A fresh stew yunao cold chain temperature data anomaly detection and compensation method

Abstract

The present application belongs to the technical field of cold chain logistics monitoring, and relates to a kind of fresh stewed bird's nest cold chain temperature data anomaly detection and compensation method, solve the problem that existing cold chain monitoring cannot meet the precise control of fresh stewed bird's nest industrialization cold chain. The present application collects multi-modal signals and uses cross-correlation method to calculate the equivalent thermal inertia characteristic quantity of fresh stewed bird's nest material liquid, and calculates the interval value basic probability assignment of the preset state level at the current time, uses discount correction and interval evidence fusion to obtain fusion confidence interval when conflict occurs, constructs a multi-level state machine supporting bidirectional transfer, drives state upgrade according to the fusion confidence interval, realizes self-recovery degradation of abnormal state based on thermal inertia characteristic quantity recovery, the state after state upgrade or the state after degradation rollback is used as the final state level, and hierarchical compensation is executed according to the final state level. The present application realizes the abnormal precise detection and active intelligent compensation of fresh stewed bird's nest cold chain transportation, and adapts to the precise control requirements of industrialization cold chain.

Description

Technical Field

[0001] This invention belongs to the field of cold chain logistics monitoring technology, and specifically relates to a method for detecting and compensating for abnormal temperature data in the cold chain of fresh stewed bird's nest. Background Technology

[0002] Freshly stewed bird's nest is prepared using a 98℃ low-temperature stewing process. The product contains no preservatives and has extremely poor quality stability. It must rely on a 0-4℃ low-temperature cold chain environment for storage and transportation, placing extremely high demands on the precision of cold chain temperature control, state stability, and tolerance for anomalies. Current cold chain monitoring and control solutions for freshly stewed bird's nest are relatively traditional, with mainstream methods mainly consisting of post-event data backtracking via temperature recorders and real-time over-limit alarms via wireless temperature and humidity tags. These methods only achieve basic monitoring of cold chain temperature. Post-event data recorders can only perform post-event reading and verification of temperature data; once a temperature exceeds the limit, the only solution is usually to scrap the entire batch of products, resulting in significant resource waste. While wireless temperature and humidity tags can provide real-time over-limit alarms, they can only warn of anomalies and cannot correct or compensate for temperature deviations that have already occurred, leading to a significant lag in control measures. Existing high-end cold chain prediction models only model and analyze the air temperature in the cold chain environment, without considering the physical state changes of the bird's nest liquid itself, resulting in limited monitoring dimensions and poor adaptability.

[0003] The existing technology system has many practical technical pain points, making it difficult to meet the high-precision cold chain management requirements for freshly stewed bird's nest: First, conventional monitoring sensors only collect the air temperature of the cold chain box, and the temperature of the core liquid inside the bird's nest bottle has a significant response lag, which easily leads to monitoring distortion problems such as the ambient temperature meeting the standard while the liquid inside the bottle has quietly exceeded the temperature, failing to accurately reflect the actual temperature control status of the product; Second, mechanical disturbances such as vibration and impact during cold chain transportation can damage the thermal boundary layer of the bird's nest liquid, causing local rapid temperature rise, and traditional single temperature monitoring methods cannot identify such hidden anomalies; Third, the existing technology lacks the ability to distinguish between minor recoverable overheating and irreversible deterioration anomalies, which can easily lead to the wrong disposal of compliant and good products, significantly increasing product storage and transportation losses and operating costs; Fourth, the existing cold chain management mode is all passive monitoring and early warning, without targeted dynamic temperature compensation and anomaly repair mechanisms. After a temperature control anomaly occurs, there is no way to actively intervene and correct the situation, and the product quality loss can only be passively borne, resulting in a serious lack of intelligence and precision in cold chain management. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, namely the limitations of current cold chain monitoring for freshly stewed bird's nest (such as limited monitoring dimensions, distorted identification of temperature control status of the liquid inside the bottle, inability to capture hidden temperature anomalies, insufficient refinement of anomaly level discrimination, lack of dynamic temperature compensation capabilities, and inability to iteratively optimize the model), which make it difficult to adapt to the precise control of the industrial cold chain for high-quality freshly stewed bird's nest, this invention proposes a method for anomaly detection and compensation of temperature data in the cold chain of freshly stewed bird's nest, comprising the following steps: Multimodal signals of fresh stewed bird's nest during cold chain transportation are collected. The multimodal signals include the bottle center temperature signal, bottle wall temperature signal, mechanical disturbance signal, box door status signal, and thermal status signal of the phase change energy storage module. Short-duration heat pulses are periodically applied to the wall of the bird's nest bottle. Based on the temperature signal at the center inside the bottle and the temperature signal on the bottle wall, the equivalent thermal inertia characteristic of the fresh stewed bird's nest liquid is calculated using the cross-correlation method. The basic probability assignment of the interval values ​​of the current state level is calculated for the multimodal signal and the equivalent thermal inertia feature quantity respectively. The preset state levels include normal, pre-cooling standby, slightly compensable, moderately irreversible, and severely scrapped. When there is a judgment conflict between the basic probability assignment of the interval values ​​of different signals, a conflict detection factor is introduced for discount correction, and then the interval evidence combination rule is used for fusion to obtain the fusion confidence interval of each state level, thereby obtaining the initial judgment of abnormal state and the fusion confidence level characterizing the uncertainty of the initial judgment. A multi-level state machine supporting bidirectional transitions is constructed, and state upgrades are driven according to the fusion confidence interval. When the equivalent thermal inertia characteristic quantity recovers to within the baseline threshold and the center temperature inside the bottle is lower than the preset value for several consecutive times, the state machine is allowed to downgrade and transition back from a mild or moderate abnormal state according to the preset recovery conditions. The state after state upgrade or the state after downgrade and transition back is taken as the final state level. Based on the final state level, graded compensation is performed: during pre-cooling standby, the phase change energy storage module is actively pre-cooled; when the fusion confidence is lower than the cancellation threshold and the current center temperature inside the bottle is lower than the preset recovery temperature threshold, reverse heating is used to restore the cold storage capacity; when the condition is slightly compensable, the subsequent allowable ambient temperature is compressed; when the condition is moderately irreversible, the dynamic remaining shelf life is calculated and the QR code is updated; when the condition is severely scrapped, an alarm is issued and compensation is disabled.

