Photovoltaic lithium battery air conditioner terminal machine room ai tuning method based on digital twinning

Abstract

The present application relates to photovoltaic lithium electricity energy storage and air conditioning system intelligent control technical field, specifically, the present application relates to photovoltaic lithium electricity air conditioner terminal machine room AI optimization method based on digital twinning, the present application is through the deployment of distributed optical fiber temperature sensor array, combined with k-epsilon turbulence model generates three-dimensional temperature field positioning hotspot area and calculates the maximum refrigerating capacity increment of air conditioner, fusion this increment, photovoltaic forecast power and lithium charge state, superimposed inverter efficiency compensation factor gets air conditioner callable green electricity quota, bind hot spot and quota generates thermal-electric coupling factor, combined with photovoltaic fluctuation intensity activates linkage mode, triggers deep deterministic policy gradient model to execute air valve opening, water temperature setting, lithium charge and discharge collaborative decision, through joint reward function binding evaluation three control quantity, output joint control vector, the method improves green power consumption rate and temperature control precision, guarantees equipment safety, applicable to machine room efficient operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of photovoltaic lithium battery energy storage and air conditioning systems, in particular to an AI optimization method for photovoltaic lithium battery air conditioner terminal machine rooms based on digital twinning. BACKGROUND

[0002] Intelligent control of photovoltaic lithium battery energy storage and air conditioning systems is an important technology. In the current low-carbon transformation and intelligent upgrading of machine rooms, this technology maps the temperature field of the machine room, the photovoltaic output and the lithium battery state in real time through digital twinning, deeply combines air conditioning refrigeration demand with green electricity supply, improves the consumption rate of photovoltaic green electricity, reduces dependence on city power to reduce energy consumption costs, accurately regulates the temperature of the machine room, avoids damage to equipment due to high temperature overload or low temperature condensation, and ensures the safe operation of the lithium battery energy storage system.

[0003] However, the existing air conditioner terminal machine room optimization technology has the core problems of insufficient heat and electricity cooperation and disconnection of multiple control links. This problem is caused by three practical limitations: first, the hot spot area is not accurately located based on the three-dimensional temperature field, and the green electricity quota allocation is only based on the photovoltaic output, which is disconnected from the local refrigeration demand of the machine room, resulting in insufficient green electricity supply in the hot spot area with high cooling demand and waste of green electricity in the low cooling demand area. Second, when the photovoltaic power fluctuates, the lithium battery charging and discharging, air conditioner damper opening and water temperature setting are independently decided, lacking linkage and constraints. Blind discharge of lithium battery during peak photovoltaic period, if the damper adjustment rate is too fast, it may exacerbate the compressor load fluctuation, and if the water temperature is not synchronously increased, it may cause air outlet condensation. Third, the evaluation of the optimization effect is not bound to multiple control variables, only a single link is judged, and the mutual influence between control variables is ignored, resulting in one-sided decision-making. These problems will have a chain effect. The refrigeration in the hot spot area is not timely, which will reduce the stability of equipment operation, the low consumption rate of green electricity will increase the consumption of city power and carbon emissions, and the risks of condensation and compressor impact caused by independent control will increase the operation and maintenance cost. Ultimately, it cannot meet the comprehensive needs of the machine room. An AI optimization scheme that can realize deep coupling of heat and electricity, collaborative decision-making of multiple control links and overall evaluation is urgently needed. In order to solve this technical problem, we provide an AI optimization method for photovoltaic lithium battery air conditioner terminal machine rooms based on digital twinning. SUMMARY

[0004] The purpose of the present application is to provide an AI optimization method for photovoltaic lithium battery air conditioner terminal machine rooms based on digital twinning to solve the problems raised in the background.

[0005] 1. Because the green electricity quota allocation is disconnected from the refrigeration demand of the machine room, resulting in insufficient cooling in the hot spot area, this case locates the hot spot through a three-dimensional temperature field, calculates the green electricity quota by integrating the refrigeration capacity increment, photovoltaic power and lithium battery state of charge, and binds the hot spot and the quota, which can accurately allocate green electricity and meet the refrigeration demand of the hot spot.

