Multi-parameter environment sensing and regulation method for solar self-powered outdoor tent

Abstract

The application discloses a solar self-powered outdoor tent multi-parameter environment sensing and regulation method, and relates to the technical field of data processing and operation strategy generation of solar self-powered equipment. The method is as follows: multi-source time sequence data such as external energy supply conditions, power generation output, energy storage state and load state are collected, time alignment verification, resampling, missing treatment and abnormality suppression are carried out on the unified time reference, and data quality indexes are output; a light-power consistency representation item is calculated in a rolling window, and disturbance labels and confidence degrees such as flexible plate deformation, periodic shielding, temporary covering, condensation water film attenuation and load step reaction are generated; the time sequence model is used to predict power generation power, and historical residual errors and confidence degrees are combined to carry out rolling correction, and uncertainty is output; the net available power is converted according to the energy storage, bottom guarantee and link loss constraints, and the load operation strategy instruction is output after conservative correction according to the prediction reliability, so that the power supply stability and key function guarantee capability under complex outdoor working conditions are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing and operation strategy generation technology for solar-powered self-powered devices, and particularly to a method for multi-parameter environmental perception and control of outdoor tents based on solar self-powered power. Background Technology

[0002] Outdoor tents are typically used in outdoor working conditions to provide windproof, rainproof, and private spaces. With increasing outdoor electricity demands, the market has seen the emergence of smart tents or smart tent kits with certain intelligent functions. These typically feature flexible thin-film solar panels and batteries on the top or outer surface of the tent to provide basic power, and can integrate environmental sensors such as temperature, humidity, and light sensors, along with supporting components such as fans and lighting. Some products also support data display and remote on / off control via mobile applications or mini-programs. Furthermore, some solutions attempt to collect multi-source data, process the data, and improve the safety and reliability of smart tents.

[0003] For example, Chinese invention patent CN120162746B discloses a performance evaluation method and device for an automatic and rapid deployment and take-up tent. The method mainly involves collecting a set of performance index test data for the automatic and rapid deployment and take-up tent. This set includes a set of deployment and take-up index test datasets, an efficiency index test dataset, and an environmental adaptability index test dataset. Each index test dataset includes several types of index test data sequences. The performance index test data set is preprocessed to obtain a preprocessed data set. Each index test dataset in the preprocessed data set is then classified and evaluated to obtain a set of sub-index test values. Finally, the sub-index test value sets are fused and evaluated to obtain the performance evaluation result value of the automatic and rapid deployment and take-up tent.

[0004] However, the existing smart tent solutions mentioned above still have significant shortcomings under actual outdoor working conditions: Firstly, flexible thin-film solar panels laid on tent fabric are prone to wrinkling and deformation, wind-induced swaying, temporary covering by equipment or tarps, and slow attenuation due to condensation or water film during outdoor operations. These factors can cause abrupt changes or drifts in the correlation between sunlight readings and power generation. Existing solutions often lack calculable identification and attribution of these disturbance sources, leading to the direct application of prediction results even after structural changes, resulting in inaccurate subsequent tent predictions. For example, in semi-shaded campsites such as woodlands / valleys, tree shadows sway in the wind, causing intermittent shading; in windy environments such as by the sea or mountain passes, the shaking of the tent fabric causes repeated deformation of the flexible panels and localized shading; outdoor workers temporarily placing tarps, curtains, sleeping bags, or equipment on the tent roof can cause sudden shading; and when there is heavy dew in the morning or dampness after rain, a water film forms on the panel surface and persists for a period of time, causing a slow decrease in output power per unit of sunlight.

[0005] Secondly, existing solutions often use power generation as the prediction target and drive control through single-point prediction. They typically do not explicitly consider constraints such as the upper limit of the battery's charge absorption capacity, the reserve power for critical functions, and link loss drift. This makes it difficult to reliably derive the load action from the predicted power generation, and under critical power supply conditions, undervoltage, reset, or frequent load start-stop can easily occur, which is not conducive to accurately predicting the subsequent available power of the tent. For example, in scenarios where the power fluctuates rapidly during the evening when sunlight is rapidly decreasing or during cloudy days with intermittent sunshine, once loads such as fans / heaters are triggered, the combined power demand from lighting, mobile phone charging, etc., may cause the bus to drop instantaneously. In low-temperature environments, the battery's charging and discharging capacity decreases, resulting in situations where there seems to be power but the voltage drops when subjected to a sudden increase in load. In outdoor conditions, frequent plugging and unplugging of USB fast charging, reverse charging of power banks, or multiple people charging simultaneously can exacerbate the step changes in load current, causing control reset or repeated strategy switching.

[0006] Third, inconsistent sampling periods for multi-source data, as well as engineering issues such as wireless link latency, packet loss, and timestamp drift, can amplify the consequences of misjudgments. Existing solutions generally lack data quality metrics, predictive reliability metrics, and risk boundary constraint mechanisms for uncertainty, resulting in insufficient policy stability and interpretability. For example, when temperature and humidity change rapidly inside and outside the tent in rainy weather, or when Bluetooth / wireless links are blocked or interfered with, there may be delays or packet loss in sensor data and load status reporting. When users are far from the tent in the campsite, when mobile phone signals are weak, or when multiple people share the network, there may be delays in the arrival of mini-program commands, which may lead to commands being executed at the wrong time after being delayed, thereby further causing undervoltage or frequent start-stop.

[0007] Therefore, a technical solution is urgently needed to address the above problems: to perform consistency verification and cause attribution on the prediction deviation caused by outdoor operating condition disturbances, to further convert the power generation prediction results into net available power prediction considering energy storage absorption capacity and power supply link loss constraints, and to construct a corresponding conservative availability boundary; at the same time, to generate an operation strategy with degradation and jitter suppression capabilities under the constraints of the availability boundary, thereby improving the stability and executability of power supply and control in outdoor operating conditions. Summary of the Invention

[0008] This invention addresses problems such as inaccurate predictions caused by outdoor working condition disturbances, difficulty in deriving action feasibility from power generation alone, and misjudgment due to unreliable data by employing a multi-parameter environmental perception and control method for outdoor tents based on solar self-powered power.

[0009] To achieve the above-mentioned objectives, the present invention provides the following technical solution: S1: Collect multi-source time-series data and unify them to the same time base, and perform timestamp consistency verification, resampling, missing data handling, and outlier suppression, while outputting data quality indicators. The multi-source time-series data includes at least: first time-series data characterizing external energy supply conditions and second time-series data characterizing power generation output. Furthermore, the multi-source time-series data also includes one or more of the following: third time-series data characterizing load-side power supply status, fourth time-series data characterizing energy storage status, and fifth time-series data characterizing load status.

[0010] S2: Calculate the consistency characterization term within a rolling window, and output the outdoor operating condition disturbance label set and the outdoor operating condition disturbance confidence level based on the consistency characterization term. The consistency characterization term is used at least to characterize abrupt changes or drifts in the correspondence between illumination and power.

[0011] S3: Call the time series prediction model to output the solar power generation prediction sequence for future periods, and perform rolling correction based on historical residuals and corresponding outdoor operating condition disturbance confidence to obtain the corrected power generation prediction sequence, while outputting the prediction uncertainty characterization.

[0012] S4: Based on at least one power supply constraint, convert the corrected power generation forecast sequence into a future net available power forecast sequence. Power supply constraints include at least one or more of the following: energy storage status constraints, critical function backup constraints, and power supply link loss constraints.

[0013] S5: Based on data quality indicators, outdoor operating condition disturbance confidence and prediction uncertainty characterization, obtain prediction credibility assessment results, and conservatively correct the net available power prediction sequence according to the prediction credibility assessment results, generate operation strategy instructions and output them to the load for execution.

[0014] The outdoor operating condition disturbance label set and its confidence level are as follows: The outdoor operating condition disturbance label set includes at least one or more of the following: flexible panel deformation disturbance labels, periodic shading disturbance labels, temporary coverage disturbance labels, condensation or water film slow decay disturbance labels, and load step reaction disturbance labels. The outdoor operating condition disturbance confidence level includes the label confidence level corresponding to each disturbance label in the outdoor operating condition disturbance label set. The label confidence level is jointly determined by the deviation degree and deviation persistence of the consistency characterization item. The deviation degree characterizes the magnitude of the deviation of the consistency characterization item from the historical stable level. The deviation persistence characterizes the continuous occurrence of deviations within multiple rolling windows. The generation of flexible panel deformation disturbance labels involves judging based on morphological evidence of sawtooth fluctuations or periodic fluctuations in power generation output within the rolling window, confirming this by combining it with joint evidence of flexible panel deformation leading to a decrease in the consistency of sunlight and power generation, and then updating the outdoor operating condition disturbance confidence level of the flexible panel deformation disturbance labels upwards. The generation of periodic shading disturbance labels involves identifying disturbances based on evidence of recurring periodic fluctuations in light intensity and power generation output within a rolling window, confirming this by combining evidence of periodic stability or recurrence frequency of these fluctuations, and then updating the outdoor operating condition disturbance confidence level for periodic shading disturbances upwards. The generation of temporary coverage disturbance labels involves identifying disturbances when the change in light intensity is less than a light stability threshold and power generation experiences a sudden change exceeding a power surge threshold, based on joint evidence for temporary coverage. Joint evidence for temporary coverage includes at least one or more of the following: light intensity and power generation consistency falling below a consistency threshold, and the predicted residuals before and after correction exceeding a residual surge threshold. The outdoor operating condition disturbance confidence level for temporary coverage disturbance labels is then updated upwards. The generation of disturbance tags for condensation or water film slow-decaying types involves identifying trends based on a continuous decrease in unit light output capacity across multiple rolling windows, enhancing the trend evidence by combining environmental evidence of indoor humidity and temperature difference between inside and outside the tent, and updating the outdoor operating condition disturbance confidence level for condensation or water film slow-decaying disturbance tags upwards. The generation of disturbance tags for load step reaction types involves identifying evidence based on the coupling strength between load state changes and bus voltage drops, confirming this by combining consequence evidence of prolonged undervoltage recovery time or increased undervoltage event density, and updating the outdoor operating condition disturbance confidence level for reaction-type disturbance tags upwards.