[0005] The beneficial effects of this invention are: 1) Six types of signals are collected: bottle center temperature, bottle wall dual-redundant temperature, ambient temperature and humidity, acceleration impact, door magnetic switch, and phase change module thermal state. A multimodal sensor evidence set is constructed to achieve comprehensive perception from the environment to the liquid, and from the thermal field to disturbances. Based on this, through cross-correlation thermal pulse active excitation and temperature normalization compensation mechanism, the equivalent thermal inertia characteristic of the liquid is extracted to achieve active identification of internal deterioration precursors such as gelation, precipitation, and bubbles. This breaks through the limitation of traditional passive temperature measurement that cannot detect changes in the internal structure of the liquid. At the same time, the accelerometer is used to detect mechanical impact in real time and dynamically correct the heat exchange coefficient to avoid missed detection of hidden temperature anomalies caused by thermal boundary layer damage. 2) By assigning basic probability values ​​to interval values, using conflict detection factors, discount corrections, and interval evidence combination rules, the uncertainty of sensor confidence is explicitly expressed. This effectively handles evidence conflict scenarios such as a normal door magnet but a temperature lag when the door is opened, overcoming the shortcomings of traditional single-threshold methods in decision-making failure when evidence conflicts occur. Simultaneously, the impact of temperature history on quality is decomposed into reversible equivalent time and irreversible high-temperature damage factors, enabling five levels of refined discrimination: normal, pre-cooling standby, slightly compensable, moderately irreversible, and severely scrapped. This distinguishes between reversible fluctuations and irreversible damage, preventing the entire batch from being scrapped due to slight overheating. 3) Construct a multi-level state machine that supports bidirectional transfer. Based on the recovery of normalized thermal inertia characteristics, the state machine is allowed to degrade from a mild or moderate abnormal state back to a normal state, breaking through the technical bias of traditional cold chain monitoring that abnormality means scrap and cannot be self-recovered. It also designs a four-level differentiated compensation closed loop, including pre-cooling standby active pre-cooling, mild compensable compression subsequent ambient temperature, moderate irreversible output of dynamic remaining shelf life, and severe scrap alarm disabling compensation, to achieve a complete closed loop from abnormal detection to compensation execution and state recovery. 4) The dynamic remaining shelf life is calculated by combining reversible equivalent time and irreversible high temperature damage factor, and encoded into a QR code for consumers to scan to obtain the real shelf life information, avoiding the scrapping of the whole batch due to slight overheating. At the same time, the compensation hit rate, accuracy and end quality inspection results of a single transportation task are used as feedback samples to trigger edge fine-tuning. Only the confidence mapping parameters of the last layer are updated and elastic weights are added to consolidate the regularization term to prevent forgetting, so as to realize the continuous self-optimization of the model and adapt to batch differences and environmental changes. Attached Figure Description

[0006] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of the steps of an anomaly detection and compensation method for cold chain temperature data of freshly stewed bird's nest according to the present invention.

[0007] Figure 2This is a flowchart illustrating the introduction of a conflict detection factor for discount correction in a method for detecting and compensating for anomalies in cold chain temperature data of freshly stewed bird's nest, as described in this invention. Detailed Implementation

[0008] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0009] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0010] To more clearly explain the method for detecting and compensating for anomalies in cold chain temperature data of freshly stewed bird's nest according to the present invention, the following is a detailed explanation. Figures 1 to 2 The steps in the embodiments of the present invention will be described in detail below.

[0011] The first embodiment of this invention discloses a method for detecting and compensating for anomalies in cold chain temperature data of freshly stewed bird's nest. (See also...) Figure 1 This includes the following steps: S1, collect multimodal signals of fresh stewed bird's nest during cold chain transportation. The multimodal signals include the bottle center temperature signal, bottle wall temperature signal, mechanical disturbance signal, box door status signal, and thermal status signal of the phase change energy storage module. In this embodiment, in order to obtain comprehensive status information of fresh stewed bird's nest during cold chain transportation, multimodal signal acquisition is adopted. The acquisition of multimodal signals is realized by the edge computing unit in the cold chain box. The edge computing unit needs to be calibrated at the factory before being put into use to ensure the accuracy of the acquired data. Subsequently, real-time acquisition is carried out during transportation. Factory calibration stage: Center temperature sensor calibration: Sensirion STS31 was used, with an accuracy of ±0.1℃@0~10℃ and a response time of 2 seconds. Calibration method: The sensor was placed in a 0℃ ice-water mixture (with thorough stirring), stabilized for 10 minutes, and 10 readings were recorded and averaged to calculate the offset Δ0. It was then moved to a 4℃ constant temperature bath (accuracy ±0.05℃), stabilized for 10 minutes, and 10 readings were recorded and averaged to calculate the offset Δ4. This process was repeated in a 10℃ constant temperature bath. The linear interpolation correction formula Tcorr = Traw + ((Traw-0) / (10-0)) was used. (Δ10-Δ0)+Δ0) stores the offset in the EEPROM of the main control MCU for lifetime use; Wall temperature sensor consistency calibration: Maxim DS18B20 stainless steel tube packaged, accuracy ±0.2℃, 12-bit resolution; Calibration method: Place two DS18B20 sensors and PT100 sensor simultaneously in a 4℃ constant temperature bath, stabilize for 10 minutes, take 10 readings and average them to calculate ΔT1 and ΔT2. If |ΔT1-ΔT2|>0.2℃, store their respective offsets in the EEPROM of the main control MCU for subsequent software correction. Phase change energy storage module thermal characteristic calibration: Place the phase change energy storage module in a 4℃ environment and heat it to 10℃ using a semiconductor cooling chip; allow it to cool naturally or actively, and record the time it takes to drop from 3.5℃ to 2.5℃. If the time is >20 minutes, it is marked as unqualified and the phase change energy storage module is replaced. The 20-minute threshold is derived from the average value of normal samples + 3σ (σ = 2 minutes, n = 30). Baseline parameters for freshly stewed bird's nest batches were obtained as follows: Each individual packaging unit (i.e., each bottle of freshly stewed bird's nest) was placed in a 4°C constant temperature chamber for 2 hours to stabilize; a 1W power thermal pulse was applied for 0.5 seconds, and the wall and center temperatures were collected at 20Hz for 30 seconds. The pulse response sequence was calculated using the cross-correlation method, and the centroid was calculated as the equivalent time constant. (Impulse response sequence, The specific calculation process is the same as below. , (Details omitted here); A linear temperature increase of the wall surface (2℃ to 6℃, rate 0.5℃ / min) is achieved using a semiconductor refrigeration chip, and the temperature coefficient θ is obtained by fitting the Arrhenius equation. Specifically, the Arrhenius equation is: ; in, For the first The dependent variable for each sampling point is the natural logarithm of the rate of change of the center temperature. For the first The rate of change of the center temperature at each sampling point, in °C / s; A is the pre-exponential factor (frequency factor) of the Arrhenius equation. Activation energy, in J / mol, represents the energy threshold required for a reaction; R = 8.314 J / (mol·K) is the universal gas constant. For the first The wall temperature at each sampling point, in Kelvin. Using least squares linear regression, we obtain the slope m = -Ea / R, and thus Ea = -mR. We then calculate the temperature coefficient. ; Wherein, θ is the temperature coefficient, in units of / ℃, which represents the relative increase in the rate of quality degradation when the temperature rises by 1℃. The normal range of θ is 0.09~0.18 / ℃. If it exceeds the range, the safety parameters of this batch need to be reassessed; otherwise, it will be rejected from the factory. For reference temperature, 277.15K (corresponding to 4℃); Will The parameters θ, Smax (total safety equivalent time, in hours, 2 bytes, typical value 240h), along with the product ID, production date, and calibration timestamp, are written into a QR code and printed on the bottle. This gives each individual packaging unit its own "thermal fingerprint" and temperature sensitivity coefficient for personalized anomaly detection and compensation, avoiding misjudgments caused by using uniform parameters. The stabilization time is 2 hours (if the bottle radius is 4cm, the thermal diffusivity is approximately 1.4e-7 m² / s, and the thermal diffusion time is approximately 1140s ≈ 19min, a 2-hour safety redundancy is used). A thermal pulse energy of 0.5J with a temperature rise of <0.3℃ does not affect quality. The normal range of θ is 0.09~0.18 / ℃, derived from 10 batches of destructive testing. When loading into the cold chain box, baseline parameters are read by scanning a QR code or using NFC and stored in the MCU Flash for subsequent judgment. Real-time data collection during transportation: After the above calibration is completed, the baseline parameters are read by scanning the QR code and stored in the Flash memory of the main control MCU during packaging. Real-time data acquisition is performed from the following sensors: The central temperature sensor is fixed to the geometric center of the bottle in the form of a stainless steel probe, with a sampling frequency of 1Hz, to directly measure the true temperature of the liquid. Two wall temperature sensors are configured and symmetrically installed on the inner surface of the bottle wall. The sampling frequency is 0.5Hz. They are used to obtain the average heat flux boundary conditions, eliminate installation position deviations, and provide dual-channel redundancy to resist single-point failures. An ambient temperature and humidity sensor is installed in the air layer inside the cold chain box, with a sampling frequency of 0.2Hz, to predict future changes in ambient temperature and detect condensation conditions. The triaxial accelerometer uses the ST LIS3DH, with a range of ±16g, and is installed on the inner wall of the bottom of the housing. It has a sampling frequency of 100Hz (burst trigger mode) and is used to detect mechanical shocks to correct the heat exchange coefficient.

