Self-adaptive multi-source heat management medicine ultralow temperature cold chain intelligent box system

Abstract

The invention discloses a self-adaptive multi-source heat management medicine ultralow-temperature cold chain intelligent box system which aims at solving the problems that an existing dry refrigerator, a liquid nitrogen tank and a single-stage semiconductor box body are poor in temperature control precision, short in endurance, high in energy consumption and difficult in data tracing. An integrated scheme of honeycomb phase change partitioning, four-stage TEC / TEG coupling, increment GRU-PID temperature control, [delta] V-SOC-[delta] T energy routing and double-chain traceability is provided. The system is composed of a multi-layer SiOaerogel heat insulation layer, a honeycomb type PCMS plate with the melting point being 63 DEG C, a four-stage thermoelectric refrigeration-power generation assembly, a lithium iron phosphate battery-super capacitor energy module, a Cat-1 / LoRa / FSK three-mode communication and posture-temperature and humidity three-stage safety state machine. The MCU samples multi-source data in a 20 ms period, the thermal load of 30 min is predicted through IIR-Kalman filtering and increment GRU, the duty ratio of PID driving TEC is self-tuned, and four-source energy is routed in real time. Experimental results show that + / -0.5 DEG C fluctuation is achieved in a target temperature zone of 80 DEG C, independent endurance is achieved for 168 h, unit energy consumption is 0.95 Wh.h.dm, data uplink can still be completed within 10 s after link disconnection, and the requirements of WHO PQS and ISO 21973-2020 are met.

Description

Technical Field

[0001] The present invention relates to pharmaceutical cold chain transportation and storage equipment technology, in particular to an adaptive multi-source thermal management pharmaceutical ultra-low temperature cold chain intelligent box system that can maintain the target temperature zone within the range of -80°C to +25°C for a long time and has energy recovery and self-learning temperature control capabilities. Background Art

[0002] With the commercialization of high-value biologics such as CAR-T cells, mRNA vaccines, and recombinant adenoviral vector-based drugs, the pharmaceutical cold chain's temperature range has expanded from the traditional +2°C to +8°C to -60°C, -70°C, and even below -80°C. WHO TRS 961, ICH Q5A(R2), and the FDA Guidance for Industry: Transporting Temperature-Sensitive Pharmaceutical Products all require traceable control of the core temperature of the pharmaceutical product within ±2°C throughout transportation, and impose a maximum deviation of no more than 5°C for the temperature difference (ΔT) between each measuring point within the box. Existing commercial equipment that meets ultra-low temperature requirements falls into three main categories: The first is a passive box using a combination of dry ice and vacuum insulation panels. Dry ice has a high latent heat of sublimation (571 kJ·kg⁻¹) and can provide 48–72 hours of insulation at a self-stable temperature of -78.5°C. However, the CO₂ partial pressure inside the container can rise from 0.04 atm to 1.2–1.5 atm within 6 hours, increasing the risk of internal pressure explosion and requiring air transport as a UN 1845 hazardous material (an additional qualification). Dry ice consumption scales linearly with the external heat load, and long-distance summer shipments often require >16 kg / 20 L, directly increasing logistics costs and carbon emissions. A second alternative is liquid nitrogen vaporizer (Dewar) transfer tanks. Liquid nitrogen has a vaporization point of -196°C and a latent heat of phase change of 199 kJ·kg⁻¹. Using a porous adsorption medium, it can achieve a 7–10 day flight duration in a low vacuum. However, it is classified as a UN 1977 hazardous material and carries numerous aviation restrictions, requiring antifreeze and nitrogen protection equipment during loading and unloading. More critically, the weight-to-volume ratio of liquid nitrogen tanks is typically greater than 0.9 kg·L⁻¹. For a 16 L core container, the entire unit can weigh up to 30 kg, making it ergonomically inefficient for intra-city or end-of-line delivery. Thirdly, single-stage TEC active cooling boxes. A Peltier chip can theoretically achieve a 67°C temperature differential at 12 V and I = 6 A. However, in a practical box, the cold end temperature cannot drop below -45°C with an outer wall temperature of 35°C. Reaching -70°C requires multiple chips in series. Furthermore, the exponential growth of the TEC drive current makes it difficult to achieve an average COP exceeding 0.2 (IEC 60072 test, 35°C → -60°C). Consequently, the battery life of a box of the same volume is often less than 12 hours, requiring vehicle DC power or frequent battery replacement. Furthermore, insufficient heat dissipation from the TEC hot end can easily lead to "heat leakage backflow," resulting in a temperature gradient > 8°C at the box core.The above schemes have exposed three common problems in actual operation: ① Poor temperature zone uniformity - the passive box relies on a single phase change material, and the active box relies on a single refrigeration unit, which makes it difficult to take into account the heat loads at different levels of the box; ② Unpredictable endurance - dry ice and batteries are greatly affected by the external climate, load capacity and compartment conditions, and cannot provide accurate estimates of remaining time; ③ Low energy utilization efficiency - dry ice completely sublimates and the battery discharges unidirectionally, and the temperature difference between the environment and the box cannot be converted into usable electricity, resulting in high energy consumption per unit of cargo value. According to the 2024 IMS Health report, the energy cost of cross-provincial transportation of mRNA vaccines has reached 7.8% of the product's ex-factory price. Current research has begun to explore coupling multi-stage TECs and phase-change energy storage panels with onboard power supplies, but this approach remains limited to a "multi-module stacking" approach, lacking a closed-loop thermal management system for the entire lifecycle. Furthermore, there is limited public documentation on the coordinated use of TEC cooling and TEG (thermoelectric generator) energy recovery within a single enclosure, with only a 2023 IEEE JESTPE report on an experimental prototype for a -30°C fry enclosure, which has yet to break through the -70°C pharmaceutical-grade temperature range. To address these technological gaps, this paper proposes an ultra-low-temperature cold chain smart enclosure featuring cellular PCMS zoning, four-stage TEC / TEG coupling, self-learning dual-mode temperature control, and multi-source cascaded power supply, aiming to achieve substantial improvements in temperature uniformity, energy efficiency, and endurance predictability.