[0015] The above technical solution has at least the following advantages compared with the existing technology: 1. By performing timestamp consistency verification, resampling, missing data handling, and outlier suppression on multi-source time-series data under the same time reference, and outputting data quality indicators such as timestamp alignment error, missing rate, and outlier rate, this technology enables explicit identification and quantitative management of engineering problems such as sensor clock drift, wireless link delay and packet loss, inconsistent sampling periods, and spike saturation jumps. Compared with existing technologies that often use simple filtering or direct data splicing and lack measurement and constraints on data availability, this technology enables subsequent attribution and prediction to automatically switch to a conservative mode when the data deteriorates, thereby reducing the risk of misjudgment and miscontrol.

[0016] 2. By calculating the consistency characterization term within a scrolling window and using it to characterize at least the abrupt changes or drifts in the correspondence between illumination and power, and simultaneously outputting the outdoor operating condition disturbance label set and the outdoor operating condition disturbance confidence, the system enables the calculable identification and cause attribution of disturbances such as wrinkle deformation, periodic occlusion, temporary coverage, slow decay of condensation film, and load step reaction in outdoor operating conditions. Compared with existing technologies that typically only provide abnormal alarms or simple threshold judgments and are difficult to distinguish the source of power anomalies, this system transforms the disturbance type and intensity into a quantitative basis that can be used for subsequent correction and strategy generation, thereby improving the interpretability of the strategy.

[0017] 3. By calling the time series prediction model, the power generation prediction sequence for future periods is output, and rolling correction is performed based on historical residuals and outdoor operating condition disturbance confidence. At the same time, the prediction uncertainty characterization is output, realizing the rapid realignment and risk boundary characterization of the prediction after abrupt changes or drifts in the mapping relationship. Compared with the existing technology, which outputs more single-point predictions and still uses the old mapping after structural changes, resulting in continuous systematic deviations, the prediction can converge in time when state changes such as shading removal, deformation recovery, and condensation disappearance occur, and the optimism level is automatically reduced when the uncertainty increases.

[0018] 4. By converting the corrected power generation prediction sequence into a net available power prediction sequence based on at least one power supply constraint, wherein the power supply constraint includes at least one or more of the following: energy storage state constraint, critical function backup constraint, and power supply link loss constraint, an engineering conversion from generation-side prediction to load-side executable capability is achieved. Compared with the existing technology that directly drives load action with power generation prediction without explicitly considering battery absorption and release capacity and link voltage drop drift, this technology can avoid erroneous triggering of high-energy-consuming loads under conditions such as limited battery capacity at low temperatures, increased cable voltage drop, or contact resistance drift, even if the generation-side prediction is high.

[0019] 5. By obtaining the prediction reliability assessment results based on data quality indicators, outdoor operating condition disturbance confidence, and prediction uncertainty characterization, and then conservatively correcting the net available power prediction sequence accordingly, the system generates operation strategy instructions and outputs them to the load for execution. This achieves safe boundary control and stable strategy output in the face of uncertainty. Compared with existing technologies that lack a unified reliability characterization and conservative boundary contraction mechanism and are prone to undervoltage reset or frequent load start-stop when data is poor and disturbances are strong, this system ensures that critical functions are continuously supplied even under conditions such as link jitter, frequent obstruction, and drastic power fluctuations, reduces strategy jitter, and improves the stability and executability of power supply and control. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the multi-parameter environmental sensing and control method for an outdoor tent based on solar self-powered power, provided in an embodiment of this application. Figure 2 This is a schematic diagram of the composition and data / energy flow of an outdoor tent system provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the time-series changes in prediction confidence and perturbation confidence provided for embodiments of this application; Figure 4 This is a comparative diagram showing the hysteresis threshold and time constraint for suppressing load start-stop jitter in the embodiments of this application. Detailed Implementation

[0022] like Figure 2 As shown, Figure 2This is a schematic diagram illustrating the composition and data / energy flow of an outdoor tent system provided in this embodiment. The outdoor tent in this embodiment can be composed of a tent body, a solar power supply system, an environmental monitoring system, a main control unit, and a load execution unit. The solar power supply system includes at least a flexible thin-film solar panel, a charge / discharge management module, and a battery pack, used to power the electrical equipment inside the tent and output solar-side electrical parameter data and energy storage status data. The environmental monitoring system includes at least one or more of a temperature sensor, humidity sensor, light sensor, and barometric pressure sensor, used to output data related to external power supply conditions. The main control unit is used to receive multi-source time-series data and perform time alignment, disturbance identification, power prediction correction, net available power conversion, and strategy command generation. The load execution unit includes at least one or more of a ventilation fan, lighting device, communication and positioning module, power supply interface, or other controllable loads, used to receive operating strategy commands and execute corresponding control actions.

[0023] Embodiment 1 of the present invention: This embodiment provides a method for multi-parameter environmental sensing and control of an outdoor tent based on solar self-powered power supply, such as... Figure 1 As shown, Figure 1 This is a flowchart illustrating a multi-parameter environmental sensing and control method for an outdoor tent based on solar self-powered electricity, provided in an embodiment of this application. The method's processing flow includes the following steps: S1: Collect multi-source time-series data and unify them to the same time reference. In this embodiment of the invention, the multi-source time-series data includes at least light intensity data characterizing external energy supply conditions, solar-side voltage and current data characterizing power generation output, bus voltage and current data characterizing load-side power supply status, battery state-of-charge data characterizing energy storage absorption and release capabilities, and load status data characterizing load bearing and electricity consumption behavior changes. Light intensity data can be collected by a light sensor; solar-side voltage and current data and bus voltage and current data can be collected by an electrical parameter acquisition module and bound to a timestamp to form an electrical parameter time-series sequence; battery state-of-charge data can be output by a battery management module; and load status data can be obtained from the control status recorded by the main controller. In another embodiment, auxiliary data such as temperature, humidity, wind speed, and attitude can be expanded to improve attribution and prediction accuracy, and unifying them to the same time reference can be achieved using a fixed sampling period. After unifying the time base, timestamp consistency verification, resampling, missing data handling, and outlier suppression are performed on multi-source time-series data. Specifically, firstly, the changing edges are extracted based on the electrical parameter sequences from the solar panel or bus side as alignment anchors. The time offset of each source data relative to the anchors is calculated to complete the timestamp consistency verification, thereby identifying misalignments caused by clock drift or link delays in each source data. Then, each source data is mapped to a unified time axis for resampling to map data from different sampling periods to a unified sampling period. Subsequently, missing segments are identified on the unified time axis, and short and long missing segments are distinguished. Short-term packet loss or disconnection segments are interpolated or conservatively marked. Finally, outliers such as spikes, saturation, and unreasonable jumps are identified and suppressed or replaced to reduce the impact of measurement noise on subsequent consistency characterization calculations and prediction corrections. The timestamp alignment error, missing ratio, and outlier ratio are output as data quality indicators. The timestamp alignment error is obtained by statistically analyzing the time offset of different source data at the same event edge. The missing ratio is obtained by the proportion of missing points within a window to the total number of points within the window. The anomaly ratio is obtained by taking the proportion of the number of anomalies out of the total number of points in the window.

[0024] S2: Calculate consistency characteristics within a rolling window. A rolling window refers to a continuous time segment of data extracted from a unified time axis, with the current time as the reference. It is updated progressively with a step size, and the window length can be determined based on the duration of typical outdoor operating condition disturbances and the system sampling period. Consistency characteristics are summarized by functional expressions that at least characterize abrupt changes or drifts in the relationship between illumination and power. In this embodiment, the consistency characteristics selected are: illumination-power generation consistency characteristics, unit illumination output capability variation characteristics, load variation and bus voltage drop coupling characteristics, and undervoltage event density and undervoltage recovery time characteristics.

[0025] The consistency characterization of illumination and power generation is used to describe whether changes in illumination can explain changes in power generation. The calculation process is as follows: Within a rolling window, the time series sequences of illumination intensity and power generation are obtained. Power generation can be obtained by multiplying the solar-side voltage and solar-side current. To reduce the interference of slow-changing trends on consistency judgment, the changes in adjacent sampling points are calculated for both the illumination intensity sequence and the power generation sequence, resulting in illumination change sequences and power change sequences. The correlation or consistency score between the two sequences is then calculated within the window. In this embodiment, the correlation can be calculated using the Pearson correlation coefficient, Spearman rank correlation coefficient, or normalized cross-correlation peak value. The consistency score can be directly obtained from the correlation, i.e., the correlation is normalized to the range [0, 1] as the consistency score. A higher consistency score indicates a more stable mapping relationship, while a lower consistency score indicates that the mapping relationship may have abruptly changed or drifted. The consistency score of the current window is compared with the consistency baseline range of historical stable operating segments. If the consistency score is lower than the lower bound of the baseline, it is determined that the correspondence between illumination and power has abruptly changed or drifted.

[0026] The unit light output capability variation characterization is used to depict whether the output capability continuously declines under similar light conditions. The specific calculation process is as follows: Within a rolling window, the light intensity sequence and power generation sequence are acquired. The unit light output capability is calculated for each sampling point within the window, which can be characterized by the ratio of power generation to light intensity. To avoid abnormal ratios due to near-zero light intensity, values ​​below a preset lower limit are excluded from the calculation. Robust statistics are performed on the unit light output capability within the window, and a representative value for the window is output. Robust statistics can use the median or truncated mean. The changing trend of the representative value of the window is tracked over multiple consecutive rolling windows. When the representative value of the unit light output capability continuously declines over multiple consecutive windows, and the cumulative decrease exceeds the allowable drift range of the historical stable operating period, a slow decay trend is identified. This is used to characterize the continuous decline in output capability under unit light intensity caused by factors such as condensation or water film. To enhance the reliability of the judgment, when the humidity inside the tent and the temperature difference between inside and outside the tent are collected, the humidity not being lower than the humidity threshold and the temperature difference being within the condensation range can be used as joint environmental evidence. When both the slow decay trend and the joint environmental evidence are satisfied, the validity of the slow decay trend is increased, and a higher upward adjustment is given in the confidence level update. When both the slow decay trend evidence and the joint environmental evidence are satisfied, the upward adjustment step size of the slow decay confidence level is set as the first upward adjustment step size. When only the slow decay trend evidence is satisfied but the joint environmental evidence is not satisfied or is unavailable, the upward adjustment step size of the slow decay confidence level is set as the second upward adjustment step size. The first upward adjustment step size is larger than the second upward adjustment step size. The initial values ​​of the first and second upward adjustment step sizes can be determined based on the fluctuation range of unit light output capacity and noise level of historical stable operation periods. Preferably, the first upward adjustment step size is 0.15 and the second upward adjustment step size is 0.05. Even in the absence of humidity or temperature difference data, a judgment can still be made based solely on trend-based evidence, but the confidence level should be adjusted accordingly to adopt a more conservative strategy.