[0012] The Hall effect magnetic switch is a normally open type with a reed switch and a magnet, with an operating distance of 10mm. It is installed at the junction of the cover and the body of the box, and the level signal sampling frequency is 1Hz. It is used to detect door opening events. The phase change energy storage module temperature sensor uses an additional DS18B20, which is mounted on the surface of the phase change energy storage module and has a sampling frequency of 0.2Hz, to monitor the thermal state of the phase change energy storage module.

[0013] All sensor data is connected to the STM32L432KC main control MCU (80MHz main frequency, 64kB RAM, 256kB Flash) via I²C or SPI bus. The MCU has a built-in real-time clock (accuracy ±2ppm, factory-calibrated to synchronize with network time) to timestamp each frame of data, forming a spatiotemporally consistent multimodal signal set; S2, a short-duration heat pulse is periodically applied to the wall of the bird's nest bottle, and the equivalent thermal inertia characteristic of the fresh stewed bird's nest liquid is calculated by cross-correlation method based on the temperature signal of the center inside the bottle and the temperature signal of the bottle wall. In this embodiment, to monitor the internal state changes of the freshly stewed bird's nest liquid in real time (such as gelation, precipitation, and the mixing of air bubbles) as an early warning of internal deterioration and to provide the current heat exchange parameters for the temperature prediction model, since conventional passive temperature measurement cannot detect changes in the internal structure of the liquid, an active thermal exciter is used to actively apply excitation to obtain a response. The active thermal exciter uses a ceramic heating element with a size of 5×5mm, a resistance of 3Ω, and a maximum power of 1W@3.7V, which is attached to the outer surface of the bottle wall. Since the deterioration process of bird's nest is slow (on the order of hours), a thermal pulse is executed every 30 minutes. Specifically, the thermal pulse is driven by a MOSFET with a 50% duty cycle PWM (frequency 1kHz) output by the MCU, applying 1W of power for 0.5 seconds. Closed-loop monitoring of voltage and current ensures power stability. The wall temperature sensor and the center temperature sensor are set at 20Hz (sampling rate 20Hz is based on: For approximately 3 seconds, the Nyquist frequency requirement is >0.67Hz, and 20Hz provides sufficient oversampling. The wall temperature Tw(t) and center temperature Tc(t) are sampled for 30 seconds (600 sampling points in total). Glitches are removed using a 5-point median filter, and the average value of the first 10 sampling points is subtracted as the initial steady-state reference. The specific process for calculating the equivalent thermal inertia characteristic quantity using the cross-correlation method is as follows: The bottle wall temperature signal before and after the application of a short-duration thermal pulse is cross-correlated with the bottle center temperature signal to obtain the pulse response sequence; the cross-correlation calculation formula is as follows: ; In the formula, This is a pulse response sequence, representing the degree of influence of the wall temperature pulse at time k on the center temperature; The difference in wall temperature at time (nk) relative to the initial steady state (in °C). The difference in center temperature at time n relative to the initial steady state (in °C); N=600 is the total number of sampling points (30 seconds × 20 Hz); M=200 is the impulse response length (corresponding to 10 seconds, because...). Approximately 3 seconds; the impulse response decays to 0 within 15 seconds (5 times the time constant, 10 seconds is sufficient for coverage); k is the delay index; n is the index of the current sampling point. Calculate the centroid of the impulse response sequence as the measured equivalent time constant; ; In the formula, The measured equivalent time constant (in seconds) reflects the current thermal inertia state of the liquid feed. The normal range is 2-5 seconds. A preliminary experiment was conducted on 30 bottles of freshly stewed bird's nest. The mean was 3.2 seconds and the standard deviation was 0.4 seconds. The 2-5 second range covers the mean ± 4.5σ. Δt is the sampling interval (0.05 seconds, corresponding to a 20Hz sampling rate). k is the delay index. h(k) is the impulse response sequence. To verify the effectiveness of this thermal pulse measurement and determine whether the acquired temperature response data has been affected by external interference (such as vibration, bubble swaying, sensor noise, etc.), it is necessary to calculate the reconstructed temperature and root mean square error (RMSE). The reconstructed temperature is the "theoretical center temperature change" calculated in reverse based on the identified impulse response sequence h(k) and the measured wall temperature change ΔT_w(n). If the identified impulse response h(k) is accurate, then the center temperature reconstructed from the wall temperature should be very close to the measured center temperature. Reconstructed temperature calculation formula: ; in, The change in center temperature reconstructed from the wall temperature response (in °C). Root mean square error (RMSE) is the reconstructed temperature. Compared with the measured temperature The quantitative evaluation index for the differences between them is calculated using the following formula: ; in, Root mean square error (in °C) is used to measure the goodness of fit between the reconstructed temperature and the measured temperature. The criteria and function of RMSE: If RMSE < 0.05℃, it indicates that the reconstruction temperature matches the measured temperature well, the measurement is valid, and the result of this calculation is accepted. And use it for subsequent state machine decisions; If RMSE ≥ 0.05℃, it indicates a significant difference between the reconstructed temperature and the measured temperature, rendering the measurement invalid. This may be due to the following reasons: abnormal fluctuations in the temperature curve caused by mechanical vibration or shaking; sudden changes in local thermal conductivity caused by bubble movement; increased signal noise due to poor sensor contact; or data glitches caused by electromagnetic interference. In this case, discard the current measurement result and use the previous valid result. And add a "physical disturbance counter" to record the disturbance event; The temperature rise caused by the thermal pulse is approximately 0.2-0.4℃, and 0.05℃ corresponds to a signal-to-noise ratio of approximately 80%. A value below this is considered to indicate that the response is overwhelmed by noise or that physical perturbation exists. This verification mechanism ensures the accuracy of the thermal inertia characteristic. To ensure reliability and avoid misjudgments caused by external interference; Anomaly detection (relative to factory baseline τ_eff0): If And if it is effective twice consecutively, it triggers "internal deterioration precursor" (L2); if However, if RMSE ≥ 0.05℃, only the "perturbation" is recorded, and no deterioration is triggered; if The sensor may be faulty or detached, triggering an alarm. A correction factor based on the temperature change rate of the thermal diffusivity coefficient of freshly stewed bird's nest liquid is used to normalize the measured equivalent time constant to obtain the normalized equivalent thermal inertia characteristic quantity, which is then used as the equivalent thermal inertia characteristic quantity. Its properties change with temperature (the higher the temperature, the faster the heat diffusion). If normalization is not performed, normal temperature fluctuations will be misinterpreted as structural changes; The correction factor is derived based on the following principle: the thermal diffusivity α(T) of freshly stewed bird's nest liquid changes linearly with temperature, α(T) = 0.0014 + 0.00001·(T-4) m² / s, where 0.0014 is the thermal diffusivity of water at 4℃ (standard physical constant), and 0.00001 is the increase in thermal diffusivity for every 1℃ increase (obtained by linear fitting of experimental data of water in the range of 0-20℃). Since the thermal response time constant τ is inversely proportional to the thermal diffusivity α, i.e., τ∝1 / α, the measured... The correction factor, normalized to a reference temperature of 4℃, is f(T_c) = α(4℃) / α(T_c). Experiments have verified that the temperature change rate of the thermal diffusivity of the freshly stewed bird's nest liquid deviates from that of water by less than 5%, thus this correction factor can be used approximately. The normalization factor is: ;in, This is the temperature normalization correction factor; is the thermal diffusivity of water (unit: m² / s); The current center temperature inside the bottle (in °C); =4℃ is the reference temperature; The normalized equivalent thermal inertia characteristic is: ; in, The normalized equivalent thermal inertia characteristic eliminates the effect of temperature on the thermal inertia. The impact only reflects changes in the internal structure of the liquid feed; This is the correction factor at the current temperature; S3, calculate the basic probability assignment of the interval values ​​of the current state level for the multimodal signal and the equivalent thermal inertia feature quantity respectively. The preset state levels include normal, pre-cooling standby, slightly compensable, moderately irreversible, and severely scrapped. When there is a judgment conflict between the basic probability assignment of the interval values ​​of different signals, a conflict detection factor is introduced for discount correction, and then the interval evidence combination rule is used for fusion to obtain the fusion confidence interval of each state level, thereby obtaining the initial judgment of abnormal state and the fusion confidence level characterizing the uncertainty of the initial judgment. In this embodiment, the decision-making conflict problem of multi-source heterogeneous sensors in a complex cold chain environment is resolved. By using interval evidence fusion theory, an abnormal state judgment result with uncertainty quantification is output. The basic probability of the current state level being in the preset state level is calculated and assigned to the multimodal signal and the equivalent thermal inertia characteristic quantity, respectively. The specific process is as follows: Each signal in the multimodal signal and the equivalent thermal inertia feature are treated as an independent piece of evidence. Based on the historical benchmark pattern library of each piece of evidence, the feature vector of each piece of evidence at the current moment is mapped to a confidence interval corresponding to the preset state level (an interval form is used instead of point values ​​because there is uncertainty in the cold chain environment, and the interval form can explicitly express this uncertainty). The confidence interval consists of a lower bound reliability and an upper bound probability, which serve as the basic probability assignment of the interval value of the evidence at that state level. Specifically, for each piece of evidence, the probability density function obtained through offline training is used. Since the 25th and 75th percentiles correspond to the 50% confidence interval, the 25th percentile is taken as the lower bound reliability α, and the 75th percentile is taken as the upper bound probability β, forming the basic probability assignment of the interval value [α, β]. The preset state levels are: S0 (normal), S1 (pre-cooling standby), L1 (mildly compensable), L2 (moderately irreversible), and L3 (severely scrapped). For example, during the cold chain transportation of a certain type of fresh stewed bird's nest, the temperature sensor inside the box showed an ambient temperature of 4.2℃, the temperature sensor at the center of the bottle showed a liquid temperature of 4.5℃, the door magnetic switch detected that the box door had been briefly opened, the accelerometer detected slight vibration, the phase change module temperature was 3.8℃, and the currently calculated normalized equivalent thermal inertia characteristic was 3.3s (the factory baseline was 3.2s, the ratio was 1.03, which is within the normal range). The basic probability assignments of each piece of evidence to the preset state level interval values ​​are shown in Table 1 below: Table 1

[0014] Since the DS evidence theory requires that the sum of the basic probability assignments of all propositions be 1, the remaining confidence is assigned to the "uncertain" state, indicating that the evidence cannot distinguish which specific state it belongs to. Therefore, the lower bound of the confidence interval of each piece of evidence for all preset state levels is calculated. If the sum is less than one, one is subtracted from the sum, and the difference is used as the basic probability assignment of the interval value of the evidence assigned to the uncertain state. This assignment is regarded as the point interval where the lower bound is equal to the upper bound. The basic probability assignments for interval values ​​of uncertain states are shown in Table 2: Table 2