[0003] In addition to the difficulty of maintaining temperature zones, ultra-low temperature cold chain equipment also faces structural shortcomings in energy management and real-time monitoring. Current active boxes generally use a single lithium iron phosphate or ternary lithium battery as their primary power source, with an energy density of 140–170 Wh kg⁻¹. However, at -40°C, the discharge capacity is reduced to only 45%–52%. The IEC 62133-2 low-temperature discharge curve shows that a 4× 90 Wh module must operate at a current of ≤0.2 C at -60°C. Otherwise, a voltage drop below 2.75 V will trigger a BMS shutdown, instantly interrupting the box's refrigeration chain. To prevent extreme power outages, operators typically install additional 12 V / 24 V DC power sources or portable gasoline generators within the vehicle compartment, but this increases reliance on vehicle specifications and fuel regulations. Recent research has attempted to incorporate thin-film TEGs to decouple hot-end waste heat from the cooling element, theoretically recovering 6–8% of thermal power. However, due to rare temperatures exceeding 55°C on the hot-end heat sink and insufficient temperature difference ΔT, the TEG output power of nearly 70% of test prototypes was insufficient to maintain standby communication between the MCU and Cat-M. This urgently requires a "high-throughput, low-resistance" thermal coupling design and a multi-source cascade scheduling strategy. At the control level, most commercial devices still use single-loop or cascaded PID control of the TEC PWM, with a duty cycle modulation resolution of 0.5–1.0 Hz. When faced with thermal load transients such as vehicles entering tunnels, sudden rain, or intense sunlight, traditional PIDs require 120–180 seconds to converge, resulting in an overshoot of 2–4°C. The IATA CEIV Pharma Audit Report (2023) states that 73% of temperature deviation events occur "within 8 minutes of a dramatic change in the external temperature of the cabinet," and 61% are accompanied by a rapid increase in humidity, leading to frosting of the label and damage to the electronic record. Existing algorithms rarely utilize prior information such as route weather, vehicle speed, and altitude for feedforward prediction. Only a few research prototypes incorporate LSTM technology into temperature control, but these model parameters require over an hour of cloud-based training, making incremental learning impossible in network outages and limiting their effectiveness in actual deployments. Monitoring and communication are also weaknesses. NB-IoT solutions have blind spots of 15–40 seconds on bridges and tunnels. While Cat-M1 can guarantee cell handover at 80 km / h⁻¹, the average reconnection time after a disconnection in white zones (areas with sparse base stations) is 28 seconds. However, the pharmaceutical abnormality alarm regulation (EU GDP 2013 / C343 / 01) stipulates that "temperature excursion signals should be accessible within 10 seconds." Many operators are piloting 5G-RedCap, but the cost of chips and modules is much higher than Cat-1, and the current power consumption does not yet meet the 24-hour independent monitoring requirements of medical portable devices.Finally, from a lifecycle economic perspective, disposable materials for dry ice solutions account for 41% of the total cost, electricity and battery depreciation for a single-stage TEC box account for 37%, and calibration and compliance for the liquid nitrogen tanks account for 26%. In response, the 2024 WHO PQS draft explicitly encourages active cold chain approaches that "convert ambient temperature differences into effective electrical energy" and proposes performance benchmarks of "energy consumption per cubic decimeter of box volume ≤ 1.2 Wh·h⁻¹, steady-state fluctuation ≤ ±0.5°C, and core-to-wall ΔT ≤ 3°C." However, no commercially available product currently fully meets these requirements. Summary of the Invention

[0004] The adaptive multi-source thermal management pharmaceutical ultra-low temperature cold chain smart box system proposed in this invention is composed of core components, including an outer shell assembly, a honeycomb composite insulation layer, a three-stage thermal zone cavity, a four-stage TEC-TEG thermoelectric coupling module, a liquid-cooled radiant plate and hot-end manifold, an integrated energy cascade management module, an environmental monitoring array, and an IoT communication and control unit. The outer shell is constructed from a double layer of 7075-T6 aviation aluminum-magnesium alloy profiles and 0.7 mm thick carbon fiber reinforced resin composite panels. The overall frame is formed through high-frequency friction stir welding and a low-temperature adhesive layer. The outer panels are treated with a PVD titanium nitride coating, reducing their solar absorptivity α_s to 0.21. This allows the outer panels to maintain a steady-state temperature below 53°C under 950 W·m⁻² irradiation, providing a safety margin for heat dissipation from the hot end. The frame is filled with a three-layer SiO2 aerogel core with an average pore size of 20 nm. Its room-temperature thermal conductivity is 0.012 W·m⁻¹·K⁻¹, 1 / 15th that of traditional PU foam. A 0.5 mm flexible graphite sheet is inserted between the aerogel and the outer panels to eliminate warping stress and quickly diffuse solar heat within the plane, reducing the risk of localized overheating. The insulation layer, facing inward, forms an integrated "aerogel-PCMS-air" sandwich structure with a high-melting-enthalpy phase-change composite material (PCMS) honeycomb panel. The PCMS utilizes a triol-decanoic acid / hexadecane mixture with a melting point of -63°C and a latent heat of fusion of 210 kJ·kg⁻¹. After emulsification and coating at 7070 rpm, it is filled into the hexagonal honeycomb cells and secured to the frame via welded joints. Compared to traditional vacuum insulation panels, this structure reduces the hot-cold surface temperature difference by 28% at -70°C and eliminates the risk of structural failure caused by internal and external pressure differentials during high-altitude transportation. The interior of the chamber is divided into a core drug compartment, an annular buffer compartment, and an outer ring environmental isolation compartment in a 5:3:2 volume ratio. The walls, bottom, and top cover of the drug compartment are inlaid with 1 mm thick, highly thermally conductive AlN ceramic plates with a thermal conductivity of 180 W·m⁻¹·K⁻¹. This allows cooling to rapidly distribute to the center of the interior volume along all three dimensions, limiting the measured steady-state temperature difference between the core and the inner wall to less than 1.4°C. A 3mm-thick, removable PCMS sheet is installed inside the annular buffer chamber to preemptively absorb sensible heat in the event of a sudden increase in external heat load, providing the cooling unit with a 3–5 minute response window. The outer ring isolation chamber primarily serves to extend the heat conduction path and cascade temperature differences, while also serving as a pre-cooling zone for the hot-end heat dissipation duct. Four-stage TEC-TEG thermoelectric coupling modules are installed on the outer walls of the core chamber. Each module consists of four 40mm×40mm Bi2Te3-based PN stacked semiconductor cooling sheets and three 220µm-thick Cu / Ni thin-film TEG sheets.The cold end of the refrigeration chip is thermally coupled to the chamber wall via a self-circulating microchannel liquid cooling plate containing 47% by mass EG-water. The hot end is bonded to a 68 cm² micro-grooved aluminum alloy heat sink via a 500 µm-thick brazed graphite pad. A thin TEG layer is inserted between the liquid cooling plate and the heat sink. At a 45°C temperature gradient, a single TEG layer outputs 2.8 W (0.52 V / 5.4 A). This output is fed through a synchronous boost-DC bus, resulting in a power recovery of 8–11 J per minute. The entire TEC stack can extract 48 W of heat flux at 12 V and 5.8 A. The cold end temperature limit is -78°C, while the hot end temperature is 49 ± 2°C, allowing a 14°C melting range buffer for PCMS phase transitions. The energy cascade management module, located in a corner of the outer ring isolation compartment, utilizes four 92 Wh lithium iron phosphate (LiFePO4) pluggable sliced ​​batteries as the primary power source. A TEG bus, an onboard 12V DC, and a 120 F supercapacitor are connected in parallel. MCU routing, based on a multivariable ΔV-SOC-ΔT model, determines if the primary power source discharge rate exceeds the limit and dynamically switches the energy path. The supercapacitor absorbs peak current surges from 0.5 s to 3 s, suppressing bus voltage fluctuations to less than 0.25 V during TEC PWM duty cycle transitions. This cascade design achieves 168 hours of independent battery life under a constant temperature load of 60°C for the outer wall and -70°C for the core. When combined with onboard DC power, this range can be extended to 340 hours in intermittent charging mode. The environmental monitoring array consists of 12 four-wire PT100 Class A platinum resistors and six digital humidity CMOS sensors, evenly spaced 120° across three thermal zones. Two optical-infrared hybrid radiometers and a three-axis IMU are located outside the box for attitude and impact detection. All sensor data is fed into the Cortex-M33 control core at a 250 ms cycle and cached in 2 MB FRAM, ensuring complete data traceability for 30 days in the event of a power outage. The communication-control unit utilizes an LTE Cat-1 dual-antenna module as the primary link, supporting 70 minutes of eDRX, 3.2µA PSM sleep, and 23 dBm uplink power. In the event of a network outage, it automatically falls back to 620 bps. W-Mbus LoRa-PHY self-organizing networking and neighboring box relay capabilities are also included. The flexible 4.7-inch E-Ink panel is concealedly attached to the inside of the box lid, which can maintain the most recent five-dimensional status information for ≥30 days in a passive state, and cooperate with the two-color LED strip to carry out 360° environmental alarm.

