operations—power failure confirmation waiting and the sending and retransmission of power failure event messages—as a sequential decision-making process that iteratively solves the same residual energy budget constraint with a fixed decision step size. The decision network then provides the action that maximizes the expected successful delivery probability at each decision step. Specifically, the acquisition unit uses a fixed decision step size at the millisecond level. The current state is refreshed, and the decision state vector is denoted as follows. ,in This is the current residual energy budget obtained through online identification and calculation. The estimated uncertainty of this budget, The elapsed power outage confirmation waiting time. For the posterior estimation of concurrent contention intensity, The uncertainty of this posterior estimate, The current channel occupancy rate. This represents the number of retransmissions that have been completed. This is a flag indicating whether AC power supply has been reset. The set of selectable actions is denoted as... These correspond to continuing to wait for confirmation, initial transmission, retransmission with adaptive backoff, and terminating the wait and sending a degraded message when the remaining energy budget is expected to be insufficient.

[0036] Specifically, the decision network uses policy functions Output the probability of choosing each available action in the current state, where The network parameters are trained offline and embedded in the acquisition unit; for each state-action pair, its value function is defined as the accumulated expected successful delivery probability until the residual energy budget is exhausted, denoted as . ,in The termination step is the one corresponding to the depletion of the remaining energy budget or the receipt of a delivery confirmation. For the first The system provides immediate feedback for each step: a value of 1 indicates successful delivery of a receipt; a value of 0 indicates no receipt received before energy is depleted; and a small negative value is used to reflect the collision cost when a message is lost due to concurrent collisions or channel obstruction. The decision network operates according to the Bellman equations during the training phase. The optimal value function is approximated through repeated iterations. Forward inference is performed only once during the deployment phase, outputting the value estimate for each action and selecting the action that maximizes the probability of successful delivery. ,in This is the output of the decision network's estimate of the state-action value. Since the inputs and outputs of this decision network are all technical quantities with definite physical meanings, such as residual energy, duration, channel occupancy rate, and concurrent contention intensity, its forward inference only involves a few layers of matrix multiplication and addition operations. It can be completed in the sub-millisecond range under the limited computing power and power consumption budget during power failure, without consuming any additional precious residual energy.

[0037] Furthermore, the decision network preferably adopts a multi-layer fully connected structure with the same input layer dimension and number of state vector components, containing 2 to 4 hidden layers with 32 to 128 neurons per layer, and an output layer dimension the same as the action set size, outputting the probability of each action selection through a normalized exponential function. When the decision scenario needs to consider the temporal dependencies between multiple consecutive decision steps, a gated recurrent unit or long short-term memory unit can be optionally inserted between the input layer and the fully connected layer, with a hidden state dimension of 16 to 64, to capture the temporal evolution characteristics of residual energy decreasing with decision steps and concurrent contention intensity continuously corrected with channel observation. Part of the training samples for the decision network are taken from the actual discharge process, channel collision process, and final delivery result recorded in historical power outage events of the field metering box; another part is generated by batch simulation of the power outage discharge model and the radio frequency channel collision model under different concurrent numbers, different initial residual energies, and different channel conditions, and then augmented with data. The samples are divided into training set, validation set, and test set, with proportions of 70%, 15%, and 15%, respectively. An adaptive moment estimation optimizer is used during training, with an initial learning rate of [missing information]. to It employs a learning rate scheduling mechanism that decays based on validation set performance, with batch sizes ranging from 64 to 256 and training epochs ranging from hundreds to thousands. The loss function is a weighted combination of temporal difference error and policy entropy regularization, denoted as... ,in Used to approximate the value function Information entropy of a strategy, used to prevent decisions from prematurely converging to a single action. The parameters are set to 0.01 to 0.1. During training, weight regularization is applied, and an early stopping mechanism is set based on the change in the successful delivery probability under the residual energy constraint on the validation set. The evaluation metric is the successful delivery probability under the residual energy constraint, which is compared with the baseline strategy with a fixed waiting time and a fixed backoff window. After training, the decision network parameters are quantized, compressed, and stored in the memory of the acquisition unit for real-time retrieval in the event of power failure.