[0006] 2. Since multiple control links are prone to risk due to independent decision-making, the case activates the linkage mode, uses the deep deterministic policy gradient model to make collaborative decisions on the air valve, water temperature and lithium battery, and evaluates the joint reward function. It can avoid device risks and improve tuning reliability.

[0007] To achieve the above purpose, one of the purposes of the present application is to provide a photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning, comprising the following steps:

[0008] According to the three-dimensional temperature field, the hot spot area is located and the maximum refrigeration capacity increment of the air conditioning system is calculated, and the maximum refrigeration capacity increment, the photovoltaic predicted power and the lithium battery state of charge are fused to calculate the green electricity quota available to the air conditioner;

[0009] The hot spot area is real-time bound with the green electricity quota available to the air conditioner, a thermal-electric coupling factor is generated, and it is activated in linkage mode with the photovoltaic predicted power fluctuation intensity to trigger the deep deterministic policy gradient model to perform collaborative decision-making, including:

[0010] Air valve opening degree decision: through water temperature set value constraint, when improving the air valve opening degree of the hot spot area, the water temperature set value is required to be improved synchronously according to the air supply dew point risk, and the water temperature improvement amplitude and the air valve opening degree increment amplitude satisfy a preset proportional relationship;

[0011] Water temperature setting decision: through lithium battery state of charge constraint, when improving the water temperature to absorb green electricity, the lithium battery is in ready mode, including state of charge and temperature safety range, otherwise the water temperature improvement amplitude is limited;

[0012] Lithium battery charging and discharging decision: through air valve action constraint, when discharging operation is performed during the photovoltaic output peak period, the air valve adjustment rate is lower than the safety threshold to prevent compressor impact;

[0013] Finally, the air valve opening degree, water temperature setting and lithium battery charging and discharging decision are overall evaluated by the joint reward function, only when the three control quantities are submitted synchronously, the complete reward value is calculated, and the joint control vector is output, and the air conditioner terminal machine room AI tuning is performed.

[0014] Compared with the prior art, the present application has the following advantages:

[0015] 1. Construct a three-dimensional temperature field by a distributed optical fiber temperature sensor array and a k-ε turbulence model, accurately locate the hot spot area of the computer room and calculate the maximum refrigeration capacity increment of the air conditioner under the anti-condensation constraint, solve the problem that the traditional green electricity quota allocation is divorced from the refrigeration demand, and integrate the photovoltaic predicted power and the lithium charge state to calculate the callable green electricity quota. Through the space mapping matrix, the geometric coordinates, temperature difference amplitude, heat source power and quota weight of the hot spot area are associated, the high weight hot spot automatically obtains the priority allocation right of green electricity, and the dynamic preemption mechanism can also respond to new hot spot demand. This design not only avoids the waste of green electricity in low cooling demand areas, but also ensures that high cooling demand hot spots obtain sufficient green electricity support, improves the photovoltaic green electricity consumption rate, and quickly eliminates the high temperature hidden danger of the computer room, ensuring the stability of the equipment operating environment.

[0016] 2. In view of the disadvantages of traditional air valve, water temperature and lithium electricity independent decision, the adaptive linkage mode is activated through a double threshold triggering mechanism. When the photovoltaic fluctuation is small, the "air valve-water temperature linkage mode" is activated, the air valve opening degree and water temperature lifting ratio are dynamically adjusted according to the nonlinear mapping table, the supply air dew point temperature is monitored in real time to avoid condensation risk, and the "lithium electricity-water temperature buffer mode" is switched when the photovoltaic fluctuation is long-term over-standard. The air valve regulation rate is locked, and the photovoltaic output fluctuation is smoothed through three-stage lithium power compensation. At the same time, a closed-loop verification chain is established. The water temperature compensation request is triggered by the air valve opening degree increment, the lithium state verification is triggered by the water temperature lifting demand, and the air valve rate is constrained by the lithium action feedback. If any link is abnormal, the global strategy reorganization is started, effectively avoiding the risks of compressor impact, lithium over-discharge and air outlet condensation.