[0015] See Figure 2 When there is a conflict in the determination of the basic probability assignments of different signals within their intervals, a conflict detection factor is introduced for discount correction, specifically: Calculate the maximum difference between the basic probability assignments of any two pieces of evidence to the same state, and use it as the conflict detection factor κ; The conflict detection factor is compared with a preset conflict threshold (0.6): If the value is less than or equal to the preset conflict threshold (κ≤0.6), the original interval value basic probability assignment remains unchanged and is directly output as the discounted interval value basic probability assignment. If the preset conflict threshold (κ>0.6) is exceeded, it is determined that there is a conflict of evidence. The lower and upper bounds of the basic probability assignment of the interval value of each piece of evidence are weighted and reduced according to the reliability coefficient λi of each signal (test samples simulating abnormal working conditions are set for various sensors before leaving the factory, and the accuracy of each piece of evidence in the correct judgment state is calculated as λi). The reduced interval value at each preset state level is obtained. The sum of the lower bounds of all preset state level reductions is calculated and subtracted. The difference is used as the basic probability assignment of the interval value in the uncertain state (considered as a point interval). The interval value after reduction of each state level is combined with the assignment in the uncertain state to obtain the basic probability assignment of the interval value after discount correction for each piece of evidence. The interval evidence combination rule is used for fusion to obtain the fusion confidence interval for each state level, thereby obtaining the initial judgment of an abnormal state and the fusion confidence level characterizing the uncertainty of the initial judgment, specifically: The interval evidence combination rule is as follows: assign basic probabilities to the interval values ​​of any two pieces of evidence, and calculate the fused interval confidence by solving the interval constraint optimization model; The fusion order is based on the reliability coefficient of the evidence from high to low (the most reliable evidence is fused first). That is, each time, the two pieces of evidence with the highest reliability coefficient are selected for fusion. The new evidence obtained after fusion replaces the original two pieces of evidence in the next round of fusion. This process is repeated for a total of five rounds of fusion until all evidence is fused. Finally, the fusion confidence interval for each state level is output. Each fusion confidence interval consists of a lower bound confidence coefficient Bel and an upper bound likelihood coefficient Pl. The fusion confidence intervals of each state level are sorted from largest to smallest according to the lower bound confidence, and the state level with the largest lower bound confidence is selected as the initial abnormal state. Calculate the fusion confidence interval (Pl-Bel) of the initial judgment of the abnormal state, and use the interval width as the fusion confidence Q(t) to characterize the uncertainty of the initial judgment. It is noted that the narrower the interval width, the more consistent the evidence and the more certain the judgment; the wider the width, the more conflicting the evidence and the more uncertain the judgment. Q(t) is used for subsequent state machine decision-making. When Q(t) is too high, a remeasurement is triggered. S4, construct a multi-level state machine that supports bidirectional transfer, and drive state upgrade according to the fusion confidence interval; when the equivalent thermal inertia characteristic quantity recovers to within the baseline threshold and the center temperature inside the bottle is lower than the preset value for several consecutive times, the state machine is allowed to downgrade and transition back from a mild or moderate abnormal state according to the preset recovery conditions; the state after state upgrade or the state after downgrade and transition back is used as the final state level. In this embodiment, the state space of the multi-level state machine includes the normal state S0 (Tc < 4.5℃, Seq < 0.4Smax, D < 1, ...). <1.3 ), pre-cooling standby state S1 (P>0.7 but no actual over-temperature), mildly compensable abnormal state L1 (actual Tc was >4.5℃ but <6℃, and Seq<0.6Smax, D≤2), moderately irreversible abnormal state L2 (Seq≥0.6Smax or D>2 or >1.5 ) and severely scrapped state L3 (Tc>8℃ or D>3 or >2.5 And it is valid twice in a row); The drive state upgrade is specifically achieved through the following transition conditions for state deterioration: When the fusion confidence interval indicates that the initial abnormal state is pre-cooling standby and the fusion confidence is higher than the first threshold (0.7, which, according to ROC curve analysis (for 1000 samples), makes the false positive rate <5% and the sensitivity >85%), the state changes from normal state S0 to pre-cooling standby state S1. When the actual center temperature inside the bottle continuously exceeds the preset value (4.5℃) for a first preset duration (5 minutes, instantaneous overheating of less than 5 minutes has no significant impact on quality), the system switches from the pre-cooling standby state S1 to the mildly compensable abnormal state L1. When the reversible equivalent time Seq exceeds the preset ratio (0.6 times) of the total safe equivalent time (Smax (Smax=240h), or the irreversible high-temperature damage factor D exceeds the first damage threshold (2), or the normalized equivalent thermal inertia characteristic exceeds the first multiple (1.5 times) of the factory baseline ( When L1 is a mildly compensable abnormal state, L2 will transition to a moderately irreversible abnormal state. When the temperature inside the bottle exceeds the scrap temperature threshold of 8℃ (8℃ is the critical point of protein denaturation), or the irreversible high temperature damage factor exceeds the second damage threshold of 3, or the normalized equivalent thermal inertia characteristic quantity exceeds the second multiple (2.5) of the factory baseline twice in a row, it is transferred to the severe scrap state L3. The state machine is allowed to degrade and transition back from a mild or moderate abnormal state according to preset recovery conditions, specifically through the following degrade and transition conditions in the state recovery direction: When the normalized equivalent thermal inertia characteristic measured by two consecutive effective thermal pulses is less than the first baseline multiple (1.3 times), and the center temperature inside the bottle is continuously lower than the preset value of 4.5°C for a second preset duration of 30 minutes, and the irreversible high temperature damage factor is less than the third damage threshold 1, it is allowed to downgrade from the moderate irreversible abnormal state L2 or the mild compensable abnormal state L1 back to the normal state S0. When the normalized equivalent thermal inertia characteristic value is less than the second baseline multiple (1.5 times) for four consecutive times, and the reversible equivalent time recovers to the total safe equivalent time of the preset ratio (0.6 times), it is allowed to downgrade from the moderate irreversible abnormal state L2 back to the mild compensable abnormal state L1. The state machine outputs the updated current state as the final state level based on the above transition conditions and degradation return conditions; the state variables are stored in the Flash circular buffer (3 redundancies) every 10 seconds, which can be recovered after power failure, and the 10-second interval balances storage overhead and recovery accuracy; the 3 redundancies can tolerate single write errors. S5, perform graded compensation according to the final state level: actively pre-cool the phase change energy storage module during pre-cooling standby; when the fusion confidence is lower than the cancellation threshold and the current bottle center temperature is lower than the preset recovery temperature threshold, reverse heating is used to restore the cold storage capacity; when the condition is slightly compensable, the subsequent allowable ambient temperature is compressed; when the condition is moderately