[0027] The coupling characterization of load changes and bus voltage dips is used to distinguish whether power changes are caused by generator-side factors or by load step reaction. The specific calculation process is as follows: The bus voltage sequence and bus current sequence are obtained within a rolling window, preferably combined with the load switch status or load power level status. First, it is detected whether a step change has occurred on the load side. If load status data exists, the load switch status change or level change is taken as the step event. Load status data refers to the time-series data used to characterize the current operating status of each electrical load within the tent. Load switch status refers to the on / off state of the load power supply; its value can be a binary state (on / off) or an equivalent on / off flag (1 / 0). Load power level status (level): refers to the discrete operating level or power level of the load, which can be represented by an integer code. Level change is an event where the level code changes between adjacent sampling times, typically including fan speed level switching, lighting brightness level switching, heating power level switching, etc., with the corresponding load current or power showing a step change in a short period. If load status data is lacking, a sudden increase in bus current within a short period is used as equivalent evidence of a step event. Subsequently, the magnitude and rate of bus voltage drop are calculated within the time neighborhood of the load step event and matched with the step magnitude or rate of change of the bus current to obtain a coupling strength index characterizing the degree of coupling. Matching includes time matching and amplitude matching. Time matching requires that the time difference between the occurrence of the bus current step event and the start of the bus voltage drop does not exceed a preset alignment threshold. Amplitude matching requires that the ratio of the bus voltage drop magnitude to the bus current step magnitude within this time neighborhood falls within a preset range. When both time matching and amplitude matching are satisfied, the load step and the bus voltage drop are considered to have a strong coupling relationship, and a higher coupling strength index is constructed accordingly. The initial value of the alignment threshold can be determined based on the system sampling period and the step alignment statistics of historical stable operating periods. The initial range of the ratio range threshold can be determined based on the statistical distribution of voltage drop magnitude / current step magnitude in historical stable operating periods. The coupling strength index can be represented using discrete gradations. Specifically, when a current step event and a voltage drop meet the alignment threshold requirements in time and the ratio range threshold requirements in amplitude, the coupling strength index is set to high, such as a value of 2. When only time matching or only amplitude matching is met, the coupling strength index is set to medium, such as a value of 1. When neither time matching nor amplitude matching is met, the coupling strength index is set to low, such as a value of 0. When the bus current step is significant and the bus voltage drops rapidly and synchronously, and the two events occur at time-aligned times, the coupling strength is determined to be high, indicating that the system is in a state of significant load impact or critical power supply. The coupling strength index is compared with the coupling baseline range of historical stable operating periods. When the coupling strength is consistently higher than the baseline threshold, disturbance evidence related to the load step reaction is output first, and this is used as a source of counter-evidence for the confidence adjustment and update of the label of shading or deformation disturbance.Otherwise, if the coupling strength does not consistently exceed the baseline threshold, or only briefly exceeds the baseline threshold within a few windows, the evidence of load step reaction is deemed invalid or of low validity, and is not used as a source of counter-evidence for the confidence upgrade update of occlusion or deformation disturbance labels, or participates in the update with only a small counter-evidence weight; in this case, the confidence upgrade update of occlusion or deformation disturbance labels is executed normally according to the corresponding consistency decline evidence and environmental evidence rules.

[0028] The undervoltage event density and undervoltage recovery time are characterized by the following calculation process: The bus voltage sequence is acquired within a rolling window, and undervoltage events are identified based on a preset undervoltage threshold. An undervoltage event occurs when the bus voltage falls below the undervoltage threshold and remains below the preset minimum duration. The event ends when the bus voltage recovers and remains above the recovery threshold. The recovery threshold can be set by adding hysteresis to the undervoltage threshold. The undervoltage event density is obtained by counting the number of undervoltage events within the window. The undervoltage recovery time is calculated from the time required for each undervoltage event to recover and stabilize above the recovery threshold, and the average, maximum, or quantile value within the window can be used as the representation of the recovery time. When the undervoltage event density increases or the undervoltage recovery time lengthens, the risk of prediction bias will be amplified. Therefore, in subsequent prediction reliability assessments and conservative corrections, the reliability should be reduced and the conservative correction magnitude increased to mitigate the risks of undervoltage, reset, and frequent start-stop. The aforementioned parameters, such as the consistency score threshold, trend drift range, coupling strength threshold, undervoltage threshold, minimum duration, and environmental threshold, can be obtained from the quantile statistics of historical stable operating periods. In the absence of historical data, they can be provided by factory stress tests or empirical initial values. Furthermore, the above-mentioned consistency characteristics are examples of consistency characteristics used to characterize abrupt changes or drifts in the relationship between illumination and power. In actual implementation, one or more of these characteristics can be selected and combined to calculate the consistency characteristics, and are not limited to the types listed in this embodiment.

[0029] After calculating the consistency representation term within a rolling window, the set of outdoor operating condition disturbance labels and the confidence level of the outdoor operating condition disturbance are output based on the consistency representation term. The confidence level of the outdoor operating condition disturbance includes the label confidence level corresponding to each disturbance label in the set of outdoor operating condition disturbance labels. The label confidence level is jointly determined by the deviation degree and deviation persistence of the consistency representation term. The deviation degree is obtained by the deviation magnitude of the consistency representation term relative to the historical stable level, and the deviation persistence duration is obtained by the number of windows in which the deviation occurs consecutively or the duration of the deviation. Optionally, the method for generating the set of outdoor operating condition disturbance labels and their label confidence levels can be implemented by rule discrimination, threshold segmentation, or a confidence level update mechanism based on historical statistics.

[0030] S3: Input the historical observation sequence and auxiliary features under a unified time reference into the time series prediction model, and output the predicted sequence of solar power generation for the future period. The time series prediction model can adopt a model structure that can characterize short-term fluctuations and lag effects, such as a long short-term memory network model, a gated recurrent unit model, or other time series regression models. Specifically, at the current moment, the historical window data consisting of the most recent N sampling points is used as the model input. The input includes at least the historical power generation sequence, and preferably further includes one or more of the following: irradiance sequence, solar side voltage and current sequence, bus voltage and current sequence, battery state of charge sequence, and outdoor operating condition disturbance confidence sequence. The above historical window is input into the time series prediction model to obtain the power generation prediction sequence with a future prediction step size of H. If some auxiliary features are missing, the available historical power generation sequence and irradiance / electrical parameter sequence are used as input to complete the prediction, or conservative padding and missing feature markers are used at the missing points to maintain the consistency of the model input dimensions. The input length N and prediction step size H can be initially determined based on a unified sampling period, the duration of typical outdoor operating condition disturbances, the load step response recovery time, and the minimum hold time / command activation window of the operating strategy. These values ​​can be configured by the user or adaptively adjusted based on historical residuals and undervoltage risk feedback. The historical duration corresponding to N preferably covers 2 to 3 times the duration of the main disturbances, and the prediction duration corresponding to H preferably covers the time scale of the primary strategy decision window or cancellation window. After obtaining the initial power generation prediction sequence, rolling correction is performed based on historical residuals and the confidence level of outdoor operating condition disturbances. Historical residuals are statistical deviations between predicted and actual observed values ​​within the historical window. Rolling correction is used to quickly converge the prediction to the new mapping after changes in the mapping relationship. The confidence level of outdoor operating condition disturbances is used to adjust the correction strength. When the confidence level is high, the correction weight for structural deviations is increased, thereby reducing the risk of systematic misjudgment caused by continuing to use the old mapping after drift. Simultaneously, a prediction uncertainty characterization is output. The prediction uncertainty characterization can be obtained by the dispersion of historical residuals, the width of the confidence interval of the model output, or the quantile prediction difference.

[0031] S4: Based on power supply constraints, the corrected power generation prediction sequence is converted into a future net available power prediction sequence. The power supply constraints include at least one or more of the following: energy storage state constraints, critical function backup constraints, and power supply link loss constraints. In this embodiment of the invention, the corrected power generation prediction sequence is converted into a future net available power prediction sequence based on the battery's absorbable charging power characterization, critical function backup reserved power characterization, and power supply link equivalent loss characterization. The battery's absorbable charging power characterization reflects the charging capacity that the battery can absorb under the current state of charge and environmental conditions; it can be estimated based on the relationship between the state of charge and historical charging current changes. The critical function backup reserved power characterization ensures uninterrupted power supply for critical functions such as security, communication, and basic lighting; it can be determined by historical experience configuration values ​​or user-set constraint values. The power supply link equivalent loss characterization reflects the reduction in available power caused by cable voltage drop, contact resistance drift, etc.; it can be estimated based on bus voltage and bus current combined with the link equivalent impedance parameters, which can be obtained from factory calibration or historical stable operation data statistics.

[0032] In this embodiment, the specific method for converting the corrected power generation forecast sequence into a future net available power forecast sequence is as follows: where, Output of the time series prediction model Obtained through rolling correction, The measured power is used for residual and correction calculations, which can be obtained from the solar side voltage. With current Calculated = . The electrical parameter module can collect data. The electrical parameter module can collect data; to account for link losses caused by cable voltage drop and contact resistance drift, the bus voltage is collected. With bus current and in the scrolling window Inner - ,in, The voltage of the upstream node in the link is at the same voltage level as the bus, preferably the output voltage of the DC / DC converter. If the system does not have a DC / DC converter or both are located at the same electrical node, then... Desirable Construct the equivalent resistance of the link. Where k is the count value of the scrolling window. For the k-th scrolling window. median{ }: Robust median statistics suppress outliers. Only in | Samples with a current value ≥ I_th are included in the statistics to avoid the ratio divergence caused by small currents. I_th is the current threshold: a threshold value used to filter out samples with small currents. When the absolute value of the link current is less than I_t, the voltage drop / current ratio is prone to divergence and noise dominates; therefore, this sample is not included in the estimation of the link equivalent resistance or link loss. This yields the link power loss. And obtain the bus-side deliverable power after considering link losses. =max( - ,0). Among them, For the current flowing through this link, the preferred ratio is PV branch current / cable current / DC / DC output current. When only the bus current is available, it is approximated only under the simplified operating condition where there are no other parallel branches. Replacement.