[0038] Preferably, the decision step size The decision step size is set between 10 milliseconds and 100 milliseconds. Too short a step size will increase the frequency of inference in the decision network, consuming additional energy, while too long a step size will cause state updates to lag behind the actual rate of residual energy loss; the adaptive backoff window width... Posterior estimation of concurrent contention intensity The relationship between them is approximately linear or piecewise linear, i.e. ,in The baseline backoff window width is set when the concurrent contention intensity approaches zero. The expansion coefficient is set to 1 to 5, so that for every unit increase in the posterior estimate of the concurrent contention intensity, the backoff window is expanded proportionally, thereby reducing the collision probability in scenarios where multiple simultaneous power outages occur at a large concurrent scale. The energy margin coefficient is preferably set to 1.1 to 1.3, that is, when the decision network judges whether the residual energy budget supports a certain action, it multiplies the expected consumption of the action by the energy margin coefficient and then compares it with the current residual energy budget. This provides a buffer for online identification of estimation uncertainties and avoids unexpected energy depletion during transmission due to identification errors.

[0039] The collaborative adaptive decision-making method described in this embodiment, under the premise of the same residual energy budget, can improve the expected success rate of delivery compared to the existing processing order where power outage confirmation waiting and retransmission are independent and each uses a fixed duration. This improvement is particularly significant in scenarios where a large number of metering boxes compete for channels simultaneously due to simultaneous power outages at the substation or line level. Since the forward inference of the decision network only requires a few layers of matrix operations, its computation time and power consumption are in the order of milliwatts to tens of milliwatts, which does not bring additional energy burden compared to the judgment logic required by the fixed strategy. Since the residual energy budget and concurrent contention intensity are estimated online rather than fixed assumptions, this decision-making method has a certain degree of adaptability to the capacity decay of supercapacitors due to service life and ambient temperature, as well as the difference between the historical power outage patterns of the substation and the actual situation. It can still give a decision that is consistent with the actual state of the box in scenarios where the aging degree of supercapacitors is different or the ambient temperature deviates significantly from the calibration conditions, demonstrating strong robustness. Of course, in certain special scenarios, such as when the protection action mode of a feeder belonging to a certain metering box is special, causing its historical power outage period distribution to deviate significantly from the overall pattern of the distribution area, the prior component weights given by the decision network may not be completely matched with the actual situation. In this case, the decision relies more on online channel observation and residual energy estimation itself for correction.

[0040] The reason for achieving the above effects, in principle, lies in the fact that this embodiment no longer treats the power outage confirmation waiting and the transmission and retransmission of power outage event messages as two independent stages with fixed durations. Instead, it unifies them into a sequential decision problem under the same residual energy budget constraint: a longer confirmation waiting time is more conducive to eliminating false alarms caused by momentary power loss and improving the reliability of power outage determination, but it also continuously consumes the already limited and non-renewable residual energy budget, compressing the energy space available for subsequent transmission and retransmission; conversely, earlier transmission can reserve more energy for retransmission to combat losses caused by channel collisions, but may generate false alarms for momentary power loss due to insufficient confirmation. The decision network, through iterative approximation of the value function, essentially seeks the optimal trade-off point between the confirmation reliability gains brought by the waiting time and the retransmission capability losses caused by energy consumption, which dynamically changes with the current residual energy, concurrent contention intensity, and the waiting time, rather than using a fixed threshold applicable to all operating conditions. Meanwhile, since the decision state vector incorporates the posterior estimate of the concurrent contention intensity, the decision network tends to send data earlier and widen the backoff window when the concurrent contention intensity is high, and tends to extend the waiting time to improve the reliability of confirmation when the concurrent contention intensity is low. This allows the same decision mechanism to adaptively switch the weight between reliability and timeliness according to the actual contention scale and energy state of the box. This is the difference in technical principle between this embodiment and the existing processing order that uses fixed waiting plus fixed backoff.