irreversible, the dynamic remaining shelf life is calculated and the QR code is updated; when the condition is severely scrapped, an alarm is issued and compensation is disabled. In this embodiment, a differentiated compensation strategy is adopted for different levels of anomalies to avoid overcompensation or undercompensation and achieve precise intervention. Specifically, a pattern matching method is used to predict the future ambient temperature. Specifically, three templates are pre-stored: cooling cycle, door opening, and constant temperature. The cooling cycle template is a sinusoidal fluctuation with a period of 20 minutes and a peak-to-peak value of 2℃, expressed as follows: Tamb ( t )= Tavg + A sin(2 πt / T ), where T=1200s, A=1℃ (peak-to-peak value 2℃), the typical start-stop cycle of the refrigeration compressor in a cold chain truck is 15-30 minutes, and the median value of 20 minutes is taken; the temperature fluctuation inside the chamber was measured to be ±1℃; The door opening temperature indicator shows an instantaneous jump of +1.5°C, followed by an exponential recovery. In actual testing, opening the door for 1 minute caused the internal temperature to rise by 1-2°C, with a recovery time of approximately 5 minutes (300 seconds). The recovery time constant is 1.5℃. For 300 seconds; The temperature template is constant; Calculate the normalized cross-correlation between the real-time ambient temperature sequence and each template. The model with the highest correlation coefficient is selected as the current model; when all correlation coefficients are below 0.6, the ARIMA(1,1,0) model is used for prediction. When the Hall effect magnetic switch detects that the door is open, it directly sets the predicted ambient temperature for the next 3 minutes to the current temperature plus 2°C, covering the pattern matching result. Use the next 90 points (15 minutes) corresponding to the current model as the ambient temperature input. Iteration using a discrete state-space model (step size Δt = 1s): ; ; in, The time was 60 seconds, determined by a step response experiment. The empty box was placed in an environment of 25°C, and the time it took for the wall temperature to reach 63.2% of the final temperature from the room temperature to 5°C was recorded to be approximately 60 seconds. The maximum predicted center temperature T_max_pred is obtained from the predicted sequence, and the standard deviation of the prediction error σ is updated online: the initial σ = 0.2℃ (from the standard deviation of 100 prediction errors in the preliminary experiment). Every 30 seconds, the error e between the current predicted value and the measured value after 30 seconds is calculated, and the variance is updated using an exponentially weighted moving average. ; A smoothing factor of 0.95 corresponds to approximately 13 updates (6.5 minutes) in half-life. The probability of overheating is calculated as follows: ; Φ is the standard normal CDF, using the approximate formula Calculate quickly on the MCU (error < 0.01) and output P value (0~1); The specific process of actively precooling the phase change energy storage module during precooling standby is as follows: The phase change energy storage module uses a paraffin-expanded graphite composite (80:20), with a melting point of 3.2℃, a latent heat of 180J / g, and a total mass of 30g. When the predicted overheat probability is higher than the preset high threshold P>0.7, the ambient temperature is higher than the preset ambient temperature threshold T_amb>5℃, and the current temperature of the phase change energy storage module is higher than the preset supercooling temperature limit T_PCM>2.5℃, the precooling is performed with a soft start. The pulse width modulation duty cycle increases from a low percentage of 20% to a high percentage of 80%, and each level is kept stable for a short time (0.25 seconds). The goal is to reduce the temperature of the phase change energy storage module to below the preset supercooling temperature limit (2.5℃) or to continue cooling for no more than the preset maximum cooling time (6 minutes), and then output the cooling action. The cancellation condition for the reverse heating to restore the cold storage capacity is as follows: when the fusion confidence is lower than the cancellation threshold of 0.3 and the current center temperature inside the bottle is lower than the preset recovery temperature threshold (Tc<4℃), the direction of the semiconductor cooling chip current is reversed, and the phase change energy storage module is heated with a low duty cycle of 10% until its temperature rises back to the preset cold storage recovery temperature of 3.0℃, and the heating action is output. The specific process of compressing the subsequent allowable ambient temperature when it is slightly compensated is as follows: Calculate the reversible equivalent time Accumulated time at an ideal constant temperature of four degrees Celsius The difference, ,in, Accumulated according to the Arrhenius model ,in, For 5 / 60 hours, θ is the factory-calibrated personalized temperature coefficient, applicable only when... Damage ≤7℃ is included; damage exceeding 7℃ is irreversible and D is calculated separately. Based on remaining transportation time (Subtract the time already used from the total estimated time at departure; if this is not available, the remaining time is assumed to be 0.5 times the total time.) Adjust the subsequent maximum allowable ambient temperature using the factory-set temperature coefficient θ. For example, ΔS=0.8h, θ=0.12, =10h, then =4-0.8 / (0.12×10)=3.33℃, and send this temperature setpoint to the cold chain transportation terminal via wireless communication, outputting the compressed ambient temperature control command; The specific process for calculating the dynamic remaining shelf life under moderate irreversibility is as follows: Based on the current accumulated reversible equivalent time, total safe equivalent time S_max (determined to be 240 hours through destructive testing), irreversible high-temperature damage factor D, and its threshold D_th (accumulated according to D = Dprev + Δthour·[max(0, Tc-7)]², where Δthour=5 / 60=0.08333 hours, and threshold D_th=3℃²·h), when D>3℃²·h, it is directly judged as a severely scrapped state. The remaining shelf life is calculated as follows: the difference between the total safe equivalent time and the reversible equivalent time is divided by 24, then multiplied by 1 and the maximum value between the ratio of the irreversible high-temperature damage factor to its threshold and 0.3 is subtracted. The formula for calculating the remaining shelf life is: Rremain = ((Smax - Seq) / 24) × max(0.3, 1 - The unit is D / Dth), rounded to the nearest day. For example, if Seq=150h, Smax=240h, and D=2, then (90 / 24)=3.75 days. The discount factor is 1-2 / 3=0.333. Taking max(0.3,0.333)=0.333, we get 1.25 days → rounded to the nearest day. The calculation result is then encoded into a QR code (JSON string). After scanning the code, the consumer will see the message "This product has experienced a short-term temperature fluctuation. Please consume it within the next X days." The updated QR code image data will then be output. When the vehicle is severely damaged and scrapped, an alarm will be triggered and compensation will be disabled. S6 further includes online incremental fine-tuning of the basic probability assignment function for the interval values, the method of which is as follows: Taking a single transportation task as the unit, the execution hit rate and compensation accuracy of the graded compensation in that task, as well as the measured results of the quality of fresh stewed bird's nest after delivery (including total bacterial count, sensory evaluation, and viscosity decay rate) are collected as feedback samples. When the number of accumulated feedback samples reaches the preset batch threshold (32), a fine-tuning process is triggered: only the parameters of the last layer of confidence mapping in the basic probability assignment function of the interval value are adjusted (all parameters of the preceding feature extraction layer are frozen), and a regularization constraint term to prevent forgetting previous knowledge is added during the fine-tuning process. After fine-tuning, the updated confidence mapping parameters are synchronized to the edge computing unit of the same batch of products via wireless communication for anomaly detection and compensation decisions in subsequent transportation tasks.