[0005] This system incorporates an integrated prediction-feedback-scheduling software core within the hardware thermal chain closed loop. The core concept is to enable the container to perceive its own and environmental conditions in real time during the transport cycle, just like an adaptive heat engine. An incremental time-series model is used to estimate future heat loads. This estimate and the real-time error are then integrated into an adaptive PID controller to mitigate transients. Simultaneously, the four energy sources—refrigeration, power generation, onboard power supply, and capacitor buffering—are dynamically reconfigured into a switchable DC microgrid. The entire process begins with a 20 ms sampling interrupt. Twenty-three signals, including temperature and humidity, voltage and current, IMU, and radiation, are filtered through an IIR-Kalman cascade filter and output to a 128-byte sliding window buffer. The control core first uses a lightweight GRU (Gated Recurrent Unit) tailored on the MCU to perform incremental inference on the past 128 seconds of data. This model has 24k parameters, and a single inference takes only 3.8 ms. The inference result is a 30-minute heat load curve and a 2.5-minute "rapid warning zone" vector. The cabinet then maps this predicted curve into a target cold-end temperature trajectory and feeds this, along with the real-time measured difference, into a self-tuning PID controller. The PID's proportional, integral, and differential coefficients are no longer fixed. Instead, they are adjusted online using a two-layer perceptron based on the sum of squared residual errors, the core-wall gradient, and the slope of the predicted curve. This allows for fine-grained adjustment of the PWM duty cycle to the 0.5% level at 1 Hz resolution. The energy-side routing logic and the temperature control circuit share a common ΔV-SOC-ΔT state equation. The algorithm first roughly classifies cooling power levels based on the battery state of charge and the hot-end-cold-end temperature difference, then prioritizes energy replenishment paths based on the TEG's real-time power, the vehicle's DC available power, and the supercapacitor's remaining charge. If the thermal load is predicted to increase by more than 20% within 5 minutes, the router precharges the supercapacitor to the upper limit of 80% to prevent voltage collapse. If the load is determined to drop below 60% within 10 minutes, the TEC automatically switches to pulse modulation mode, recharging excess current to the battery pack. In actual road tests, the peak-to-valley power ratio can be reduced from 3.6 to 1.9, and the maximum amplitude of the SOC curve can be compressed to 7%. The algorithm also embeds an "uplink punctuality" strategy for Cat-1 links: when the network is normal, a 14-dimensional feature vector is reported every 10 seconds. If no ACK is received within 15 seconds, the frequency is immediately reduced to 100 seconds, and the LoRa-PHY is simultaneously awakened to form a neighboring box self-organizing network, and the latest 120 seconds of key summaries are broadcast in frames, ensuring that the EU GDP regulatory requirement of traceability within 10 seconds can still be met in base station blind spots.The entire prediction-feedback-scheduling logic completes a closed-loop operation in 14 ms on a 120 MHz MCU, significantly shorter than the 20 ms sampling period. Public road test results show that during a 90-second peak shock, when the outer wall temperature rose from 34°C to 61°C, the box core overshoot was limited to 0.42°C, 78% lower than simple PID control. The overall energy recovery efficiency was increased to 10.6%, extending the independent battery life from 118 hours to 168 hours, fully meeting the WHO PQS draft targets of "energy consumption ≤ 1.2 Wh·h⁻¹ per cubic decimeter, fluctuation ≤ ±0.5°C."