[0033] To account for energy storage state constraints, the battery state of charge (SOC(t)) and the upper limit of charging power given by the battery management module are read. With upper limit of discharge power . The proportional coefficient representing the "allowed charging" of the battery under the current SOC is used to map the upper limit of charging power given by the battery management module to the actual usable charging limit according to the SOC constraint. The proportional coefficient characterizing the battery's "allowable discharge support" at the current SOC is used to map the upper limit of discharge power given by the battery management module to the actual usable discharge limit according to the SOC constraint. in, For the boundary convergence bandwidth, clip( This indicates that the value is limited to between 0 and 1 to obtain the rechargeable capacity. The battery management module provides the maximum allowable discharge power at time t. = · And based on the target state of charge. ;in, To compensate for bandwidth issues; and to ensure minimum power consumption for critical functions. With safety margin After deduction, the predicted net available power is obtained: a conservative boundary value can be taken. ;in, It can be obtained by summing the critical load ledger; This is used to cover measurement errors and unknown loads. When energy storage discharge support is permitted, reinforced boundaries can be used. Through the above conversion, the predictive capability of the power generation end is transformed into the available capability of the load end, avoiding the problem of undervoltage, reset, and frequent start-stop caused by relying solely on the prediction of power generation to drive regulation, while ignoring the battery absorption capacity and link losses.

[0034] S5: Based on data quality indicators, outdoor operating condition disturbance confidence levels, and prediction uncertainty characteristics, the prediction reliability assessment results are obtained, and the net available power prediction sequence is conservatively corrected according to the prediction reliability assessment results. Conservative correction is used to reduce the optimistic bias of the prediction, which can be achieved by selecting conservative coefficients or conservative quantiles. This ensures that when data quality deteriorates, outdoor operating condition disturbance confidence levels increase, or prediction uncertainty increases, the net available power prediction converges to a more conservative availability boundary. Based on the conservatively corrected net available power prediction, operating strategy instructions are generated and output to the load for execution. The operating strategy instructions not only include the start / stop and power level control of ventilation, lighting, or other electrical loads, but also include executable constraint parameters for power supply boundary fluctuations in outdoor operating conditions, ensuring the instructions are effective and do not trigger undervoltage resets or frequent start / stop operations.

[0035] In this embodiment, the operation strategy instruction includes at least one or more of the following fields: load object identifier, control action type, target power level or duty cycle, instruction activation window, minimum hold duration, cooling time, start threshold, stop threshold, segmented start parameters, concurrency limit parameters, critical function safety net identifier, degraded level and cancel window parameters. The load object includes at least ventilation fans, tent lighting, heating modules, communication and positioning modules, USB power ports, or other controllable electrical equipment; the control action type includes at least start, stop, degrade, delayed execution, and segmented upgrade. The instruction fields are obtained as follows: the load object identifier and target power level are obtained from the load ledger and load capacity table in the main control firmware; wherein, the load ledger is a pre-set load list and its metadata table in the main control firmware, used to record the object identifier, control channel identifier, and control method information of each type of controllable load; the load capacity table is a capacity parameter table associated with the load object identifier, used to record the set of power levels supported by the load and the target duty cycle / switching duration / rated power range, etc., corresponding to each level. When generating strategy instructions, the main controller first determines the target load's object identifier and control channel based on the load ledger. Then, it queries the available power level set for the load in the load capacity table and selects the target power level that meets the constraints by combining the net available power prediction value and concurrency limit rules, forming an instruction field of "load object identifier + target power level". The ledger and capacity table can be written to the firmware by factory calibration or set by the user in the configuration interface and then sent to the main controller for storage and updating. The minimum hold time and cooling time are configured based on the load's physical characteristics and outdoor operating conditions, or obtained by statistical analysis of frequent start-stop jitter segments in historical operating data and rolled correction. The start-up threshold and stop-down threshold are at least based on the conservatively corrected net available power prediction sequence, the critical function's backup power reserve, and undervoltage sample statistics. The segmented start-up and concurrency limit parameters are at least based on the bus voltage recovery characteristics, load start-up surge characteristics, and historical undervoltage event statistics. The cancellation window parameter is at least used to promptly withdraw high-risk actions when the prediction confidence decreases or the disturbance confidence increases, so that the control strategy remains stable and interpretable.

[0036] By expanding the operation strategy instructions from a single start / stop command to a set of instructions that include constraint parameters such as threshold, timing, concurrency, and degradation, the load control can automatically converge to a more conservative action intensity and suppress frequent start / stop when data quality deteriorates, outdoor operating condition disturbance confidence increases, or prediction uncertainty increases. This reduces the risk of undervoltage reset and improves the executability of power supply control in outdoor operating conditions.

[0037] Embodiment 2 of the present invention: Based on Embodiment 1, this embodiment further provides a method for generating the outdoor operating condition disturbance label set and the outdoor operating condition disturbance confidence level, which is used to transform the cause of inaccurate prediction from a qualitative description into a calculable, verifiable and correctable attribution result, thereby supporting subsequent rolling correction and net available power conversion.

[0038] After calculating the consistency characterization term, an outdoor operating condition disturbance label set and an outdoor operating condition disturbance confidence score are output based on the consistency characterization term. The outdoor operating condition disturbance label set includes at least one or more of the following: flexible panel deformation disturbance, periodic shading disturbance, temporary coverage disturbance, condensation or water film slowly decaying disturbance, and load step reaction disturbance. In this embodiment, the outdoor operating condition disturbance confidence score is used to quantify the credibility of flexible panel deformation disturbances. The outdoor operating condition disturbance confidence score includes the label confidence score corresponding to each disturbance label in the outdoor operating condition disturbance label set. The label confidence score is jointly determined by the deviation degree and deviation persistence of the consistency characterization term. The label confidence score corresponding to each disturbance label is obtained using an executable update rule. For any perturbation label, the evidence satisfaction status corresponding to that label is first calculated within the current rolling window, and the evidence is broken down into deviation degree and deviation persistence. Deviation degree characterizes the magnitude of deviation of the label's related consistency indicators from the historical stable level, and can be obtained by the amount or proportion by which the current window statistic exceeds the baseline allowable range. Deviation persistence characterizes the continuous occurrence of deviations, and can be obtained by counting windows that continuously satisfy the deviation condition or by the duration of such occurrences. Based on this, this embodiment performs window-by-window updates on the label confidence: when the evidence for that label is valid in the current window, the label confidence is updated upwards; when the counter-evidence condition for that label is valid or subsequent windows return to the stable baseline, the label confidence is updated downwards or suppressed. The magnitude of the upward update increases at least with the deviation degree factor and at least with the deviation persistence factor; the magnitude of the suppressed or downward update increases at least with the counter-evidence strength. To ensure stable confidence output and avoid single-window noise amplification, this embodiment sets upper and lower bounds for the label confidence and employs smoothing constraints. The label confidence level can be represented by a normalized numerical value, ranging from 0 to 1, with the initial value being the baseline confidence level. The baseline confidence level, upward adjustment threshold, downward adjustment threshold, and counter-evidence threshold can be obtained from statistics of historical stable operating periods. The specific generation of the outdoor operating condition disturbance label set and the corresponding confidence level update of the outdoor operating condition disturbance labels are as follows: To generate flexible panel deformation disturbance labels, this embodiment extracts morphological evidence from a scrolling window. Morphological evidence characterizes whether the power generation output waveform exhibits a typical deformation pattern, including at least one of the following: sawtooth fluctuations, repetitive undulations, or quasi-periodic undulations in the power generation output within the window; fluctuations occur repeatedly within a short period, and the fluctuation rhythm is inconsistent with the illumination change rhythm. The judgment parameters such as sawtooth fluctuation intensity, fluctuation amplitude, and fluctuation period stability are not required to be fixed and can be achieved using a baseline comparison method: the fluctuation amplitude, fluctuation amplitude, and period stability within the current window are compared with the corresponding statistics of historical stable operating segments; if they exceed the baseline range, the morphological evidence is considered valid. The baseline range can be a typical interval of the statistical distribution of stable segments, such as the allowable fluctuation band near the quantile interval or mean; in the absence of historical data, the initial range obtained from factory pressure testing can be used as the baseline, and replaced with the field statistical baseline after accumulating sufficient stable samples. The combined evidence of flexible plate deformation is used to eliminate confounding factors such as overall changes in external illumination and the reaction effect of load step changes. It includes at least the following two types of combined criteria: Decreased consistency between illumination and power generation: manifested as small changes in illumination but significant power fluctuations, or a significant decrease in the correlation between illumination and power within a window; Exclusion of power generation step evidence: the load state on the bus side does not show a step change matching the fluctuation in power generation output, and there is no significant simultaneous occurrence of a rapid voltage drop and a current step on the bus side, used to exclude power supply fluctuations caused by sudden load increases. When both morphological evidence and combined evidence of flexible plate deformation are valid, it is determined that both morphological evidence and combined evidence of flexible plate deformation are simultaneously satisfied, generating a flexible plate deformation-type disturbance label, and updating the confidence level of the flexible plate deformation disturbance corresponding to this label upwards. Upward adjustments should comprehensively consider at least the following: Morphological evidence strength: graded as weak, medium, and strong, with grading boundaries provided by stable segment statistics or factory stress tests and updated on a rolling basis; Consistency decline magnitude: graded based on the magnitude of deviation of consistency characteristics from the stable level, with larger deviations resulting in higher grades; Number of consecutive satisfaction windows: recording the number of consecutive satisfactions, with adjustments made based on the baseline when a threshold is reached, and enhanced adjustments for higher thresholds. The thresholds are determined by historical deformation event duration statistics or empirical initial values ​​and are subject to rolling correction. Simultaneously, upward adjustments can be achieved through incremental layering at each grade level: morphological evidence strength determines the baseline adjustment magnitude, consistency decline magnitude amplifies or reduces it, and the number of consecutive satisfaction windows cumulatively enhances the upward adjustment, thereby increasing the confidence level of flexible plate deformation perturbation as the strength and persistence of evidence increase.To avoid misattribution, this embodiment sets up a suppression and fallback mechanism: when a load state step is detected and coincides with a rapid drop in bus voltage or a bus current step, it is determined that there is a load step counter-evidence. Then, the confidence level of the deformation disturbance of the flexible plate is suppressed, including no longer increasing or falling back the already increased portion. The fallback magnitude increases with the strength of the counter-evidence. The counter-evidence threshold is obtained from historical undervoltage samples, factory voltage tests, or on-site statistics. When the waveform recovers to the allowable range of the stable baseline within the subsequent window, and the consistency recovers to near the stable level and continues to meet the recovery window number, it is determined that the disturbance has subsided. The confidence level of the flexible plate deformation disturbance is updated by attenuation and fallback to make it smoothly return to the baseline. The recovery window number is determined by historical recovery process statistics or empirical initial values ​​and is adaptively corrected.