[0041] Example 3 In another embodiment of the present invention, as an optional alternative, the acquisition unit does not use an offline-trained decision network to perform forward reasoning decisions on optional actions. Instead, it uses a value iteration method based on dynamic programming to pre-solve an offline state-action lookup table. After a power outage occurs, it completes the collaborative decision-making between power outage confirmation waiting and the sending and retransmission of power outage event messages by looking up the table. This achieves a similar technical effect to Embodiment 2 with lower online computational overhead and more easily verifiable decision logic. Specifically, the residual energy budget, the consumed power outage confirmation waiting time, the posterior estimate of concurrent contention intensity, and the number of retransmissions are discretized into a finite number of value intervals according to a preset quantization interval, and combined to form a finite set of states. The four optional actions of continuing to wait for confirmation, sending immediately, retransmitting with adaptive backoff, and sending in a degraded manner are used as a finite set of actions. For each combination in the state set and action set, based on the typical discharge characteristic curve of the supercapacitor and the typical collision probability model of the radio frequency channel, the transition probability and immediate cost of the state-action pair to obtain successful delivery before the residual energy budget is exhausted are calculated offline, and then calculated according to the value iteration formula. The process is iterated repeatedly until the difference between the value functions obtained from two adjacent iterations is less than the preset convergence threshold. In this way, the optimal action that maximizes the expected successful delivery probability is obtained for each discrete state. The mapping relationship between the state and the corresponding optimal action is solidified into a lookup table and stored in the non-volatile memory of the acquisition unit.

[0042] Preferably, the discretization interval for the residual energy budget can be 10 to 20 levels, the discretization interval for the posterior estimate of concurrent contention intensity can be 5 to 10 levels, the discretization interval for the consumed waiting time can be 8 to 16 levels in multiples of the decision step size, and the number of retransmissions can be discretized according to the actual maximum allowed retransmission limit. The resulting state-action lookup table is in the thousands to tens of thousands of entries, which can be completely stored in the existing non-volatile storage resources of the acquisition unit without constituting an additional hardware burden. After a power outage, the acquisition unit maps the real-time residual energy budget, consumed waiting time, posterior estimate of concurrent contention intensity, and number of retransmissions to the nearest state item in the lookup table according to the above discretization intervals, directly reads the optimal action corresponding to the state item, and controls the wireless communication unit to perform the corresponding operation accordingly. No forward neural network operation is required; only numerical comparison and one memory lookup are involved. Its online computation time and power consumption are lower than the forward inference overhead of the decision network in Embodiment 2, which is especially suitable for low-cost metering box products with more limited computing power and storage resources.

[0043] Furthermore, considering the quantization errors that may be introduced by state discretization, after obtaining the optimal action from the lookup table, the acquisition unit can also fine-tune the execution parameters of the action based on the offset direction of the current state relative to the mapped discrete level. For example, when the actual residual energy budget is slightly higher than the center value of the mapped level, the upper limit of the waiting time corresponding to that level is appropriately extended; when the actual residual energy budget is slightly lower than the center value of the mapped level, the transmission is initiated earlier. This value iteration lookup table can be periodically resolved offline and updated based on the power outage event records accumulated by the field metering box during actual operation, in order to absorb the changes in the aging law of the supercapacitor and the historical power outage characteristics of the corresponding transformer area over time. The update process is completed on a computing platform outside the metering box, and the updated lookup table is downloaded through the maintenance channel after verification to replace the original lookup table. As an application boundary of this alternative, when the concurrent contention intensity or residual energy budget distribution of the metering box's operating condition deviates significantly from the typical discharge characteristic curve and channel collision model used in the offline solution value iteration table, the optimality of the lookup table decision may not be as good as the adaptability obtained by the decision network in Example 2 based on actual sample training. In this case, the decision network described in Example 2 and the lookup table of this example can be deployed in parallel, and the lookup table results can be used as backup decision basis when the decision network reasoning is abnormal, taking into account both computational overhead and decision adaptability.

[0044] Using the value iteration lookup table method described in this embodiment, the power outage confirmation waiting and the sending and retransmission of power outage event messages can also be coordinated and weighed under the same residual energy budget constraint. Moreover, the posterior estimation of the concurrent contention intensity still drives the adaptive widening of the backoff window. Its decision basis and solution objective are consistent with those of Embodiment 2. The only difference is that offline value iteration replaces offline neural network training and online table lookup replaces online forward inference in the method of obtaining the optimal decision. This provides another feasible technical path for metering box products under different hardware conditions and different cost constraints.