[0006] To ensure the reliability and compliance of the entire cold chain container during the transportation of high-value pharmaceuticals, this invention integrates safety protection and data integrity mechanisms beyond the closed loop of the thermal and energy chains, and provides a full lifecycle operation and maintenance solution centered around "manufacturing-calibration-operation-maintenance." The 4.7-inch flexible electronic ink display on the inside of the container lid maintains the most recent five-dimensional status (core temperature, core humidity, bus voltage, predicted remaining battery life, and risk level) for at least 30 days after the MCU loses power, thanks to microcapsule electrophoresis. Sixteen dual-color status LEDs surround the display, emitting a silent green light under normal operating conditions. Temperature, humidity, drop, and voltage anomalies trigger flashing patterns of red, yellow, blue, and purple, respectively. A 78 dBC buzzer provides 360-degree audible and visual alerts. The cabinet's attitude is monitored by a dual MEMS gyro-accelerometer system. If a tilt angle of 35° or greater is detected for two seconds, the system locks the lid solenoid valve, automatically reduces the TEC load by 30%, and broadcasts a "Class I drop event" message on the Cat-1 channel. If the tilt persists for more than 40 seconds or the Z-axis acceleration exceeds 4 g, the control core immediately activates the PCMS phase change residual heat channel to completely shut off the TEC power supply, preventing localized explosive heating of the coolant in the event of structural damage. The cabinet's exterior is inlaid with a highly flame-retardant EVA cushioning tire and features 16 kg·cm self-resetting hinges. After a drop test (1.2 m onto 50 mm thick steel plate), the cabinet maintains a core temperature gradient of no more than 0.9°C for one hour, meeting the ISO 21973-2020 limits for temperature excursion after impact for Class IV packaging. On the communication side, Cat-1 dual-antenna diversity serves as the primary link. If the base station blind spot persists for more than 15 seconds, LoRa-PHY switches to a 620 bps self-organizing network, forming a three-hop tree-like forwarding structure between adjacent boxes. This compresses core message bandwidth to 45 bytes over distances up to 1.2 km, maintaining the "accessibility within 10 seconds" requirement stipulated by the EU GDP. In the more extreme scenario of complete link loss, the system broadcasts summary information for a 120-second period using 433 MHz FSK, with each frame containing 34 bytes. This information can be read by a portable handheld base station within a radius of 400 meters. All uplink data is hashed in two layers: the first layer uses Blake3 to solidify a 256-bit hash of every three-second detection frame, while the second layer submits a one-minute summary to the Fabric consortium chain main channel. High-frequency power data from the TEC-hot-end-TEG is written to a lightweight DAG sidechain and synchronized with the mainchain time series using a verifiable delay function (VDF, τ = 256 steps), forming a legally tamper-proof, fully traceable ledger. Based on the actual logs of 3,300 boxes shipped from Q4 2024 to Q2 2025, the on-chain-off-chain hash comparison error rate is less than 2.4 × 10⁻.7 The manufacturing and calibration phase utilizes full module pre-calibration: Four-stage TEC-TEG components undergo two-point (-20°C / -60°C) ΔT-IV calibration at room temperature (23°C), and the PID-GRU coefficients are solidified after 120 minutes of self-learning. The temperature and humidity array undergoes a three-point calibration in a quartz constant-temperature water bath (-70°C / -20°C / 0°C), with an RMS residual of 0.08°C. Each batch of cabinets undergoes a six-hour step-and-hold test in a 35°C environmental chamber before shipment, with a steady-state mean square error (RMS) of no more than 0.3°C. The operational maintenance strategy is based on a cloud-based digital twin model: Upon initial power-up, the cabinet is paired with the cloud to generate a 1:1 3D twin. Real-time traffic is rendered using Web-GL, and abnormal quadrants are visualized using color blocks and vectors. If the GRU model detects a deviation of more than 8% in the average energy consumption-temperature difference curve over a seven-day period, an OTA calibration command is issued, prompting on-site replacement of the silicone grease or inspection of the cabinet wall integrity. The battery cells utilize 2p-2s hot-swappable technology, issuing an early warning and locking the 0.8°C discharge after 2000 cycles. This allows for hot swaps within 90 seconds without interrupting the temperature range. These safety and data integrity features ensure that the system is drop-resistant, thermally shock-resistant, self-healing, and forensically traceable throughout its lifecycle. Furthermore, it maintains long-term thermal management stability through a digital twin-cloud-based operations and maintenance framework. Combining the structural / thermal chain innovations of the previous two components with a self-learning energy chain algorithm, this invention achieves highly uniform temperature maintenance (core-to-wall ΔT ≤ 1.5°C, fluctuation ≤ ±0.5°C) for pharmaceutical preparations at -80°C for over 168 hours, while maintaining a specific energy consumption of 0.92 Wh·h⁻¹·dm⁻³. This significantly outperforms current international standards and competing products, meeting the stringent cold chain equipment requirements for future high-end biologics, personalized cell therapies, and remote clinical trials. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0008] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0009] Attachment Figure 1 It is the overall system flow chart.

[0010] Attachment Figure 2 It is an exploded CAD drawing of an industrial structure.

[0011] Attachment Figure 3 It is the flow chart of the prediction-feedback-scheduling algorithm.

[0012] Attachment Figure 4 It is a state machine diagram for security protection and data integrity. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0014] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationships and movement of various components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indicators will also change accordingly.

[0015] In addition, the descriptions of "first", "second", etc. in the present invention are for descriptive purposes only and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of such features. The technical solutions between the various embodiments can be combined with each other, but they must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0016] In today's biologics distribution system, the temperature control accuracy, energy efficiency, and full traceability of ultra-low temperature (≤ -70°C) cold chain equipment are key indicators of transport safety. Traditional dry freezers, liquid nitrogen tanks, or single-stage semiconductor boxes struggle to simultaneously meet the multiple requirements of sub-Celsius temperature uniformity, battery life exceeding 160 hours, and legal-grade data integrity, hindering the global distribution of high-value-added pharmaceuticals such as mRNA vaccines and cell therapy products. To address this, this paper proposes a pharmaceutical ultra-low temperature cold chain intelligent box system featuring "honeycomb phase change partitioning + four-level TEC / TEG coupling + self-learning GRU-PID temperature control + multi-source cascade energy management + dual-chain traceability." Through integrated innovations in architecture, algorithm, and operation and maintenance, this system achieves key performance features such as ±0.5°C fluctuation at a sustained -78°C temperature environment, 0.92 Wh·h-1·dm-3 per unit energy consumption, and self-recovery within 10 seconds of disconnection, providing comprehensive temperature control and regulatory compliance assurance for future high-end pharmaceuticals. Example 1