[0039] Periodic shading disturbance labels are used to characterize intermittent shading caused by tree shadow swaying or wind-induced swaying. Evidence of periodic fluctuations in light intensity and power output is extracted within a rolling window, including at least: repeated fluctuations in both light intensity and power output, with these fluctuations recurring within the window; and estimation of the stability or recurrence frequency of the fluctuation period to distinguish between occasional noise and periodic disturbances. Periodic stability can be determined using baseline comparison, for example, by statistically analyzing the "natural fluctuation period range" of the stable segment; if the fluctuation exceeds the natural fluctuation range and recurs, periodic evidence is considered valid. To avoid misjudging deformation or load steps as periodic shading, this embodiment further requires joint evidence of periodic shading to meet one of the following criteria: light intensity and power output are synchronous or have similar rhythms in terms of period, while typical sawtooth evidence of deformation is not dominant; and there are no corresponding periodic repeated steps in the load state on the bus side, to reduce interference caused by periodic load start-stop cycles. When evidence of periodic fluctuations is established and confirmed by joint evidence of periodic occlusion, a periodic occlusion perturbation label is generated, and the confidence level of the periodic occlusion perturbation is increased based on the following factors: the more times the period repeats and the more stable the period, the greater the increase; the more significant the decrease in illumination and power consistency and the more persistent the window, the greater the increase. When subsequent windows periodically disappear and a stable baseline is restored, a fallback update is performed; when counter-evidence of load periodic start-stop is detected, a suppression or rollback update is performed.

[0040] Temporary coverage disturbance tags are used to characterize sudden shading caused by canopies, curtains, sleeping bags, or equipment placement. Their typical characteristics include stable illumination changes but abrupt changes in power generation output, leading to a short-term surge in prediction residuals. To generate temporary coverage disturbance tags, this embodiment extracts joint evidence of temporary coverage based on stable illumination changes and abrupt changes in power generation output within a rolling window. This joint evidence includes at least one of the following: illumination change amplitude is below the allowable range of the stable period or below the shading sensitivity threshold, but power generation output experiences a step drop within a short period and continues for more than the minimum duration; the consistency between illumination and power generation is significantly reduced within the short window; and a surge in prediction residuals before and after correction, with the surge in residuals and the power change occurring at a temporal coincidence. The illumination stability threshold, power change amplitude threshold, and minimum duration can be obtained from historical shading events; in the absence of samples, they are given by factory calibration or empirical initial values ​​and are updated on a rolling basis during subsequent operation. When joint evidence of temporary coverage is established, a temporary coverage disturbance label is generated, and the confidence level of this label is increased. The confidence level increase should increase at least with the following factors: larger and longer-lasting power surges; larger decreases in consistency; and more pronounced residual surges that are better aligned with the power surges. When the power recovers in subsequent windows and the consistency returns to near the baseline, a fallback update is performed. If counter-evidence of a load surge causing a drop in bus voltage is detected, the confidence level of this label is suppressed or rolled back.

[0041] Condensation or water film slow-decay disturbance tags are used to characterize the slow decrease in output capacity per unit of illumination caused by the formation of a water film on the plate surface. Typical characteristics include trend, slow variability, and persistence. To generate condensation or water film slow-decay disturbance tags, this embodiment extracts trend evidence and combined environmental evidence from a rolling window. Trend evidence of a continuously decreasing output capacity per unit of illumination is extracted within multiple consecutive rolling windows, including at least: under similar illumination conditions, the output capacity per unit of illumination continuously decreases as the window progresses, and the decrease exceeds the allowable drift range of the stable segment; the decrease process is continuous rather than a single-window abrupt change, to distinguish it from temporary coverage disturbances. The trend threshold can be obtained statistically from historical operating data, such as statistically analyzing the slow variability range caused by normal aging / temperature changes; exceeding this range is considered valid trend evidence. When historical data is lacking, initial values ​​can be calibrated using the factory environment and corrected during rolling operation. To improve the reliability of the judgment, this embodiment combines environmental evidence such as indoor humidity and temperature difference between the inside and outside of the tent for enhancement: when the humidity is not lower than the humidity threshold and the temperature difference is within the condensation-prone range, the trend evidence is enhanced, resulting in a larger increase in the confidence level of the label; if there is no humidity or temperature difference data, the label can still be generated based solely on trend evidence, but the confidence level of the label will be increased more conservatively. The humidity threshold and temperature difference range can be obtained from empirical values, environmental calibration, or user configuration, and will be continuously corrected during operation based on actual false alarms. The more obvious the trend, the more continuous the window, and the more sufficient the environmental evidence, the greater the increase in the confidence level of the label; when the humidity decreases, the temperature difference changes leave the condensation-prone range, and the unit light output capacity returns to a stable trend, a decay reduction is performed.

[0042] Load step reaction disturbance tags are used to characterize sudden load activation or multiple load stacking leading to bus voltage drops, passive power derating on the generator side, or power supply fluctuations, avoiding misattribution to obstruction or deformation. To generate load step reaction disturbance tags, this embodiment extracts coupling strength evidence and consequence-based evidence within a rolling window. The coupling strength evidence between load changes and bus voltage drops is extracted within the rolling window, including at least: a step change in load status, or a bus current change rate exceeding a threshold; and a bus voltage drop rate exceeding a threshold, with both occurring at consistent times. Thresholds can be obtained from factory voltage test data or historical undervoltage samples and can be maintained separately for different load configurations. To enhance the reliability of the judgment, this embodiment combines consequence-based evidence such as prolonged undervoltage recovery time or increased undervoltage event density. When undervoltage events occur more frequently within the window, or the time required to recover to a stable voltage is significantly prolonged, the confidence level of the load step reaction tag is increased; consequence-based evidence is used to distinguish between normal load fluctuations and significant reactions caused by power supply criticalities. When the evidence of coupling strength is established and confirmed by consequential evidence, a load step reaction label is generated and the confidence level is increased.

[0043] In real-world outdoor scenarios, multiple disturbances may coexist. This embodiment allows multiple labels and their confidence levels to be output in the same window, while setting conflict handling principles: if a temporary coverage label and a periodic occlusion label are both valid, the label with stronger evidence of abrupt change receives a higher confidence level; if the confidence level of a load step reaction label is significantly higher, occlusion / deformation labels are suppressed or their confirmation is delayed; if a slow condensation change label is valid, it usually represents long-term trend evidence and can coexist with other short-term labels, but different correction weight strategies are used during rolling correction. Simultaneously, this embodiment uses this label as a source of counter-evidence for other labels: when the confidence level of a load step reaction label is high, the confidence level of deformation and occlusion labels can be suppressed to avoid misattribution.

[0044] On the unified time axis, let t represent the sampling time (unit: s, given by the master control unified sampling time axis), let... This represents the k-th scrolling window, determined by the window length. With sliding step size Δ Generate, where k is the window counting index. Bus voltage (t) (unit: V) and bus current (t) (unit: A) are all obtained by the electrical parameter acquisition module on the bus side; photovoltaic side voltage (t) (unit: V) was obtained by the electrical parameter acquisition module on the photovoltaic side. (In the window...) Within this framework, we define the rate of change of bus current and the rate of change of bus voltage (in discrete differential form for ease of engineering implementation): (t)= (t)- (t-Δ ), ;where Δ To standardize the sampling period (unit: seconds, set by resampling, e.g., 1 second), a window coupling strength index is constructed to measure the degree of synchronization of "load step causing bus drop". For example, take the normalized combination of the maximum negative voltage change and the corresponding current change within the window: ;in, The normalized reference value for bus voltage (unit: V) can be taken as the rated bus voltage or the median value of the stable section. The reference value for current normalization (unit: A) can be taken as a typical load current or the median value of the steady-state range; clip(x) represents the cutoff calculation that restricts x to [0, 1], and is defined as clip(x) = min(max(x, 0), 1). ≥ At that time, the evidence for the coupling strength of "load step - bus drop" was determined to be valid, among which The coupling threshold can be obtained from factory voltage tests, historical undervoltage samples, or statistical analysis of stable periods in the field, and can be continuously corrected. (In the window...) Within, with undervoltage threshold Unit: V, given by policy configuration or factory calibration. Identify undervoltage events: when... And lasting longer than the minimum duration Each undervoltage event is recorded as one undervoltage event. Let the number of undervoltage events within the window be . And statistically analyze robust representative values ​​of undervoltage recovery time. Unit: s, for example, to take the maximum value or quantile value, then when the following conditions are met: or In such cases, consequential evidence is considered valid. Among them, (number of times threshold) and The (duration threshold) can be obtained from factory voltage testing or statistical analysis of historical undervoltage samples. When both the coupling strength evidence and the consequence evidence are valid, the load step reaction is considered to occur within the window. The internal structure serves as proof against the misattribution of occlusion / deformation. The flag for proof by contradiction is 1, otherwise 0. Let the overall confidence level of the mapping drift-type perturbation such as occlusion / deformation at time t be . The range is 0–1, obtained by weighted synthesis of the confidence scores of each occlusion / deformation label, and the confidence score of the load step reaction is... (Range 0–1, obtained from Example 2 / Load Reaction Label Update). Define the effective mapping drift confidence after proof by contradiction suppression: );in: The maximum value is the counter-evidence suppression coefficient (suggested to be between 0 and 1), used to control the strength of the counter-evidence; max( ) is the maximum value operator, ensuring Not negative. The rolling correction in S3 and the prediction confidence assessment in S5 will preferentially adopt [the following method / approach]. replace As input, this reduces the risk of miscorrection and malfunction when the bus drop is mainly caused by a sudden increase in load.