[0045] Example 4 This embodiment provides an electricity information collection and metering box with power outage monitoring and wireless reporting, corresponding to the method described in Embodiment 1. That is, it implements the aforementioned method from a device perspective, such as... Figure 1 As shown, the metering box includes a metering box body, and a data acquisition unit, a wireless communication unit, a supercapacitor, and a bus voltage detection unit disposed within the metering box body. The metering box body also includes a supercapacitor terminal voltage sampling unit, a load current sampling unit, and a temperature sampling unit. The bus voltage detection unit is used to detect the three-phase phase voltage of the AC bus of the metering box body and its power-off signal. When the AC power supply is normal, the supercapacitor is charged to its operating voltage via a charging circuit and maintained thereafter. After the AC power supply is interrupted, it supplies power to the data acquisition unit and the wireless communication unit. The supercapacitor terminal voltage sampling unit is used to acquire the terminal voltage of the supercapacitor. The load current sampling unit is used to acquire the load current currently drawn by the data acquisition unit and the wireless communication unit. The temperature sampling unit is used to acquire the current temperature value. The charging, holding, and power-off switching power supply circuit of the supercapacitor is as follows... Figure 3 As shown, the circuit configuration of the bus voltage detection unit is as follows: Figure 4 As shown, its circuit implementation is the same as that described in Example 1.

[0046] The acquisition unit includes a residual energy identification module, a concurrent contention inference module, a collaborative decision-making module, and a transmission self-calibration module. Each module corresponds one-to-one with steps one through four in Embodiment 1, and their data flow relationships are as follows: Figure 2 As shown.

[0047] The residual energy identification module is used to receive the terminal voltage sequence obtained by the supercapacitor terminal voltage sampling unit and the load current sequence obtained by the load current sampling unit at a fixed sampling period after the bus voltage detection unit detects a power failure signal. Based on the discharge state model established that the terminal voltage equals the voltage across the capacitor minus the voltage drop across the load current on the equivalent series resistance, and the voltage across the capacitor decreases linearly with the discharged charge, the module uses recursive least squares method or extended Kalman filter to identify the current capacitance and equivalent series resistance of the supercapacitor online. Then, based on the identified capacitance, current terminal voltage, and cutoff operating voltage... The residual energy is calculated, converted into a residual energy budget, and its estimation uncertainty is determined. The residual energy identification module also maintains an aging count that accumulates with the number of charge and discharge cycles and the operating temperature obtained by the temperature sampling unit, which is used to provide initial values ​​and priors for identification.

[0048] The concurrent contention inference module is used to read the power outage characteristics recorded by the bus voltage detection unit, and to give a prior estimate of the concurrent contention intensity based on whether the three phases drop simultaneously and the rate and depth of the bus voltage drop. It controls the wireless communication unit to perform short-time carrier sensing before transmission to measure the channel occupancy rate, the received signal strength, and count the preamble of the neighbor power outage event message. It establishes the observation likelihood with the number of concurrent contentioning units as the latent variable, and fuses the prior estimate, the channel observation, and the historical power outage patterns stored in this box in a Bayesian manner. Based on the time synchronization and slope correlation of the three-phase phase voltage drop waveform, it inverts the scale of the fault subgroup to which this box belongs, and obtains the posterior estimate of the concurrent contention intensity and its uncertainty.

[0049] The collaborative decision-making module is used to construct a decision state vector from the residual energy budget and its uncertainty output by the residual energy identification module, the currently consumed power outage confirmation waiting time, the indicator of whether AC power supply has been restored, the posterior estimate of the concurrent contention intensity and its uncertainty output by the concurrent contention inference module, the current channel occupancy rate and the current number of retransmissions. This vector is pre-trained offline and embedded in the decision network of the acquisition unit with a fixed millisecond-level decision step size. The residual energy budget is used as the only shared budget for the power outage confirmation waiting and the transmission and retransmission of power outage event messages to compete for each other according to their respective energy consumption. The module selects the action that maximizes the expected successful delivery probability until the residual energy budget is exhausted.