[0017] Based on Example 1, the following further details the interactions between each functional layer during a complete operational cycle, along with the calibration process and key performance data, are provided to enable those skilled in the art to directly replicate the smart box described herein. After the box is loaded with explosives and powered on in a room temperature of 22°C, the acquisition layer S101 initiates self-calibration of the 24-bit Σ-Δ ADC during the first 20 ms interrupt cycle, generating a quantization baseline with 0.8 µV resolution from an internal 2.5 V reference source. The twelve PT100 sensors are compensated using Callendar-Van Dusen polynomials, with a zero-degree resistance of 100.003 Ω. After this initial compensation, the full-scale error is reduced to ±0.05°C. To prevent self-heating during startup, the current sampling adopts an intermittent mode of "1 ms power-on, 9 ms power-off." The six-channel CMOS humidity sensor completes on-chip temperature compensation within 200 ms and transmits back a CRC-16 checksum. The battery pack's initial state of charge was approximately 98.6%, and the supercapacitor bank voltage was 2.2 V. At this point, the ΔV-SOC-ΔT router determined the load was light and operated the TEC at an idle cooling state with a 34% duty cycle and I≈1.9 A. The capacitor bank was precharged with a current of 0.18 C. This preparatory phase lasted 2.7 minutes. When the box was pushed into the −35°C test chamber, the outer wall temperature sensor dropped to 0°C within 11 seconds. The Kalman prediction noise covariance R_k automatically increased from 1.3×10⁻ to 0°C based on the temperature drift. 5 Increased to 1.1×10⁻ 4The filtering and preprocessing layer S102 quickly relaxed its response speed. The incremental GRU inference output a curve with a heat load ramp rate of 8.6 W·min⁻¹ and a predicted peak of 42 W in 30 minutes at 18 seconds. The target trajectory generator S104 lowered the cold-end setpoint from −72°C to −74°C and generated a linear ramp of −0.18°C·min⁻¹ for the next 15 minutes. The adaptive PID dynamically increased Kp from 2.8 to 4.6 within two 20-ms cycles and shortened the integral limiter Ti to 28 seconds. As a result, the TEC duty cycle rapidly increased to 84% between 40 and 52 seconds. Simultaneously, the energy router S107 detected rising bus ripple and added a supercapacitor in parallel to limit the peak pulse current to 6.2 A and the bus voltage drop to no more than 0.32 V. After 9 minutes of ramp cooling, the temperature difference at six points in the drug compartment fell below 1.4°C, and the core temperature error converged to −0.46°C. Thereafter, intermittent TEC pulses maintained a temperature difference drift of 0.18°C. During transportation, the test vehicle entered a long tunnel at an altitude of 2,138 meters. The base station lost signal for 46 seconds. After not receiving a Cat-1 ACK in the 15th second, the communication gateway S108 switched to LoRa-PHY and, utilizing the neighboring box's ad hoc network, sent a "low pressure altitude event" summary frame via a three-level relay. Upon exiting the tunnel, it automatically reconnected to the base station and added 184 cached second-level temperature and humidity data points to the DAG sidechain. The air pressure in the tunnel dropped by approximately 4.8 kPa, causing trace outgassing from the phase change plate inside the compartment to cause the humidity to rise to 58% RH. Although the threshold had not yet been exceeded, the predictor S103 mapped this trend into a 5-minute warning vector, increasing the hot-end cooling wind speed by 16% in advance. The results showed that the dew point inside the box was always 7°C lower than the temperature of the cold end metal plate, and no condensation occurred. At the end of the entire 168-hour independent endurance, the battery pack had a remaining SOC of 16.7%, the supercapacitor was discharged, and the TEG recovered a total of 76 Wh of energy, accounting for 8.9% of the total consumption; 10,099 blocks were written to the Cat-1 main chain, and 6.04×10 5 High-frequency records were verified by a double-layer cloud-based hash check, with no mismatches. The steady-state temperature RMS deviation was 0.26°C, with a peak of 0.48°C, exceeding the ±0.5°C upper limit of the WHO PQS draft. The specific energy consumption was 0.92 Wh·h⁻¹·dm⁻³, representing a 37.8% energy saving compared to a single-stage TEC control chamber (1.48 Wh·h⁻¹·dm⁻³). This demonstrates that the process and functional layers illustrated in Figure 1 can achieve the proposed goals of high-uniform temperature control, long battery life, and full traceability in extreme operating environments. Example 2

[0018] Example 2 describes the physical structure, material selection, and assembly process of the pharmaceutical ultra-low temperature cold chain smart box. The overall frame is first cut into a U-shaped base and four columns from 7075-T6 aviation aluminum-magnesium alloy extrusion profiles on a five-axis CNC machine tool. It is then subjected to stir friction welding to form an integrated fillet weld under process parameters of 180 rpm and an axial force of 14 kN. Metallographic examination of the weld shows that the grain size has been refined to 4.8 µm, ensuring the structural strength for subsequent extreme drop tests. The outer panel is made of 0.7 mm thick heat-cured carbon fiber reinforced epoxy resin sheet. After PVD titanium nitride coating to reduce solar absorption, it is fixed to the aluminum frame using a mixture of −35°C cold-resistant epoxy glue and ball-blind rivets, achieving a strength better than 2.8 × 10⁻. 4The thermal insulation system consists of three layers of SiO2 aerogel core, a 0.05 mm aluminized film, and a 0.5 mm flexible graphite heat diffuser laminated from the outside inward. Cured in a vacuum hot press at 0.1 MPa and 85°C, the in-plane thermal conductivity is only 0.012 W·m⁻¹·K⁻¹. After the insulation panel is slid into the inner slot of the shell, it is filled with 5 gm⁻¹ polytetrafluoroethylene micropowder under a vacuum of 10⁻³ Pa, providing both vibration damping and acoustic absorption. A honeycomb phase change composite panel is mounted on the inner surface of the aerogel via dovetail grooves: polydecanoate triol-hexadecane liquid PCMS is first injected into the 0.3 mm thick aluminum honeycomb, then sealed using 60 kHz ultrasonic roll welding. The melting point is −63°C and the latent heat of fusion is 210 kJ·kg⁻¹. 0.8 mm EPDM flat seals are inserted around the honeycomb panels to form a replaceable heat sink module, designed for at least 500 assembly and disassembly cycles. The internal three-stage cavity is secured by two sets of X-shaped glass-fiber-reinforced PPS truss baffles. The central drug compartment is lined with 1 mm aluminum nitride ceramic plates. Their high thermal conductivity of 180 W·m⁻¹·K⁻¹ ensures rapid and even distribution of the cooling energy extracted from the TEC cold end. The annular buffer compartment is pre-embedded with 3 mm PCMS sheet slots for sudden load absorption. The outer ring isolation compartment serves as the hot end air duct. The four-stage TEC-TEG module is embedded in a milled stepped groove in the long side wall. The cold end interfaces with a microchannel liquid cold plate, then contacts the AlN inner plate via a 20 µm thermal grease containing a silane coupling agent. The hot end couples heat flow to the outer ring air duct via a 500 µm graphite sheet-aluminum alloy heat sink. A 220 µm Cu / Ni bimetallic film TEG is sandwiched between these two elements, delivering 2.8 W of DC power at a 45°C temperature differential with the hot end. The liquid cooling circuit utilizes an embedded BLDC micropump and a 47% ethylene glycol-water refrigerant, with a flow rate of 0.95 L / min⁻¹ at a static pressure of 10 kPa. The bottom of the box features an integrated guide rail bracket, into which four 92 Wh lithium iron phosphate batteries and a 120 F supercapacitor slide and lock via spring-loaded connectors. The bracket is milled with 1.5 mm thick copper-based heat sink fins, supplemented by low-resistance silicone pads to ensure a cell temperature gradient of less than 3°C. All PT100 sensors and digital humidity chips are routed between the housing and the insulation layer using laser-engraved routing channels. These are then wrapped with shielded copper foil and graphite sheet composite tape. After routing, the wiring is secured with room-temperature vulcanized silicone adhesive, with a bend radius of at least 5 mm to prevent breakage during transportation vibration. The top cover is a 4.7-inch flexible electronic ink display composite. Rectangular silicone sealing rings are embedded on all four sides and locked into a stainless steel strip with a torque of 0.9 N·m. A two-point Hall effect lock and electromagnetic lock work together to provide IP67 protection. Once sealed, the entire box is tested in a 15 m³ helium inspection chamber with a 1 × 10⁻ 6The Pa·m³·s⁻¹ leak rate is determined, and passing is marked as qualified. After a 1.2 m drop onto a 50 mm steel plate impact test and a six-cycle thermal shock test from −70°C to +60°C, the maximum ΔT between the core and the wall remained within 1.5°C, with no cracks observed in the structural components. This meets all the drop and thermal shock requirements of ISO 21973-2020 Class IV packaging. The entire container has an empty mass of 18.6 kg and a net volume of 18 L, making it suitable for single-person transport. This demonstrates that the assembly structure shown in Figure 2 not only achieves extreme strength and ultra-low thermal conductivity, but also provides a reliable physical support for the subsequent implementation of the thermal chain-energy chain closed loop and safety-traceability functions described in Examples 3 and 4. Example 3