[0045] To further reduce the risk of miscorrection, this embodiment freezes or limits the "mapping update" when the proof of contradiction is significant. Let α(k) be the weighting coefficient used in S3 to update the residual baseline / correction strength, ranging from 0 to 1, and let this symbol represent the correction update strength only within this proof of contradiction section. When the proof of contradiction is valid and the load reaction confidence exceeds the proof of contradiction threshold... When executing freeze or limit rules: The strongest suppression freeze rule: during continuous... Setting α(k) = 0 within a window indicates that new residuals are not absorbed into the mapping update temporarily, avoiding the learning of load consequences as generation mapping drift during load shocks. A weaker, smoother limiting rule: in continuous... Within each window, let α(k) = min(α(k), );in, The number of continuous freeze / limiting windows is obtained from factory voltage tests or historical undervoltage samples, and can also be adjusted by feedback rolling correction based on strategy jitter caused by on-site miscalibration. To trigger the confidence threshold for a rebuttal freeze, historical samples can be used to identify windows with significant load impact. Quantile values ​​are determined. A low upper limit is used to restrict the update rate without completely freezing it. Unfreezing / limiting and resuming normal mapping updates will occur when one of the following conditions is met: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Within each window (t)≤ , The release threshold is typically less than This creates hysteresis to avoid frequent switching; coupling strength index Falling back to The following are the consequences of undervoltage indicators It returns to a stable range. The number of persistent windows is used to ensure the persistence of counter-evidence fading and to avoid premature fading due to single-window noise.

[0046] Through the aforementioned mechanism of counter-evidence determination—suppression effect—intensity and duration—relief condition, the load step reaction can serve as a source of counter-evidence for misattribution of obstruction / deformation at the power supply critical point: on the one hand, it suppresses... On the one hand, the upward adjustment and the driving force of the correction, on the other hand, the freezing or limiting of the mapping update, thereby reducing the risk of miscorrection and malfunction when the consequences of load shock or undervoltage are significant, and improving the stability of prediction correction and strategy output in outdoor working conditions.

[0047] Embodiment 3 of the present invention: Based on the outdoor working condition disturbance label set and its confidence level output in Embodiment 2, this embodiment further provides an executable calculation rule for the rolling correction intensity, so that the prediction can quickly converge to the new mapping when the illumination-power mapping relationship changes abruptly or drifts, and reduce the risk of systematic misjudgment caused by using the old mapping.

[0048] The time-series prediction model can employ Long Short-Term Memory (LSTM) networks, gated recurrent units (GRUs), or other time-series regression models. Training data must originate from at least the following time-series data: historical illumination, solar-side electrical parameters, bus-side electrical parameters, battery status, and load status. Training can be grouped or incrementally trained according to outdoor working conditions and season. Historical residuals are statistically obtained from the deviation between predicted and actual observed values ​​within a historical window. To avoid errors being amplified by occasional noise, this embodiment aggregates residuals by window and maintains stable segment residuals and disturbed segment residuals separately: stable segment residuals are used to update the normal baseline; disturbed segment residuals are used to trigger rapid correction, but the correction intensity must be controlled in conjunction with the disturbed confidence level to prevent over-correction during misattribution. Rolling correction involves updating historical residuals with weights based on the set of outdoor operating condition disturbance labels and their corresponding confidence levels. The weighting coefficients increase with increasing confidence levels, and include at least the following rules: when the confidence level of a certain type of disturbance label is high, the weight of the residuals in that window is increased in the update, enabling the correction to respond more quickly to sudden changes or drift; when data quality indicators are poor or there is a counter-evidence of load step reaction, the upper limit of the correction weight is limited or the correction is delayed to suppress oscillations caused by misjudgments; when the disturbance subsides and the confidence level falls, the correction weight is gradually reduced to make the prediction smooth regression stable mapping. The initial value and upper limit of the weighting coefficients can be determined by historical operation statistics; when statistical data is lacking, empirical initial values ​​can be used, and rolling adjustments can be made during operation based on whether the corrected residuals converge and whether the strategy jitters.

[0049] This embodiment presents a residual baseline update and correction output method based on a rolling window. On a unified time axis, let the sampling time be t, and let the initial power generation prediction value output by the time series prediction model be... (t), unit: W, output by the model. Actual observed power generation is... (t), unit: W, obtained from photovoltaic side electrical parameters. For example, from photovoltaic side voltage. With photovoltaic side current . Define the instantaneous residual as e(t) = (t)- (t); robust aggregation of the residuals within the window yields the window residual statistic. To reduce the impact of spike noise: =Agg{e(t)∣t∈ }; where Agg{ } is a robust aggregation operator, which can be selected from the window median, truncated mean, or Huber robust mean; for example, when using the median, Agg{ }=median{ }. The confidence level of the outdoor operating condition disturbance is set in the window. The representative value on is denoted as The range is 0–1, and it can take the mean or maximum value within the window. It can be synthesized from the confidence scores of "mapping drift" labels such as occlusion / deformation / condensation. To enable faster absorption of structural biases when the mapping relationship changes abruptly or drifts, this embodiment defines window correction weights. For follow Monotonically increasing functions, for example: ;in: The minimum correction weight is used to ensure slow self-correction even under low disturbances; The maximum correction weight is used to ensure fast convergence under high disturbances; This is the dynamic upper limit of the k-th window, used to limit overcorrection when data quality is poor or there is a load step counter-proof. In this embodiment, the minimum value of the window correction weight is... The initial configuration uses a target convergence window size approach and allows for adaptive updates based on statistics during runtime. This mapping ensures that the confidence level is affected by outdoor operating condition disturbances. At higher levels, Increase the weighting, thereby "improving the correction weight for structural biases," and avoiding systematic misjudgments caused by continuing to use the old mapping after drift; when At lower levels, near The correction is smoother, avoiding noise-triggered oscillations.

[0050] This embodiment maintains the residual baseline. , representing the slowly varying baseline of the prediction bias under the current mapping relationship, and updated using exponential smoothing: + In the window Within this period, the power forecast for the future time period is corrected to obtain the corrected forecast value. : Under this update framework, when mapping mutations / drifts cause the window residuals to continuously deviate, will be by The controlled speed absorbs this deviation, thus enabling the prediction to converge quickly to the new mapping. To prevent the error from being amplified when data is missing, alignment errors are large, or load step reaction is dominant, this embodiment... Introducing dynamic upper limit Let the data quality score of the k-th window be . It can be obtained by weighting the normalized timestamp alignment error, missing proportion, and outlier proportion; a representative value for the load step reaction confidence window is set as follows: Output by the load reaction label. Then it can be defined as: Where η is the inhibition coefficient, and This is used to control the strength of the suppression of mapping updates by load counteraction. When data quality deteriorates or the confidence level of load counteraction increases, the dynamic upper limit automatically decreases, making... Even in Even at higher levels, the correction will not be over-amplified. In the absence of historical statistics, The desired convergence window number can be set: Let's assume we want approximately [number missing] convergence windows under steady operating conditions. Each window absorbs most of the baseline deviation, approximately [percentage missing] under high disturbances. If the window converges quickly, then it is acceptable: =2 / ( +1), =2 / ( +1); where, > When historical data is available, it can be... The quantile statistics of the number of windows lasting as a typical occlusion / deformation event are taken. The statistical result of the number of drift fading windows in the stable segment is taken. During operation, the system can adjust the settings based on whether the corrected residuals converge and whether the strategy jitters. Or η performs scroll adjustments: for example, in a window Calculate the corrected window residuals =Agg{ - }, when | | When the voltage continues to decrease and the density of undervoltage events does not increase, the voltage level can be appropriately increased. Or reduce η to speed up convergence; when | | If the load starts and stops more frequently or the density of undervoltage events increases instead of decreasing, then reduce the load. Alternatively, η can be increased to suppress strategy jitter caused by overcorrection. After performing rolling correction, the corrected power generation forecast sequence is obtained, and a characterization of forecast uncertainty is output. Uncertainty can be derived from the dispersion of historical residuals, the range of recent residual fluctuations, or quantile forecast differences. If the data quality index deteriorates or the perturbation confidence increases, the uncertainty characterization will increase accordingly.

[0051] Embodiment 4 of the present invention: Based on S5 of Embodiment 1, this embodiment provides a detailed implementation of prediction reliability assessment and conservative correction, so that the net available power prediction can automatically converge to a safer availability boundary when uncertainty increases, and reduces the risk of undervoltage, reset and frequent start-stop at the load execution level.

[0052] The prediction reliability assessment should comprehensively consider at least the following inputs: data quality indicators: timestamp alignment error, missing proportion, and outlier proportion; outdoor operating condition disturbance confidence: confidence of each label and its persistence; and prediction uncertainty characterization: recent residual fluctuation range or interval width. The assessment principle is: when data quality deteriorates, disturbance confidence increases, or uncertainty increases, prediction reliability decreases, and the conservative correction magnitude increases. In this embodiment, the prediction reliability assessment is implemented using a window reliability scoring method. The window reliability score ranges from zero to one; a larger value indicates a more reliable prediction within the rolling window, and a smaller subsequent conservative correction magnitude; a smaller value indicates a less reliable prediction within the rolling window, and a larger subsequent conservative correction magnitude. Figure 3 As shown, when the confidence level of mapping drift-type disturbances increases during the disturbance occurrence phase, the window confidence score decreases accordingly; during the disturbance resolution phase, after the disturbance confidence level falls back and data quality recovers, the window confidence score gradually recovers. The window confidence score is used to drive the hierarchical selection of conservative margins, so that the safe and executable net available power boundary automatically tightens when the confidence level decreases, and gradually widens when the confidence level recovers. For each rolling window, three types of data quality indicators are calculated: Timestamp alignment error: refers to the statistical value of the time offset of each source data near the anchor point when multiple source data are aligned to the same alignment anchor point. The alignment anchor point can be selected as event points that are easy to occur synchronously in multiple source data, such as bus current sudden change points, bus voltage sudden drop points, and load switch state change points; the time offset can be obtained by statistically analyzing the difference between the occurrence times of the corresponding events of each source data. Missing ratio is the proportion of the number of missing sampling points in the window to the total number of sampling points in the window. The anomaly ratio is the proportion of the number of points within the window that are judged as abnormal, such as spikes, saturation, out-of-bounds, or unreasonable jumps, to the total number of sampling points within the window.