[0050] The self-calibration module is used to control the wireless communication unit to send a power outage event message according to the action decided by the collaborative decision-making module and listen for the receipt within a set waiting window. If no receipt is received before the residual energy budget is exhausted and the budget still supports transmission, the module returns to the collaborative decision-making module to re-determine the decision based on the updated decision state vector. The module also records the actual energy consumed, the predicted amount, and the delivery result of each transmission in non-volatile storage for online correction of the parameters of the residual energy identification module, the concurrent contention inference module, and the decision network.

[0051] In terms of hardware deployment, the acquisition unit can be implemented by a microcontroller or embedded system-on-a-chip with low-power digital signal processing capabilities. The four modules of residual energy identification, concurrent contention inference, collaborative decision-making and self-calibration are run on the same processor in firmware form. The offline training of the decision network is completed on a general computing platform outside the metering box with the help of a graphics processor. The network parameters obtained from the training are quantized, downloaded through the maintenance channel and solidified in the non-volatile memory of the acquisition unit. The above hardware is a common configuration in the field and is not limited to specific brands and models.

[0052] Example 5 This embodiment illustrates the implementation process and technical effects of the present invention through a specific application case. Suppose a low-voltage distribution area is powered by a distribution transformer, with several low-voltage feeders connecting approximately 180 user meter boxes equipped with the metering boxes described in this invention. On one occasion, the outgoing switch of the distribution transformer trips due to an upstream fault, causing all 180 metering boxes in the area to simultaneously lose AC power. The supercapacitors in each box begin discharging almost synchronously and enter the power outage reporting process. Meanwhile, the upper-level nodes in the area that could act as aggregation relays also lose power simultaneously, leaving no relays available. Because such simultaneous power outages are subject to four constraints—unknown residual energy budget drift, no surviving relays for aggregation, large-scale concurrent channel contention, and unobservable concurrency—the conventional processing sequence of first fixing the waiting confirmation and then fixing or randomly backing off transmission is inadequate: the longer the waiting time, the more likely the energy will be exhausted before transmission; and backoff parameters, not adjusted according to the actual concurrency scale, are difficult to suppress collisions.

[0053] Each metering box is executed according to the steps described in Example 1. Taking one metering box that has been used for about five years and whose supercapacitors have aged to some extent as an example: after a power outage, its residual energy identification module identifies within about 30 milliseconds that the actual capacitance has dropped from the nominal 20 farads to about 15 farads and the equivalent series internal resistance has increased to about 1.4 times the nominal value. Based on this, the estimated available working time is shortened by about 25% compared to the nominal value, thus avoiding overestimation of energy. Its concurrent contention inference module determines that the power outage is caused by a simultaneous power outage at the substation level based on the three-phase synchronous rapid drop. Combining the channel occupancy rate and preamble count measured by 5 millisecond carrier sensing, it inverts the scale of the same feeder tripping fault subgroup to which this box belongs to about 40 units, and expands the backoff time window accordingly. Under the residual energy budget and concurrent estimation, the collaborative decision module selects to appropriately shorten the power outage confirmation waiting, send in advance, and follow the action sequence of expanded backoff retransmission.

[0054] To evaluate the effectiveness, the reporting success rate of power outage event messages across the entire distribution area was statistically analyzed under different concurrent scales under the condition of simultaneous power outages in the same cluster. This success rate was calculated as the percentage of messages successfully delivered to the metering box of the concentrator before the residual energy was exhausted. The results were compared with two control schemes: Control Scheme 1 used an existing processing sequence where power outage confirmation waiting and retransmission were independent and both used fixed-duration sequences; Control Scheme 2 was an ablation scheme based on the present invention, which removed the concurrency contention intensity inference and did not adaptively widen the backoff window with the concurrent scale. With a concurrent scale of approximately 50 devices, the reporting success rate of the present invention was approximately 96.8%, Control Scheme 1 was approximately 88.5%, and the ablation scheme was approximately 93.1%. With a concurrent scale of approximately 120 devices, the present invention achieved approximately 94.2%, Control Scheme 1 approximately 79.3%, and the ablation scheme approximately 87.6%. In a full-area cluster power outage with a concurrent scale of approximately 180 devices, the present invention achieved approximately 91.5%, Control Scheme 1 approximately 71.8%, and the ablation scheme approximately 83.4%. It is evident that the larger the concurrency scale, the more significant the success rate advantage of this invention compared to a fixed processing order. Furthermore, the success rate of the ablation scheme after removing the inference of concurrency contention intensity drops significantly, indicating that this innovative step makes an independent contribution to suppressing concurrent collisions.