[0019] Example 3, centered around the prediction-feedback-scheduling algorithm shown in Figure 3, describes the real-time operational mechanism of temperature control and energy management in a smart box for urban distribution. The box is loaded with two pallets of 120 mL of cell therapy stock solution, which, according to pharmacopoeia requirements, must be transported across cities for 14 hours at −78°C ± 0.5°C. After packing, the system enters the 20 ms sampling interrupt phase S301, where the MCU synchronously acquires temperature, humidity, voltage, current, and IMU data. During the first cycle, the initial value of the Kalman prediction noise covariance, R0, is set to 3.2 × 10⁻. 5 , corresponding to the steady-state noise at room temperature of 22 ℃. The second interrupt cycle filter-preprocessing layer S302 has automatically set R k Increase to 1.1 × 10⁻ 4The processed stream data is then pushed into a 128-second sliding window (S303). When the delivery truck exits the GMP workshop and enters the 32°C loading and unloading dock, the window data reaches full amplitude at 300 seconds. The incremental GRU inference module (S304) begins to infer a 30-minute heat load trajectory every 20 milliseconds. At this point, the predicted curve slope reaches 7.9 W·min⁻¹. The model's output rapid warning vector indicates a "cold-end temperature overshoot risk > 0.8°C within 5 minutes." Consequently, the target trajectory generator (S305) pre-sets the original -78°C setpoint to -79°C and adds a temporary trapezoidal slope of -0.2°C·min⁻¹, forming a system feedforward. The instantaneous error between the new target temperature curve and the real-time sampled temperature was fed into the self-tuning PID controller S307. When the sum of squares of the online residual error, δ(t), reached 0.042, the PID controller automatically increased the proportional coefficient Kp from 3.2 to 4.1 and shortened the integral limit Ti to 22 seconds. Subsequently, the PWM drive duty cycle increased sharply from 38% to 81% within 36 seconds. The four-stage TEC-TEG actuator module S308 began operating at a 5.6 A rectifier current. The cold-junction temperature dropped to −79.2°C within 180 seconds, with the overshoot reduced to less than 0.46°C. The sudden surge in power at the warm end caused a bus current spike. The energy routing logic S309 immediately consulted the ΔV-SOC-ΔT table and connected a 120 F supercapacitor in parallel. Simultaneously, the hot-end-cold-end temperature difference ΔT, detected to have exceeded 42°C, prompted the router to boost the 2.7 W output of the TE-G thin-film capacitor to the grid, reducing the load on the main battery. The bus voltage regulation loop S310 maintained the voltage ripple at 0.3 V peak-to-peak for 15 seconds, limiting the battery's instantaneous discharge rate to 0.78°C. After the delivery truck entered an elevated tunnel, it lost contact with the base station for 27 seconds, triggering a communication layer degradation. The Cat-1 module logged three "NO ACK" signals and forwarded them to a neighboring LoRaPHY relay box. Due to the tunnel's dual-hole structure, the LoRa packet was captured by the exit base station after 800 meters and transmitted back to the cloud, ensuring temperature data traceability within the regulatory 10-second limit. The air pressure in the tunnel dropped by 4.1 kPa, causing a slight increase in the humidity inside the box. The incremental learning weights of the GRU model adapted in real time, and the prediction curve was corrected to a gentler slope. The PID tuning then fell back to the steady-state coefficient group, and the PWM duty cycle dropped to 46%.At the end of the 14-hour process, the battery pack had a remaining SOC of 68.4%, with the TEG recovering 24 Wh and the supercapacitors utilizing 31 Wh of peak kinetic energy. The RMS core temperature deviation was 0.24°C, with a maximum instantaneous peak of 0.48°C, both below the clinical internal control limit of ±0.5°C. The unit energy consumption was converted to 0.95 Wh·h⁻¹·dm⁻³, a 34% reduction compared to the control of a traditional single-stage TEC enclosure (1.44 Wh·h⁻¹·dm⁻³). This demonstrates that the prediction-feedback-scheduling algorithm enables the system to maintain sub-Celsius temperature control and high energy efficiency under dynamic thermal load and link jitter conditions, fully confirming the effectiveness of the S301-S310 process shown in Figure 3. Example 4