[0053] This embodiment pre-configures corresponding thresholds for the three indicators mentioned above. The thresholds can be derived from factory stress test recommendations. The scoring rules are as follows: when the timestamp alignment error is less than a multiple of the threshold, the missing rate is less than a multiple of the threshold, and the anomaly rate is less than a multiple of the threshold, the data quality is considered good, and the data quality score is set to the largest value; when any indicator exceeds the threshold, the data quality is considered to have decreased, and the data quality score is reduced accordingly. This embodiment scores the three indicators separately and then averages the scores: each indicator is scored as full marks when it is better than the threshold, and as zero marks when it is significantly worse than the threshold, decreasing linearly according to the degree of deviation. Finally, the average of the three scores is taken to obtain the data quality score.

[0054] For each rolling window, the confidence scores for various disturbances are output from step two and normalized to zero to one. In this embodiment, the comprehensive confidence score for mapping drift is defined as the largest among the four confidence scores: flexible plate deformation, periodic shading, temporary coverage, and slow decay of condensation or water film. Subsequently, a counter-evidence suppression method for load step reaction disturbances is introduced. The counter-evidence determination uses a threshold plus persistence approach: when the load step reaction confidence score is higher than a preset counter-evidence threshold and continuously reaches a preset number of windows, the counter-evidence is determined to be valid; when the load step reaction confidence score falls below the threshold and remains below it for a certain number of windows, and the bus undervoltage recovery characteristics return to normal, the counter-evidence is determined to be resolved. The counter-evidence threshold and the number of consecutive windows can be determined by factory pressure testing, historical undervoltage sample statistics, or on-site operation statistics, and rolling corrections are allowed based on false alarms and missed alarms. The counter-evidence mechanism operates in two levels: suppression and freeze / delay. Suppression rule: When counter-evidence is valid, the overall confidence level for mapping drift needs to be reduced by a preset suppression strength. The suppression strength is determined by a configurable suppression coefficient; the larger the coefficient, the stronger the suppression of drift conclusions such as shading / deformation by the load reaction. When the result after reduction is less than zero, it is counted as zero. Subsequent rolling corrections and confidence assessments prioritize the effective confidence level after suppression to avoid miscorrection and erroneous actions. Freeze / delay rule: When the confidence level of the load step reaction exceeds the counter-evidence threshold and continues to meet stricter continuous window requirements, this embodiment freezes or delays the "mapping update and correction strength": within a preset freeze window period, the correction update strength is limited to an extremely low level or updates are directly suspended; the freeze is lifted only after the load reaction confidence level falls back and the bus undervoltage recovery characteristics return to normal. Both the freeze window period and the lifting conditions are configurable, and the sources can be load testing or statistical experience, with rolling corrections allowed.

[0055] For each rolling window, the prediction uncertainty is given using available data, such as the dispersion of recent residuals, the quantile difference of the absolute value of residuals, or the width of the confidence interval of the model output. A threshold is configured for the uncertainty. When the uncertainty is significantly lower than the threshold, the prediction fluctuation range is considered small; when the uncertainty exceeds the threshold, the prediction fluctuation range is considered large, the confidence level should be reduced, and the conservative correction should be increased. The uncertainty threshold can be determined by factory stress testing or historical stable period statistics, and rolling correction is allowed. In this embodiment, the data quality score is directly proportional to the window confidence score; the higher the effective confidence level of the suppressed mapping drift, the lower the window confidence score; the greater the prediction uncertainty, the lower the window confidence score. This embodiment uses a fixed-weighted synthesis method and limits the results to the range of zero to one; the weights are configurable, and by default, the data quality proportion can be slightly higher to prioritize avoiding aggressive decisions when the data is unreliable. For the net available power prediction sequence obtained in step four, this embodiment introduces a conservative margin, which means an additional safety power reserve to cover measurement errors, mapping drift, and uncertainty. The conservative margin is determined using a tiered rule: a high-confidence threshold and a low-confidence threshold are set. The minimum conservative margin is applied when the window confidence score is higher than the high-confidence threshold; a medium conservative margin is applied when the window confidence score is between the two thresholds; and the maximum conservative margin is applied when the window confidence score is lower than the low-confidence threshold. All three conservative margins are configurable values, obtainable from factory voltage tests or historical undervoltage samples, and matched with the equipment's rated power and bus drop sensitivity. The safe and executable net available power boundary is obtained by subtracting the corresponding conservative margin from the predicted net available power value, with a lower limit not less than zero. This ensures that when data quality deteriorates, a strong drift risk remains even after counter-evidence suppression, or prediction uncertainty increases, the window confidence score automatically decreases and the conservative margin automatically increases, causing the safe and executable net available power boundary to automatically converge to a more conservative range. This reduces the risks of undervoltage, reset, and frequent start-stop, and improves the uninterrupted supply capability of critical functions.

[0056] Conservative correction involves selecting a conservative correction magnitude based on the prediction reliability assessment results to reduce the optimistic bias of the net available power prediction sequence and improve power supply stability. The selection of the conservative correction magnitude involves classifying reliability into high, medium, and low levels, corresponding to different conservative correction intensities. At low reliability, a more conservative availability boundary is selected, such as a smaller availability value or a larger safety margin. The initial classification boundary can be given by factory voltage testing, historical undervoltage sample statistics, or empirical initial values, and is continuously adjusted during operation based on the undervoltage occurrence rate and strategy jitter. In this embodiment, at least one power supply constraint can be selected for conversion, including at least: energy storage state constraints, critical function backup constraints, and power supply link loss constraints. Energy storage state constraints can be obtained from the state of charge output by the battery management module or from the integration of charge and discharge current combined with capacity calibration; critical function safety constraints can be configured by the user or given by a preset strategy table, and can be switched according to outdoor operating mode; power supply link loss constraints can be estimated from the bus voltage and current combined with the link equivalent parameters. The link equivalent parameters can be given initial values ​​by factory calibration and updated on a rolling basis through electrical parameter statistics during the stable section. Based on the conservatively corrected net available power prediction, the operation strategy command is generated and output to the load for execution. To prevent frequent start-stop, this embodiment introduces a jitter suppression mechanism in the strategy output: a minimum hold time and cooling time are set for the same load to avoid repeated switching caused by short-term fluctuations; when the prediction confidence is low or the load step reaction confidence is high, a degradation strategy is prioritized, such as reducing the power of non-critical loads or delaying their start-up; a safety net power supply is maintained for critical loads, and non-critical loads are scheduled according to a priority queue. This embodiment also updates the calculation benchmark for the deviation of historical stability level or consistency characterization item based on the bus voltage recovery after load execution and subsequent residual changes: if the bus voltage recovers quickly and the residual converges after load execution, the conservative boundary is allowed to be gradually relaxed and the stability baseline is updated; if the bus voltage recovers slowly or undervoltage events increase, the conservative boundary is tightened and the sensitivity to relevant disturbance labels is increased, while the benchmark range for deviation judgment is updated, so that the system enters the conservative strategy earlier for power supply vulnerability.

[0057] Embodiment 5 of the present invention: Based on Embodiment 1, Embodiment 2, Embodiment 3 or Embodiment 4, this embodiment further provides a mechanism for generating and executing operation strategy instructions, so that net available power prediction can be transformed into load regulation actions, and reduces the risk of undervoltage reset, frequent start-stop and malfunction when there are outdoor operating condition disturbances and data uncertainties.

[0058] Load grading and safety margin constraint strategy generation: This embodiment divides the load into critical loads and non-critical loads, and sets safety margin power constraints for critical loads. Critical loads include at least communication and positioning, basic lighting, and security alarms, while non-critical loads include at least comfort and extended power loads such as fans, heating, and high-speed charging. When generating the strategy, the safety margin reserved power for critical functions is deducted first, and then the remaining net available power is allocated to non-critical loads. This ensures that critical functions remain uninterrupted in the evening when sunlight decreases rapidly or when power fluctuations are significant on cloudy days, while non-critical loads are downgraded or postponed according to priority. Load grading and priority can be obtained from a preset strategy table. The safety margin reserved power for critical functions can be obtained from historical experience configuration values, or it can be obtained based on historical critical load power consumption statistics and continuously adjusted.

[0059] Threshold hysteresis and time constraint-based jitter suppression strategy: To avoid repeated switching caused by fluctuations in net available power within a short window, this embodiment sets an on / off threshold for each load, and sets a minimum hold duration and cooling time. Specifically, an on or upgrade command is generated only when the conservatively corrected net available power is higher than the on threshold and the prediction confidence is not lower than the threshold for multiple consecutive rolling windows; a off or downgrade command is generated only when the net available power is lower than the off threshold or the prediction confidence significantly decreases for multiple consecutive windows; after the load is on, it must maintain a minimum hold duration, and after it is off, it must undergo at least a cooling time before being allowed to be on again. The on and off thresholds can be obtained from historical undervoltage samples. The minimum hold duration and cooling time can be set based on load characteristics and user experience, and can be optimized by combining statistical results of start-stop jitter occurrence periods. Figure 4As shown, the upper curve represents the safe and executable net available power boundary updated over time, and indicates the hysteresis interval formed by the start threshold Th_on and the stop threshold Th_off. This hysteresis is used to characterize the short-term fluctuations in the net available power boundary near the threshold under the combined effects of outdoor power supply fluctuations, load changes, and prediction uncertainties. The lower curve shows the strategy action output under the same boundary input conditions: If the "threshold hysteresis and time constraint jitter suppression strategy" of this embodiment is not adopted, and control is only based on threshold hysteresis (red line), the load will be triggered to start and stop multiple times at the moment the boundary repeatedly crosses the threshold, forming short-cycle switching and start-stop jitter; After adopting the strategy of this embodiment (green line), the system only confirms and submits the start / up action when the conservatively corrected net available power boundary is higher than Th_on for multiple consecutive rolling windows and the prediction confidence is not lower than the threshold. After the action is submitted, the minimum hold time suppresses the immediate shutdown caused by short-term sway; when the boundary is continuously lower than Th_off or the prediction confidence is significantly reduced, the shutdown / degradation action is generated, and after shutdown, the cooling time prevents the immediate restart caused by the short-term rebound of the boundary. Therefore, in scenarios where the safety boundary fluctuates briefly near Th_on / Th_off, this embodiment significantly reduces the number of start-stop cycles and maintains stable operation through the joint constraints of "hysteresis threshold + confirmation window + minimum hold time + cooling time (which may include cancellation / rollback)," thereby reducing the risk of undervoltage reset and malfunction, and improving the executability and robustness of outdoor power supply regulation.