[0055] Furthermore, by removing the online residual energy identification from the aforementioned aging metering boxes and instead using the nominal value of the supercapacitor to estimate the energy budget, approximately 15% of the samples showed instances of unexpected energy depletion and reported loss before transmission. However, after adopting online identification, these losses were largely eliminated, indicating that online residual energy identification also independently contributes to preventing premature depletion due to overestimation of energy caused by aging. The above data shows fluctuations of several percentage points across different batches of supercapacitors and under different ambient temperatures, consistent with the dispersed characteristics of real-world operating conditions. This application case demonstrates that, under special circumstances where there are simultaneous power outages at the distribution level, unknown drift in residual energy budgets, and unobservable concurrent scale, this invention can improve the probability of successful delivery of power outage event messages for a single metering box within the mutually exclusive coupling constraints of energy, timeliness, and reliability. This ensures the integrity of the power supply side's outage range definition and distribution topology restoration, and shortens the response time for emergency repair location.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An electricity information acquisition metering box with power-off monitoring and wireless reporting, comprising a metering box body, and an acquisition unit, a wireless communication unit, a super capacitor and a bus voltage detection unit arranged in the metering box body, the bus voltage detection unit is used for detecting the power-off signal of the AC bus of the metering box body, and the super capacitor supplies power to the acquisition unit and the wireless communication unit after the interruption of AC power supply; characterized in that, The acquisition unit is configured to, upon detecting the power failure signal, identify the current capacitance and equivalent series internal resistance of the supercapacitor online and convert them into a residual energy budget. Based on the power failure characteristics and channel occupancy observations, it infers the concurrent contention intensity of the unit. The residual energy budget is used as the sole shared budget. The power failure confirmation waiting and the transmission and retransmission of power failure event messages are deducted from the residual energy budget according to their respective energy consumption, causing them to compete with each other. The decision network outputs the expected successful delivery probability of each optional action until the residual energy budget is exhausted. The action that maximizes the expected successful delivery probability is selected to control the wireless communication unit to perform transmission and retransmission.

2. The electricity information collection and metering box according to claim 1, characterized in that, The acquisition unit is configured to acquire the terminal voltage sequence of the supercapacitor and the load current sequence currently sampled by the acquisition unit and the wireless communication unit at a fixed sampling period after detecting the power failure signal. Based on the fact that the terminal voltage is equal to the voltage across the capacitor minus the voltage drop across the load current on the equivalent series resistance, and that the voltage across the capacitor decreases linearly with the discharged charge, a discharge state model is established with capacitance and equivalent series resistance as the parameters to be identified. Then, using the recursive least squares method or extended Kalman filter, with the terminal voltage sequence and the load current sequence as the observations, the current capacitance and equivalent series resistance of the supercapacitor are identified online, so that the capacitance decrease and internal resistance increase caused by aging and temperature are reflected in the identification results.

3. The electricity information collection and metering box according to claim 2, characterized in that, The acquisition unit is configured to, based on the identified capacitance, current terminal voltage, and cutoff operating voltage, perform... Calculate the current residual energy of the supercapacitor, where To identify the obtained capacitance, This is the current terminal voltage. The cutoff operating voltage is used to calculate the available operating time before the residual energy is exhausted, and its estimated uncertainty is calculated based on the power consumption corresponding to the current load current. The acquisition unit also maintains an aging count that accumulates with the number of charge and discharge cycles and the operating temperature, which provides initial values ​​and priors for the identification of the capacitance and equivalent series internal resistance.

4. The electricity information collection and metering box according to claim 1, characterized in that, The acquisition unit is configured to read the power outage characteristics recorded by the bus voltage detection unit within a short time window before and after the power outage. The power outage characteristics include whether the three phases drop simultaneously, the rate of drop and the depth of drop of the bus voltage. Simultaneous and rapid drop of the three phases is determined as a simultaneous power outage at the line level or transformer substation level, while single-phase or slow drop is determined as a local power outage. Based on this, a priori estimate of the concurrent contention intensity of this box is given.