[0016] Therefore, through a dual preprocessing architecture coupling multimodal signal acquisition and factory baseline calibration, six types of signals are collected: bottle center temperature, bottle wall dual-redundant temperature, ambient temperature and humidity, acceleration impact, door magnetic switch, and phase change module thermal state. Spatiotemporal consistency calibration is then completed, enabling precise acquisition of the bird's nest liquid thermal field distribution, mechanical disturbances, door events, and energy storage status. Compared to traditional single ambient temperature monitoring methods that suffer from center temperature lag and lack of disturbance perception, the comprehensiveness and real-time performance of cold chain anomaly detection are significantly improved. Furthermore, by utilizing cross-correlation thermal pulse identification and temperature normalization compensation mechanisms, the input wall temperature and center temperature response signals are used to output normalized equivalent thermal data. Inertial characteristics accurately eliminate the interference of temperature on thermal inertial measurement, breaking the limitation of traditional passive temperature measurement that cannot detect changes in the internal state of the liquid, and greatly improving the sensitivity of identifying precursors of internal deterioration such as gelation, precipitation, and bubbles; through the interval evidence fusion and conflict discount correction architecture, multimodal signals and thermal inertial characteristics are input, and the fusion confidence interval and fusion confidence of each state level are output, realizing decision conflict handling of multi-source heterogeneous sensors in complex cold chain environment. Combined with explicit modeling of uncertain states, the initial judgment of abnormal state and confidence index are output, avoiding the shortcomings of traditional single sensor threshold judgment that is susceptible to environmental interference and decision failure when evidence conflicts occur; Based on the linkage design of bidirectional state machine and thermal inertia self-recovery, the input is the fusion confidence interval and fusion confidence signal, and the output is the final state level to drive the graded compensation execution. At the same time, the state machine is degraded and transferred back according to the recovery of normalized thermal inertia characteristic quantity. It accurately adapts to the decision-making needs of automatic recovery of reversible anomalies and graded handling of irreversible anomalies, and solves the adaptation problem of traditional cold chain monitoring where anomalies are scrapped and cannot be self-recovered. By modeling the equivalent time accumulation and irreversible damage separation, the system inputs the bottle's internal center temperature and the factory-calibrated temperature coefficient, and outputs the reversible equivalent time and irreversible high-temperature damage factor. This achieves the separation of the reversible and irreversible components of the temperature history's impact on quality. Combined with the dynamic remaining shelf life calculation formula and QR code output, it provides consumers with dynamic shelf life information based on real thermal history, avoiding the scrapping of the entire batch due to slight overheating. Through multi-level state machines and graded compensation closed-loop linkage, differentiated treatment is achieved for four levels: pre-cooling standby, mild compensable, moderate irreversible, and severe scrap. During pre-cooling standby, active pre-cooling is based on over-temperature probability prediction, and reverse heating is performed to restore cold storage capacity when the risk is eliminated. During mild compensable, the subsequent allowable ambient temperature is reduced. During moderate irreversible, the dynamic remaining shelf life is calculated and the QR code is updated. During severe scrap, an alarm is issued and compensation is disabled, forming a complete closed loop from anomaly detection to compensation execution and state recovery. Meanwhile, factory baseline calibration and QR code binding enable personalized thermal fingerprint registration for each bottle of bird's nest. Real-time data acquisition by the edge computing unit and closed-loop control by the MCU ensure signal transmission and collaborative stability for anomaly detection, status decision-making, and compensation execution, significantly reducing the management costs of manual temperature recording, post-event traceability, and batch disposal. Furthermore, this method does not rely on high-end dedicated hardware or complex manual operations, lowering the threshold for industrial application. It provides reliable technical support for the entire process management of anomaly detection, proactive compensation, and quality traceability in the cold chain transportation of freshly stewed bird's nest, driving the high-end freshly stewed bird's nest industry to upgrade from passive monitoring to proactive intelligent compensation.