[0060] Concurrency limiting and segmented startup peak shaving strategy: For high-power loads such as fans, heaters, or fast chargers, this embodiment incorporates concurrency limiting and segmented startup control into the strategy. Concurrency limiting restricts multiple high-power loads from starting simultaneously within the same time window; segmented startup gradually increases the load from a low level to the target level, or progressively increases the duty cycle to suppress the instantaneous drop in bus voltage caused by startup surges. The concurrency limiting threshold and segmented startup step size can be obtained from the bus voltage recovery curve, load startup surge characteristics, and historical undervoltage event statistics; when the system detects an increase in bus recovery time or undervoltage event density, the strategy automatically tightens the concurrency limit and reduces the segmented startup speed, thereby reducing the reset probability in scenarios such as multiple users charging simultaneously, frequent USB plugging and unplugging, or reverse charging of power banks.

[0061] Differentialized strategy mapping based on disturbance tags: This embodiment links the operation strategy with outdoor operating condition disturbance tags and their confidence levels. Different disturbance types trigger different strategy branches to improve strategy targeting and interpretability. When the confidence level of flexible panel deformation disturbances is high, it is determined that there is an abnormal power output shape and an unstable mapping relationship. The strategy prioritizes entering a conservative mode, restricting the start-up of high-power loads and fast charging upgrades, and increasing the start threshold and confirmation window length to avoid triggering large current steps during abnormal waveforms on the power generation side, leading to undervoltage. When the confidence level of periodic shading disturbances is high, it is determined that there are periodic fluctuations caused by intermittent shading. The strategy prioritizes strengthening jitter suppression constraints, extending the minimum hold time, and increasing the shutdown confirmation conditions to avoid frequent start-ups and shutdowns of the load following periodic fluctuations. When the confidence level of temporary coverage disturbances is high, it is determined that there is a sudden power step drop caused by shading. The strategy prioritizes rapid degradation or delayed execution, immediately reducing the power level of non-critical loads and setting cancellation window parameters. When the shading is removed and the net available power recovers and continuously meets the recovery window number, the load is gradually restored according to the segmented start-up rules. When the confidence level of disturbances such as condensation or slow decay of water film is high, it is determined that the unit light output capacity is continuously decreasing. The strategy does not rely on short-term rebounds in a single window, but instead tightens the available boundary according to the trend and gradually lowers the target level of non-critical loads, thereby avoiding the risk of cumulative undervoltage during periods of heavy morning dew or post-rain dampness. When the confidence level of disturbances such as load step reaction is high, it is determined that the bus drop is mainly caused by a sudden increase in load. The strategy prioritizes peak shaving control, reduces the concurrency of high-power loads, and performs rollback or postponement on recently triggered loads. This label is used as a source of counter-evidence for shading and deformation labels to reduce erroneous actions caused by misattribution.

[0062] Two-phase execution and reversal window mechanism: To reduce the cost of misjudgment, this embodiment allows a two-phase strategy for outputting non-critical loads. The first phase is a preparatory state instruction, which executes low-disturbance actions and carries reversal window parameters, such as only enabling low power, delaying execution, or limiting the rate of increase. The second phase is a commit state instruction, which commits strong actions such as high power or fast charging when the prediction confidence and net available power meet the conditions within the continuous confirmation window. If data quality deteriorates, disturbance confidence increases, or bus recovery slows down within the reversal window, the commit state is revoked and the system reverts to the preparatory state action, thereby reducing the risk of incorrect execution timing caused by wireless link latency, packet loss, or data backfilling.

[0063] For example, in semi-shaded campsites in woodlands or valleys, when tree shadows cause periodic shading, the strategy uses threshold hysteresis and minimum hold time to prevent fans and lighting from frequently starting and stopping due to periodic fluctuations, thereby reducing the number of start-stop cycles and minimizing repeated drops in bus voltage. In windy environments such as coastal areas or mountain passes, when flexible panels deform under stress, causing sawtooth fluctuations in the output waveform, the strategy uses concurrency limiting and segmented startup to reduce startup surges, and limits fast charging upgrades when deformation confidence increases, thereby reducing the probability of undervoltage reset. When outdoor users temporarily place awnings or equipment on the roof, creating sudden shading, the strategy uses rapid degradation and window cancellation mechanisms to promptly roll back non-critical loads, and then restores them in segments only after net available power recovers and continuously meets the recovery window requirements, thereby reducing the duration of malfunctions. When morning condensation or water film causes a slow decline in output capacity, the strategy tightens the available boundary according to the trend and gradually lowers the non-critical load levels, thereby avoiding the risk of continuous undervoltage caused by overestimation of net available power. Through the above-mentioned operation strategy generation and execution mechanism, the net available power prediction sequence can be transformed into a set of executable instructions that include thresholds, timing, concurrency limits, segmented start-up, degradation and cancellation windows. When there are outdoor operating condition disturbances and data uncertainties, the intensity of actions is automatically reduced and jitter is suppressed, thereby improving the stability and executability of power supply control in outdoor operating conditions.

Claims

1. A method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered power, characterized in that, Includes the following steps: S1: Collect multi-source time series data and unify them to the same time base, and perform timestamp consistency verification, resampling, missing data handling and outlier suppression, while outputting data quality indicators; The multi-source time-series data includes at least: first time-series data for characterizing external energy supply conditions and second time-series data for characterizing power generation output; Furthermore, the multi-source time-series data also includes one or more of the following: third time-series data for characterizing the power supply status of the load side, fourth time-series data for energy storage status, and fifth time-series data for load status. S2: Calculate the consistency characterization term within the scrolling window, and output the outdoor operating condition disturbance label set and the outdoor operating condition disturbance confidence level based on the consistency characterization term; The consistency characterization term is used at least to characterize abrupt changes or drifts in the relationship between illumination and power; S3: Call the time series prediction model to output the solar power generation prediction sequence for future periods, and perform rolling correction based on historical residuals and corresponding outdoor operating condition disturbance confidence to obtain the corrected power generation prediction sequence, while outputting the prediction uncertainty characterization. S4: Based on at least one power supply constraint, convert the corrected power generation prediction sequence into a future time period net available power prediction sequence; The power supply constraints include at least one or more of the following: energy storage state constraints, critical function backup constraints, and power supply link loss constraints. S5: Based on the data quality indicators, the outdoor operating condition disturbance confidence level, and the prediction uncertainty characterization, obtain the prediction credibility assessment result, and conservatively correct the net available power prediction sequence according to the prediction credibility assessment result, and generate an operation strategy instruction to be output to the load for execution.

2. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered power supply according to claim 1, characterized in that, The data quality indicators include at least timestamp alignment error, missing rate, and anomaly rate; The timestamp alignment error is obtained by statistically analyzing the time offset of different source data at the same alignment anchor point. The missing percentage is obtained by the ratio of the number of missing points in the scrolling window to the total number of points in the scrolling window. The anomaly ratio is obtained by the proportion of the number of anomalies to the total number of points in the window.

3. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered electricity according to claim 1, characterized in that, The specific contents of the outdoor operating condition disturbance label set and the outdoor operating condition disturbance confidence level are as follows: The outdoor working condition disturbance tag set includes at least one or more of the following: flexible plate deformation disturbance tags, periodic occlusion disturbance tags, temporary coverage disturbance tags, condensation or water film slow decay disturbance tags, and load step reaction disturbance tags. The outdoor operating condition disturbance confidence includes the label confidence corresponding to each disturbance label in the outdoor operating condition disturbance label set, and the label confidence is jointly determined by the degree of deviation and the persistence of deviation of the consistency characterization item. The degree of deviation represents the magnitude of the deviation of the consistency indicator from the historical stable level; The deviation persistence characterizes the continuous occurrence of the deviation within multiple scrolling windows.

4. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered electricity according to claim 3, characterized in that, The generation of the flexible panel deformation disturbance tag includes judging based on morphological evidence of sawtooth fluctuations or periodic fluctuations in power output within a rolling window, confirming it by combining the joint evidence of flexible panel deformation with the decrease in consistency between illumination and power generation, and performing an upward update on the outdoor operating condition disturbance confidence of the flexible panel deformation disturbance tag.

5. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered electricity according to claim 3, characterized in that, The generation of the periodic occlusion disturbance label includes discrimination based on evidence of recurring periodic fluctuations in light intensity and power generation output within a rolling window, confirmation by combining the periodic stability or recurrence frequency of the periodic fluctuations with joint evidence of periodic occlusion, and an upward update of the confidence level of the outdoor operating condition disturbance of the periodic occlusion type.

6. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered power supply according to claim 3, characterized in that, The generation of the temporary coverage disturbance label includes: when the change in light intensity is less than the light stability threshold and the power generation power changes by more than the power change threshold, a judgment is made based on the joint evidence of temporary coverage. The joint evidence for temporary coverage includes at least one or more of the following: the consistency between illumination and power generation is lower than the consistency threshold and the prediction residual before and after correction exceeds the residual surge threshold. The outdoor operating condition disturbance confidence of the temporary coverage disturbance label is then updated upwards.

7. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered electricity according to claim 3, characterized in that, The generation of the condensation or water film slow decay type perturbation tag includes judging based on trend evidence that the unit light output capability is continuously decreasing within multiple rolling windows, and combining environmental evidence of indoor humidity and temperature difference between indoor and outdoor environments to enhance the trend evidence, and performing an upward update on the outdoor working condition perturbation confidence of the condensation or water film slow decay type perturbation tag.

8. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered electricity according to claim 3, characterized in that, The generation of the load step reaction type disturbance tag includes judging based on the coupling strength evidence between load state change and bus voltage drop, and confirming it by combining the consequence type evidence of prolonged undervoltage recovery time or increased undervoltage event density, and performing an upward update on the outdoor operating condition disturbance confidence corresponding to the reaction type disturbance tag.

9. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered electricity according to claim 1, characterized in that, The rolling correction includes weighted updating of historical residuals based on the confidence level of the outdoor operating condition disturbance.

10. The method for multi-parameter environmental sensing and control of outdoor tents based on solar self-powered power supply according to claim 1, characterized in that, The conservative correction includes selecting a conservative correction magnitude based on the prediction reliability assessment results to reduce the optimistic bias of the net available power prediction sequence and improve power supply stability, and rolling updates the calculation benchmark of the deviation of the historical stability level or consistency characterization item based on the bus voltage recovery after load execution and subsequent residual changes.

Images