5. The electricity information collection and metering box according to claim 4, characterized in that, The acquisition unit is configured to control the wireless communication unit to perform short-time carrier sensing before transmission, measure the channel occupancy rate and received signal strength of the target channel, and count the preamble of neighbor power outage event messages superimposed on the channel. Based on the monotonic variation of channel occupancy rate and collision statistics with the number of concurrent devices when multiple devices transmit concurrently on the radio frequency channel, an observation likelihood with the number of concurrent contending devices as a latent variable is established. The prior estimate, the channel occupancy rate and the preamble count, as well as the historical power outage periods stored in this box and the historical group power outage patterns of the corresponding transformer area are fused in a Bayesian manner to infer the posterior estimate of the concurrent contention intensity and its uncertainty.

6. The electricity information collection and metering box according to claim 4, characterized in that, The acquisition unit is configured to read the three-phase voltage drop waveforms acquired by the bus voltage detection unit, and based on the synchronicity of the drop time and the correlation of the drop slope of the metering boxes on the same feeder in the same tripping event, invert the scale and boundary of the fault subgroup to which the box belongs, so as to distinguish between strong concurrent subgroups that trip on the same feeder and weak concurrent subgroups that only belong to the adjacent domain, and use the scale of the fault subgroup to replace the scale of the entire distribution area to drive the inference of the concurrent contention intensity.

7. The electricity information collection and metering box according to claim 1, characterized in that, The acquisition unit is configured to construct a decision state vector from the residual energy budget and its estimation uncertainty, the currently consumed power outage confirmation waiting time, the AC power supply recovery flag given by the bus voltage detection unit, the posterior estimate and uncertainty of the concurrent contention intensity, the current channel occupancy rate, and the current number of retransmissions, and input the decision state vector into the decision network with a fixed millisecond-level decision step size; the decision network is trained offline using historical or simulated power outage discharge processes, channel collision processes, and delivery results as training samples, with the goal of maximizing the successful delivery probability under the residual energy constraint, and its input and output are physical quantities of residual energy, duration, channel occupancy rate, and concurrent contention intensity.

8. The electricity information collection and metering box according to claim 7, characterized in that, The optional actions include continuing to wait for confirmation, sending immediately and retransmitting with adaptive backoff after sending, and terminating the waiting and sending in a degraded manner when the residual energy budget is expected to be insufficient; the acquisition unit is configured to widen the backoff window width of the adaptive backoff as the posterior estimate of the concurrent contention intensity increases, and when the residual energy budget is insufficient to support the complete power outage event message and its retransmission, the power outage event message is downgraded to a minimum power outage event message containing only the local box identifier and the time of power failure.

9. The electricity information collection and metering box according to claim 7, characterized in that, The acquisition unit is configured to control the wireless communication unit to send a power outage event message according to the decided action and listen for a receipt within a set waiting window. If no receipt is received before the remaining energy budget is exhausted and the remaining energy budget still supports transmission, the unit will re-determine whether to retransmit and back off based on the updated decision state vector. If a receipt is received or the remaining energy budget is insufficient to support retransmission, the process ends. The acquisition unit also records the actual energy consumed, the predicted amount of energy consumed, and the result of whether the delivery was successful after each transmission. This information is used to correct the identification of the capacitance and equivalent series internal resistance, the inference of the concurrent contention intensity, and the parameters of the decision network online.

10. The electricity information collection and metering box according to claim 1, characterized in that, The metering box body is also equipped with a supercapacitor terminal voltage sampling unit, a load current sampling unit, and a temperature sampling unit. The supercapacitor terminal voltage sampling unit is used to collect the terminal voltage of the supercapacitor. The load current sampling unit is used to collect the load current currently drawn by the acquisition unit and the wireless communication unit. The temperature sampling unit is used to collect the current temperature value. The supercapacitor is charged when the AC power supply is normal. The bus voltage detection unit is also used to detect the phase voltage of the AC bus of the metering box body.