[0017] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related explanations of the methods described above can be found in the corresponding processes in the foregoing system embodiments, and will not be repeated here.

[0018] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0019] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0020] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0021] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for detecting and compensating for anomalies in cold chain temperature data of freshly stewed bird's nest, characterized in that, Includes the following steps: Multimodal signals of fresh stewed bird's nest during cold chain transportation are collected. The multimodal signals include the bottle center temperature signal, bottle wall temperature signal, mechanical disturbance signal, box door status signal, and thermal status signal of the phase change energy storage module. Short-duration heat pulses are periodically applied to the wall of the bird's nest bottle. Based on the temperature signal at the center inside the bottle and the temperature signal on the bottle wall, the equivalent thermal inertia characteristic of the fresh stewed bird's nest liquid is calculated using the cross-correlation method. The basic probability assignment of the interval values ​​of the current state level is calculated for the multimodal signal and the equivalent thermal inertia feature quantity respectively. The preset state levels include normal, pre-cooling standby, slightly compensable, moderately irreversible, and severely scrapped. When there is a judgment conflict between the basic probability assignment of the interval values ​​of different signals, a conflict detection factor is introduced for discount correction, and then the interval evidence combination rule is used for fusion to obtain the fusion confidence interval of each state level, thereby obtaining the initial judgment of abnormal state and the fusion confidence level characterizing the uncertainty of the initial judgment. A multi-level state machine supporting bidirectional transitions is constructed, and state upgrades are driven according to the fusion confidence interval. When the equivalent thermal inertia characteristic quantity recovers to within the baseline threshold and the center temperature inside the bottle is lower than the preset value for several consecutive times, the state machine is allowed to downgrade and transition back from a mild or moderate abnormal state according to the preset recovery conditions. The state after state upgrade or the state after downgrade and transition back is taken as the final state level. Based on the final state level, graded compensation is performed: during pre-cooling standby, the phase change energy storage module is actively pre-cooled; when the fusion confidence is lower than the cancellation threshold and the current center temperature inside the bottle is lower than the preset recovery temperature threshold, reverse heating is used to restore the cold storage capacity; when the condition is slightly compensable, the subsequent allowable ambient temperature is compressed; when the condition is moderately irreversible, the dynamic remaining shelf life is calculated and the QR code is updated; when the condition is severely scrapped, an alarm is issued and compensation is disabled.

2. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The specific process for calculating the equivalent thermal inertia characteristic quantity using the cross-correlation method is as follows: The bottle wall temperature signal before and after the application of a short-duration thermal pulse is cross-correlated with the bottle center temperature signal to obtain the pulse response sequence. Calculate the centroid of the impulse response sequence as the measured equivalent time constant; A correction factor based on the temperature change rate of the thermal diffusivity of the freshly stewed bird's nest liquid is used to normalize the measured equivalent time constant to obtain the normalized equivalent thermal inertia characteristic quantity, which is then used as the equivalent thermal inertia characteristic quantity.

3. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The basic probability of the current state level being in the preset state level is calculated and assigned to the multimodal signal and the equivalent thermal inertia characteristic quantity, respectively. The specific process is as follows: Each signal in the multimodal signal and the equivalent thermal inertia feature are treated as an independent piece of evidence. Based on the historical benchmark pattern library of each piece of evidence, the feature vector of each piece of evidence at the current moment is mapped to the confidence interval corresponding to the preset state level. The confidence interval consists of a lower bound reliability and an upper bound probability, which are used as the basic probability assignment of the interval value of the evidence at the state level. Calculate the sum of the lower bounds of the confidence intervals for each piece of evidence for all preset state levels. If the sum is less than one, subtract the sum from one and use the difference as the basic probability assignment of the interval value to the uncertain state for that piece of evidence. This assignment is regarded as a point interval where the lower bound is equal to the upper bound.

4. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, When there is a conflict in the determination of the basic probability assignments of different signals within their intervals, a conflict detection factor is introduced for discount correction, specifically: Calculate the maximum difference between the basic probability assignments of any two pieces of evidence to the same state, and use it as the conflict detection factor; The conflict detection factor is compared with a preset conflict threshold: If the value is less than or equal to the preset conflict threshold, the original interval value basic probability assignment remains unchanged and is directly output as the discounted interval value basic probability assignment. If the preset conflict threshold is exceeded, it is determined that there is a conflict of evidence. The lower and upper bounds of the basic probability assignment of the interval value of each piece of evidence are weighted and reduced according to the reliability coefficient of each signal in the pre-calibrated signal, so as to obtain the reduced interval value at each preset state level. The difference between the sum of the lower bounds of all preset state levels after reduction is calculated and used as the basic probability assignment of the interval value at the uncertain state. The basic probability assignment of each evidence discount-corrected interval value is obtained by combining the interval values ​​after the reduction of each state level with the assignments on the uncertain state.

5. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The evidence is combined using interval rules to obtain the fusion confidence intervals for each state level, thereby obtaining the initial judgment of an abnormal state and the fusion confidence level representing the uncertainty of this initial judgment, specifically: The interval evidence combination rule is as follows: assign basic probabilities to the interval values ​​of any two pieces of evidence, and calculate the fused interval confidence by solving the interval constraint optimization model; The fusion order is carried out in descending order of evidence reliability coefficients. That is, each time, the two pieces of evidence with the highest current reliability coefficients are selected for fusion. The new evidence obtained after fusion replaces the original two pieces of evidence in the next round of fusion. This process is repeated for a total of five rounds of fusion until all evidence is fused. Finally, the fusion confidence interval for each state level is output. Each fusion confidence interval consists of a lower bound confidence level and an upper bound likelihood level. Sort the fusion confidence intervals of each state level from largest to smallest according to their lower bound confidence, and select the state level with the largest lower bound confidence as the initial judgment of an abnormal state; calculate the interval width of the fusion confidence interval of the initial judgment of an abnormal state, and use the interval width as the fusion confidence level that characterizes the uncertainty of the initial judgment.

6. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The drive state upgrade is specifically achieved through the following transition conditions for the state deterioration direction: When the fusion confidence interval indicates an initial abnormal state of pre-cooling standby and the fusion confidence is higher than the first threshold, the system transitions from the normal state to the pre-cooling standby state. When the actual internal center temperature of the bottle continuously exceeds the preset value for a first preset duration, the system transitions from the pre-cooling standby state to a mildly compensable abnormal state. When the reversible equivalent time exceeds a preset proportion of the total safe equivalent time, or the irreversible high-temperature damage factor exceeds the first damage threshold, or the normalized equivalent thermal inertia characteristic exceeds the first multiple of the factory baseline, the system transitions from a mildly compensable abnormal state to a moderately irreversible abnormal state. When the internal center temperature of the bottle exceeds the scrap temperature threshold, or the irreversible high-temperature damage factor exceeds the second damage threshold, or the normalized equivalent thermal inertia characteristic exceeds the second multiple of the factory baseline twice consecutively, the system transitions to a severely scrapped state.

7. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The state machine is allowed to degrade and transition back from a mild or moderate abnormal state according to preset recovery conditions, specifically through the following degrade and transition conditions for state recovery direction: When the normalized equivalent thermal inertia characteristic measured by two consecutive effective thermal pulses is less than a multiple of the first baseline, and the center temperature inside the bottle remains below the preset value for a second preset duration, and the irreversible high-temperature damage factor is less than the third damage threshold, a downgrade from a moderately irreversible abnormal state or a slightly compensable abnormal state back to a normal state is permitted; when the normalized equivalent thermal inertia characteristic is less than a multiple of the second baseline for four consecutive times, and the reversible equivalent time recovers to below the total safe equivalent time of the preset proportion, a downgrade from a moderately irreversible abnormal state back to a slightly compensable abnormal state is permitted. The state machine outputs the updated current state as the final state level based on the aforementioned transition conditions and degradation / return conditions.

8. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The specific process of actively precooling the phase change energy storage module during precooling standby is as follows: When the predicted overheat probability is higher than the preset high threshold, the ambient temperature is higher than the preset ambient temperature threshold, and the current temperature of the phase change energy storage module is higher than the preset supercooling temperature upper limit, the precooling is performed with a soft start. The pulse width modulation duty cycle is gradually increased from a low percentage to a high percentage, and each level is kept stable for a short time. The goal is to reduce the temperature of the phase change energy storage module to below the preset supercooling temperature lower limit or to continue cooling for no more than the preset maximum cooling time, and then output the cooling action. The cancellation condition for the reverse heating to restore the cold storage capacity is as follows: when the fusion confidence is lower than the cancellation threshold and the current center temperature inside the bottle is lower than the preset recovery temperature threshold, the direction of the semiconductor cooling chip current is reversed, and the phase change energy storage module is heated with a low duty cycle until its temperature rises back to the preset cold storage recovery temperature, and the heating action is output.

9. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, The specific process of compressing the subsequent allowable ambient temperature when it is slightly compensated is as follows: The difference between the reversible equivalent time and the ideal four-degree constant temperature cumulative time is calculated. The subsequent maximum allowable ambient temperature is adjusted according to the remaining transportation time and the temperature coefficient specified at the factory. The temperature setpoint is then sent to the cold chain transportation terminal via wireless communication, and the compressed ambient temperature control command is output. The specific process for calculating the dynamic remaining shelf life under moderate irreversibility is as follows: Based on the current accumulated reversible equivalent time, total safe equivalent time, irreversible high-temperature damage factor and its threshold, the remaining shelf life is calculated by dividing the difference between the total safe equivalent time and the reversible equivalent time by 24, multiplying by 1 and subtracting the maximum value between the ratio of the irreversible high-temperature damage factor and its threshold and 0.

3. The unit is days and rounded down. The calculation result is encoded into a QR code and output as the updated QR code image data.

10. The method for anomaly detection and compensation of cold chain temperature data of freshly stewed bird's nest according to claim 1, characterized in that, It also includes online incremental fine-tuning of the basic probability assignment function for the interval values, the method of which is as follows: Taking a single transportation task as the unit, the execution hit rate, compensation accuracy, and measured results of the quality of fresh stewed bird's nest after delivery are collected as feedback samples. When the number of accumulated feedback samples reaches the preset batch threshold, a fine-tuning process is triggered: only the parameters of the last layer of confidence mapping in the basic probability assignment function of the interval value are adjusted, and a regularization constraint term to prevent forgetting previous knowledge is added during the fine-tuning process. After fine-tuning, the updated confidence mapping parameters are synchronized to the edge computing unit of the same batch of products via wireless communication for anomaly detection and compensation decisions in subsequent transportation tasks.

Images