[0020] Example 4, based on the security protection and data integrity state machine shown in Figure 4, illustrates how the smart box implements multi-level responses to attitude shocks, temperature and humidity drift, electrical anomalies, and link interruptions in a combined scenario of long-distance air transport and road connection, while simultaneously completing dual-chain judicial evidence storage. The box is handed over by a loading supervisor at the −18°C apron loading point and then placed in the belly cargo hold. Upon startup, the sensor array S401 establishes a zero baseline, with attitude readings of 0 g in all three axes, −77.8°C temperature, 34% RH, and 14.6 V bus voltage. During the flight's climb phase, the outer wall pressure drops sharply with altitude. At 4500 meters, air turbulence causes the box to drop 18 cm vertically with a 22° tilt angle, resulting in an IMU peak reading of 2.7 g. Threshold determination module S402 confirmed within 30 ms that the tilt angle was >15° and lasted for >1 second, triggering a Level 1 alert. The LED bar flashed red rapidly, a buzzer sounded every 5 seconds, and an event frame was sent to the cloud via Cat-1. Due to wireless attenuation in the cargo hold, the RSSI of the signal was only −107 dBm, extending the uplink latency to 5 seconds, but still within the regulatory limit of 10 seconds. Fifteen minutes later, as the aircraft entered the stratosphere, the cargo hold temperature dropped from 12°C to −30°C. An external infrared radiometer detected a reverse heat flow, and the GRU predictor inferred a significant decrease in thermal load in the background. The adaptive PID controller reduced the TEC load to 24% duty cycle. However, the bus voltage, after automatic balancing, had dropped to 13.2 V. A further drop below 12.8 V would trigger load shedding. The system, through energy routing, injected 2.1 W of TEG power into the bus to recover and maintain the voltage level. During the second hour of flight, the cargo box experienced another strong shock due to frontal airflow: the IMU registered a peak of 4.4 g on the Z axis, with a momentary tilt of 38° that persisted for three seconds. The threshold determination (S402) initiated the second-level emergency path. Within 200 milliseconds, the second-level emergency control (S404) locked the lid solenoid valve, executed a 30% load reduction command for the PWM, and simultaneously opened the PCMS temperature trap link for 35 seconds to allow for latent heat to maintain stability while the TEC was paused. Because the drop impact exceeded the third-level protection threshold but was below the damage threshold, the system did not disconnect the TEC, maintaining the second-level state until the attitude returned to normal. After landing, the flight switched to a highway relay.When a container was loaded onto a truck and entered a mountain tunnel, the Cat-1 link experienced three consecutive NOACKs within 17 seconds. Link integrity check (S406) determined the primary link had failed, automatically switching to LoRa-PHY and broadcasting node information. Metal diffraction from the tunnel walls also caused LoRa packet loss for 11 seconds, prompting the system to enter level 3 degradation (S407). Activating 433MHz FSK broadcast, the system transmitted a 34-byte emergency summary frame every 5 seconds. A portable base station carried by the convoy's lead vehicle successfully received the signal at a distance of 280 meters and reconnected to the public network. A sudden rockfall at the tunnel exit caused the truck to brake suddenly, causing the container to slide 0.9 meters, tilting 44 degrees, with a delay of 6 seconds. The threshold immediately escalated to level 3 safety protection (S405). The self-locking solenoid valve remained closed, the system forcibly disconnected the TEC power supply, the anti-condensation relay disconnected the refrigerant and liquid cooling circuits, the PCMS took over full temperature control, and the buzzer switched to a continuous alarm. Since the Cat-1 link had been restored, the Level 3 protection event frame was encrypted and sent to the cloud within 10 seconds, synchronously written to the Fabric main chain (S408). To avoid write latency, the second-level temperature and humidity data was directly transmitted to the DAG side chain (S409). The two-layer hash was then calculated and cross-written using Blake3, achieving dual-layer verifiable delay function curing (S410). Subsequent inspections showed that the core temperature remained at −77.4°C within the 6-minute protection window for TEC disconnection, with a maximum drift of 0.46°C, fully meeting pharmacopoeia requirements. After the event was closed, maintenance personnel replaced the collapsed buffer tire and loose graphite sheet according to the cloud-based operation and maintenance instructions. The system released the Level 3 lock and resumed TEC operation. A total of 13,821 main chain blocks and 1.09 × 10-second side chain records were generated throughout the process. 6 The hash comparison error between the chain and the chain is < 2.3 × 10⁻ 7 All sound and light warning responses are < 0.3 s. After one cycle of the four-level state machine 0→3→0, the hardware is intact and the temperature zone is not out of control. This fully demonstrates that the S401–S410 state machine shown in Figure 4 can achieve security protection and data integrity goals under multiple shocks and link failure environments.

[0021] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention. Technical terms that need to be explained to help understand the present invention

[0022] To facilitate reading and accurate understanding of this disclosure, the following definitions are provided for technical terms that frequently appear in this specification or have specific meanings. If there are any discrepancies with commonly used industry definitions, the following interpretations shall prevail. Terms not listed hereby shall be understood according to their common meanings in the art.

[0023] (1) Thermoelectric Cooler (TEC): A solid-state cooling device composed of a PN pair of semiconductor materials such as Bi2Te3, which uses the Peltier effect to absorb heat at one end and release heat at the other end under direct current drive. The control method often uses pulse width modulation (PWM) to adjust the duty cycle to adjust the cooling power.

[0024] (2) Thermoelectric Generator (TEG): This is similar to the TEC material system, but operates in a thermoelectric power generation mode: the temperature difference between the hot and cold ends generates an electromotive force that outputs DC power for energy recovery or busbar recharging. In this invention, a TEG film is sandwiched between the TEC hot end and a heat sink to form a "cooling-power generation" coupling.

[0025] (3) Honeycomb Phase-Change Material Structure (PCMS): A plate-shaped energy storage module made by integrally infusing an aluminum honeycomb substrate and a eutectic-long-chain alkane phase-change filler. It can absorb or release a large amount of latent heat during the melting / solidification process, which is used to reduce peaks and fill valleys and slow down the temperature drift of the cabinet.

[0026] (4) SiO2 aerogel insulation layer: Nanoporous silica material produced by supercritical drying process, with a thermal conductivity as low as 0.012 W·m⁻¹·K⁻¹ at room temperature, is one of the lightest and most efficient commercial insulation media currently available.

[0027] (5) Four-stage TEC / TEG coupled module: This refers to a multi-stage thermoelectric component in which four stacked TECs are connected in series and parallel, and three layers of Cu / Ni thin film TEGs are used for hot-end power generation. The stacked structure achieves large temperature difference and low freezing point cooling, and the TEGs are responsible for converting the useful temperature difference at the hot end into DC energy recovery.

[0028] (6) ΔV-SOC-ΔT energy router: A multivariate function decision table with bus voltage drop ΔV, battery state of charge SOC and hot-end-cold-end temperature difference ΔT as independent variables. Through the MOS array, seamless switching of the four sources of battery-TEG-on-vehicle DC-supercapacitor is achieved to ensure the balance between instantaneous peak load and endurance.

[0029] (7) Incremental GRU predictor: A pruned Gated Recurrent Unit (GRU) network running on an embedded MCU; incremental forward inference is performed only on the latest sliding window data, outputting the next 30-minute heat load curve and a 2.5-minute rapid warning vector.

[0030] (8) Self-tuning PID controller: The proportional-integral-derivative (PID) algorithm adds dynamic tuning rules such as the sum of squares of residual errors and adaptive gradients to adjust K in real time. p , Tᵢ, T_d, so as to minimize the overshoot and steady-state error of the core temperature.

[0031] (9) Cat-1 cellular IoT communication: The LTE Cat-1 narrowband IoT standard proposed by 3GPP Release 13 combines a peak rate of 10 Mbit / s⁻¹ with the low-power sleep characteristics of eDRX / PSM, and is suitable for high-frequency, small-packet, and low-latency reporting of mobile cold chain boxes.

[0032] (10) LoRa-PHY self-organizing network: A 620 bps low-speed wireless link based on the Semtech LoRa spread spectrum physical layer, used as a neighboring box cascade relay when the main link is disconnected.

[0033] (11) 433 MHz FSK broadcast: A low-frequency broadcast channel modulated by frequency shift keying (FSK). It is enabled only when both Cat-1 and LoRa fail, ensuring that the emergency summary frame can be captured by a handheld base station within a radius of 400 m within 10 s.

[0034] (12) Fabric Main Chain / DAG Side Chain: Fabric is a consortium chain framework responsible for writing minute-level blocks; the directed acyclic graph (DAG) side chain records high-frequency flows at the second level. The dual-chain architecture provides both judicial-level immutability and write throughput.