[0017] 3. A joint reward function is used to control the submission of control variables and multi-dimensional weighted evaluation. The air valve opening degree, water temperature setting and lithium charge and discharge instructions are required to be submitted synchronously and have consistent time stamps. Otherwise, only local rewards are calculated and errors are fed back to avoid one-sidedness caused by single instruction evaluation. During evaluation, the photovoltaic consumption rate, hot spot elimination rate, lithium loss index and condensation risk coefficient are integrated, the comprehensive score is calculated according to the weight, and the collaborative effectiveness level is divided. This mechanism not only ensures the integrity and objectivity of the evaluation of the optimization effect, but also continuously optimizes the decision-making strategy through level feedback, so that the collaborative decision-making of the air valve, water temperature and lithium always meets the operation goal of the computer room. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The overall workflow of the present application is shown in the figure. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0020] Referring to Figure 1 As shown in the embodiment, the photovoltaic lithium battery air conditioner terminal machine room AI optimization method based on digital twinning is provided, including the following steps:

[0021] According to the three-dimensional temperature field, the hot spot area is located and the maximum refrigerating capacity increment of the air conditioning system is calculated, and the maximum refrigerating capacity increment, the photovoltaic predicted power and the lithium battery state of charge are fused to calculate the green electricity quota that can be called by the air conditioner;

[0022] The hot spot area and the green electricity quota that can be called by the air conditioner are real-time bound, a thermal-electric coupling factor is generated, and the thermal-electric coupling factor is combined with the photovoltaic predicted power fluctuation intensity to activate a linkage mode, triggering a deep deterministic policy gradient model to execute a cooperative decision, including:

[0023] Air valve opening degree decision: through water temperature set value constraint, when the air valve opening degree in the hot spot area is improved, the water temperature set value is required to be improved synchronously according to the air supply dew point risk, and the water temperature improvement amplitude and the air valve opening degree increment amplitude satisfy a preset proportional relationship;

[0024] Water temperature setting decision: through the lithium battery state of charge constraint, when the water temperature is improved to consume green electricity, the lithium battery is in a ready mode, including the state of charge and the temperature safety range, otherwise the water temperature improvement amplitude is limited;

[0025] Lithium battery charging and discharging decision: through the air valve action constraint, when the discharging operation is performed in the photovoltaic output peak period, the air valve adjustment rate is lower than a safety threshold to prevent the compressor from being impacted;

[0026] Finally, the air valve opening degree, the water temperature setting and the lithium battery charging and discharging decision are overall evaluated through a joint reward function, only when the three control amounts are submitted synchronously, the complete reward value is calculated, and a joint control vector is output, for the air conditioner terminal machine room AI optimization.

[0027] In the photovoltaic lithium battery air conditioner terminal machine room AI optimization process based on digital twinning, according to the three-dimensional temperature field, the hot spot area is located and the maximum refrigerating capacity increment of the air conditioning system is calculated, which is a core basic link. This step not only provides a quantitative basis for the refrigeration demand for the subsequent green electricity quota calculation, but also clearly defines the priority target of the air conditioner optimization through accurate positioning of the hot spot area, avoids blind allocation of cold quantity and green electricity resources, lays a data foundation for the generation of the thermal-electric coupling factor and the execution of the cooperative decision, and the specific implementation is as follows:

[0028] A distributed fiber-optic temperature sensor array is deployed, which adopts a non-uniform density design. The core logic is to arrange sensors differently according to the thermal distribution characteristics of the machine room cabinet, to balance the monitoring accuracy and cost. The high heat density area refers to the area where the heat is concentrated due to the high load of equipment on the surface of the cabinet, usually corresponding to the upper part of the cabinet, the area near the power interface, and the area where the cooling fan fails frequently. This area has a fast heat accumulation speed and a high temperature peak, and is the focus of air conditioning cooling. Therefore, the sensor deployment density needs to be doubled to ensure that subtle temperature changes can be captured. The temperature data collected synchronously by the sensors will be transmitted to the digital twin platform through the optical fiber, providing raw data support for subsequent three-dimensional temperature field construction.