[0035] (13) Blake3 hashing & Verifiable Delay Function (VDF): Blake3 is a high-throughput 256-bit hashing algorithm; VDF outputs a time lock proof within τ = 256 steps, ensuring that the on-chain time series cannot be reversed or forged in parallel.

[0036] (14) Kalman filter (IIR-Kalman cascade): IIR first suppresses high-frequency noise, and Kalman filter dynamically iterates the optimal estimate in each sampling period through the state space model. The cascade of the two can take into account both speed and robustness.

[0037] (15) Σ-Δ ADC: Σ-Δ type high-precision analog-to-digital converter, which uses oversampling and noise shaping technology to achieve 24-bit quantization accuracy and is suitable for microvolt-level thermoelectric signal acquisition.

[0038] (16) Friction stir welding: A solid phase welding process for low melting point alloy materials. The metal in the welding zone undergoes controlled plastic flow and there is no molten pool. The weld has refined grains and low residual stress.

[0039] (17) PVD titanium nitride coating: Physical vapor deposition (PVD) forms a TiN coating about 400 nm thick on the surface of the outer panel to reduce the solar absorption rate α s To 0.21, improve the scratch resistance and corrosion resistance of the shell.

[0040] (18) E-Ink flexible display: The electrophoretic electronic ink screen can maintain the image for several months when the power is off, and is used to display the box core temperature, residual energy and risk level in a passive environment.

[0041] (19) WHO PQS: World Health Organization Performance, Quality and Safety Specification, which proposes standards such as ±0.5 °C fluctuation and 10 s traceability for pharmaceutical cold chain equipment.

[0042] (20) ISO 21973-2020: "Biopharmaceutical Products - Cell Therapy - Transport Requirements" issued by the International Organization for Standardization, which stipulates the limits for drop, thermal shock and temperature retention, and is the basis for the drop resistance and temperature drift test of the box of the present invention.

[0043] Other abbreviations such as SOC (State of Charge), RMS (root mean square), PWM (pulse width modulation), and BLDC (brushless DC motor) all use their commonly used meanings in the industry.

Claims

1. A pharmaceutical ultra-low temperature cold chain intelligent box system with adaptive multi-source thermal management, characterized by: include: A) A multi-layer thermal insulation and cold storage assembly, consisting of an outer shell, a SiO2 aerogel insulation layer, a honeycomb phase change composite material plate, and an aluminum nitride inner wall. The melting point of the phase change material is −63°C. B) a four-stage TEC / TEG coupled refrigeration-energy recovery assembly, wherein the cold end is thermally conductively connected to the inner wall, the hot end is connected to the microchannel liquid cooling heat sink via a graphite interface material, and a thin film TEG sheet is sandwiched between the hot and cold ends; C) The energy subsystem consists of a lithium iron phosphate battery pack, a supercapacitor pack, a thermoelectric DC output, and an external 12 V DC input. These four sources are switched via a ΔV-SOC-ΔT energy router. D) Data acquisition and control subsystem, including 12 PT100 and 6 digital humidity sensors, IMU and 24-bit Σ-Δ ADC, with an incremental GRU predictor and self-tuning PID controller running in the MCU; E) Communication and traceability subsystem, including Cat-1, LoRa, and 433 MHz FSK modules. The collected data is written to the Fabric main chain and DAG side chain in a dual layer and solidified using Blake3-256 hashing and VDF; F) Safety state machine, with posture-drop detection, temperature and humidity threshold comparison, electromagnetic lock, and three-level sound and light-load reduction-heat sink protection logic; The MCU collects data at a 20 ms cycle, uses a GRU to predict the 30-minute heat load, and drives a self-tuning PID controller to adjust the four-stage TEC. Duty cycle control maintains the core temperature at −80°C ± 0.5°C, with a specific energy consumption of ≤ 0.95Wh·h⁻¹·dm⁻³.

2. The system according to claim 1, wherein: The honeycomb phase change composite material plate is made of 0.3 mm thick aluminum honeycomb and polydecanoic acid triol-hexadecane eutectic filler by integral infusion molding, with a latent heat of ≥ 210 kJ·kg⁻¹.

3. The system according to claim 1, wherein: The ΔV-SOC-ΔT energy router automatically switches between batteries, TEGs, on-board DC, and supercapacitors within 5 ms based on the combined state of the bus voltage drop ΔV, state of charge (SOC), and hot-end-cold-end temperature difference ΔT.

4. The system according to claim 1, wherein: The number of parameters of the incremental GRU predictor is ≤ 24 k, and the single inference time is ≤ 4 ms. The proportional, integral, and differential coefficients of the self-tuning PID are updated online based on the residual error sum of squares and the temperature difference slope.

5. The system according to claim 1, wherein: The microchannel liquid cooling heat sink uses a 47% ethylene glycol-water mixture as the refrigerant, with a flow rate of 0.9 L·min⁻¹ and a static pressure of 10 kPa.

6. The system according to claim 1, wherein: The communication-tracing subsystem automatically downgrades to LoRa-PHY if the Cat-1 link is unresponsive for 15 seconds. If LoRa also fails, 433 MHz FSK broadcast is enabled, and the emergency summary frame can be received by a portable base station within a 400 m radius within 10 seconds.

7. The system according to claim 1, wherein: The dual-chain write frequency is: second-level temperature and humidity streams are written to the DAG side chain, and minute-level summaries are written to the Fabric main chain. The dual-chain write delay is ≤ 7 seconds.

8. The system according to claim 1, wherein: The three-level protection of the safety state machine includes: the first level is LED-buzzer warning; the second level is electromagnetic lock closing and reducing the TEC load to 70% of the rated value; the third level is cutting off the TEC and the PCMS heat sink takes over the temperature control.

9. A temperature control and energy coordination method based on the smart box according to claim 1, characterized in that: a) Multi-source data is collected at a 20 ms period and filtered using IIR-Kalman filtering; b) Input the 128-second sliding window data into the GRU to obtain the heat load curve for the next 30 minutes and generate the target temperature trajectory; c) The self-tuning PID drives the four-stage TEC based on the error, and the ΔV-SOC-ΔT router synchronously dispatches the power supply; d) Temperature, humidity and event frames are uploaded via Cat-1. When the connection is lost, they are downgraded to LoRa and FSK broadcasts in turn and written to the dual chain at the same time; e) When the IMU detects that the Z-axis acceleration is > 4 g and the tilt angle is > 40° for 2 s, the third level protection is triggered, the TEC is cut off and the box cover is locked.

10. Use of the system or method according to any one of claims 1 to 9 in the cold chain field of mRNA vaccines, cell or gene therapy products, blood or other biological samples that need to be transported at ≤ −70°C.

Images