[0029] Subsequently, the synchronously collected temperature data is input into a fluid dynamics solver based on the k-ε turbulence model, which is a turbulence model commonly used to simulate indoor airflow and temperature distribution. Through two transport equations, it describes the turbulent kinetic energy (k, reflecting the intensity of turbulent motion) and the turbulent dissipation rate (ε, reflecting the dissipation speed of turbulent energy), which can accurately calculate the diffusion path of cold air and the accumulation law of hot air in the machine room. It is especially suitable for complex air flow scenarios such as forced air supply in air conditioning end machine rooms and equipment heat dissipation, avoiding temperature field distortion caused by traditional simplified models. The solver will first grid the machine room space, then use the temperature data collected by the sensors as boundary conditions, and through iterative calculation, it will obtain the real-time temperature value of each grid. Finally, a three-dimensional temperature field cloud map is generated, which visually presents the temperature distribution with different colors, making the temperature distribution of the machine room clear at a glance. After generating the three-dimensional temperature field cloud map, further identification of polygon hotspot areas with temperature difference exceeding the threshold is required. The hotspot area refers to a local area in the machine room where the temperature is significantly higher than the surrounding environment and may affect the normal operation of equipment. The specific identification process is as follows:

[0030] The solver first calculates the temperature difference of each grid with the same height channel area, filters out all the grids above the threshold, clusters the adjacent super threshold grids through region growing algorithm, forms a continuous high temperature area, connects the vertices of each cluster area boundary grid to generate a closed polygon, and labels the key information of each polygon hotspot area, such as the polygon hotspot area covers the upper 3 grids of cabinet A, the maximum temperature is 32℃, and is labeled as "hot spot 1: cabinet A-middle upper, Tmax=32℃, grid number=3", which ensures accurate positioning of the refrigeration target in subsequent optimization. At the same time of positioning the hotspot area, the dew point temperature safety boundary on the air supply path is calculated based on the temperature field. The dew point temperature refers to the temperature at which water vapor in the air begins to condense into liquid water under the current atmospheric pressure and relative humidity. If the air supply temperature is lower than the dew point temperature, it will cause air outlet condensation, and then cause equipment short circuit, computer room humidity exceeding standard, etc. Therefore, the dew point temperature safety boundary needs to be calculated, that is, the minimum difference between the air supply temperature and the dew point temperature. The specific calculation process is as follows: First, extract the real-time relative humidity (φ,) and atmospheric pressure (P,) of the grid at the air supply outlet from the three-dimensional temperature field. The atmospheric pressure in the computer room is standard atmospheric pressure 101.325kPa. Then, the Magnus-Tetens equation is used to calculate the saturation water vapor pressure (E_s) under the current humidity. The equation form is E_s=0.61078×exp[(17.27×T) / (T+237.3)], where T is the current air supply temperature (22℃). After substitution, E_s≈2.64kPa can be obtained. According to the relative humidity formula φ=E / E_s (E is the actual water vapor pressure), the actual water vapor pressure E=φ×E_s=0.6×2.64≈1.58kPa is calculated. Finally, the Magnus-Tetens equation is used again to back calculate the dew point temperature (T_d), that is, the temperature corresponding to E_s=E. After substituting E=1.58kPa, T_d≈15.2℃ can be calculated. To ensure no condensation, the safety boundary is set to dew point temperature +2℃, that is, the air supply temperature needs to be ≥17.2℃. This value is the dew point temperature safety boundary on the air supply path. Finally, the difference between the current air supply temperature and the dew point temperature is multiplied by the maximum equivalent air supply quantity of the air conditioning system to obtain the upper limit value of the refrigerating capacity increment under the anti-condensation constraint. The maximum refrigerating capacity increment of the air conditioning system refers to the upper limit of the additional refrigerating capacity that the air conditioning system can improve based on the current basic refrigerating capacity. This value is subject to the anti-condensation constraint to avoid excessive increase of refrigerating capacity leading to air supply temperature lower than the dew point. The specific calculation process is as follows:

[0031] Determine the difference between the current supply air temperature and the dew point temperature (ΔT), for example, the current supply air temperature is 22°C, and the dew point temperature is 15.2°C, then ΔT = 22-15.2 = 6.8°C, determine the maximum equivalent supply air volume (Q) of the air conditioning system, which is the actual effective supply air volume after deducting the air duct loss and filter resistance from the rated supply air volume of the air conditioner, for example, the rated supply air volume of the air conditioner is 1000 m³ / h, and the air duct loss rate is 15%, then Q = 1000 x (1-15%) = 850 m³ / h, according to the refrigerating capacity calculation formula (refrigerating capacity = air density x specific heat capacity x supply air volume x temperature difference), wherein the air density (ρ) is 1.2 kg / m³, and the air specific heat capacity (c_p) is 1.005 kJ / (kg·℃), then the refrigerating capacity increment upper limit value (Q_cool) = ρ x c_p x Q x ΔT, and the numerical calculation can be obtained Q_cool = 1.2 x 1.005 x 850 x 6.8 ≈ 6917 kJ / h, which is converted into a refrigerating capacity unit (1 cold ton ≈ 3517 kJ / h) of about 1.97 cold tons, that is, the maximum refrigerating capacity increment upper limit of the air conditioning system is about 2 cold tons. This calculation process not only quantifies the additional refrigerating capacity that the air conditioner can provide, but also ensures that there is no risk of condensation through the constraint of ΔT, providing accurate refrigeration demand boundaries for subsequent fusion of photovoltaic predicted power and lithium charge state calculation of green electricity quota, so that the entire optimization process has a clear goal and safety constraints from the beginning.

[0032] After completing the three-dimensional temperature field hot spot area positioning and air conditioning system maximum refrigerating capacity increment calculation, the maximum refrigerating capacity increment, photovoltaic predicted power and lithium charge state need to be further fused to calculate the green electricity quota that the air conditioner can call, which is a key link connecting refrigeration demand and green electricity supply, which not only connects the refrigerating capacity increment upper limit obtained in the foregoing, but also combines the photovoltaic output and lithium energy storage state, and through multi-dimensional constraints, ensures that the green electricity quota not only meets the air conditioning refrigeration demand, but also meets the safe operation logic of the photovoltaic-lithium battery system, and provides quantitative green electricity allocation basis for subsequent thermal-electric coupling factor generation, and the specific implementation of fusing and calculating the green electricity quota that the air conditioner can call is as follows:

[0033] Convert the maximum refrigeration capacity increment into the equivalent electric power requirement boundary. Since there is a fixed energy efficiency correlation between air conditioner refrigeration capacity and power consumption, the energy efficiency ratio (COP, i.e. the ratio of refrigeration capacity to power consumption) of the air conditioning system needs to be converted. The COP of the machine room air conditioner is usually stable at 3.0-4.0, and is pre-calibrated to 3.5 according to the equipment model. The calculation formula of the equivalent electric power requirement boundary is refrigeration capacity increment boundary power value = maximum refrigeration capacity increment / COP. For example, the maximum refrigeration capacity increment under the anti-condensation constraint calculated in the foregoing is 6917 kJ / h (about 1.97 tons of cold), which is converted to kilowatt unit, i.e. 6917÷3600≈1.92kW, and then divided by COP=3.5, to obtain refrigeration capacity increment boundary power value≈0.55kW. This value represents the upper limit of the additional electric power required by the air conditioner to achieve the maximum refrigeration capacity increment, and is the minimum demand benchmark that the green electricity quota needs to meet. Avoiding insufficient green electricity supply leading to failure to achieve the refrigeration target, convert the state of charge of lithium battery to real-time callable power coefficient through the health state evaluation model. The state of charge (SOC) of lithium battery refers to the percentage of the current remaining capacity of lithium battery to the total capacity. However, only through SOC, the power that can be output by lithium battery cannot be directly judged. For example, although the SOC is high, the health state of the battery is poor, and the actual callable power will decrease significantly. Therefore, the health state evaluation model needs to be introduced. The model takes SOC as the basic input, and combines real-time temperature, cycle number, and historical charge and discharge attenuation rate of lithium battery in three dimensions to calculate the real-time callable power coefficient through weighted summation. The coefficient ranges from 0 to 1, and 1 represents the full power output of lithium battery, and 0 represents no output. For example, the current SOC of a certain lithium battery is 70% (corresponding to the basic coefficient 0.7), the temperature is 25℃ (no attenuation, the coefficient remains 0.7), the cycle number is 800 times (no attenuation), and the monthly attenuation rate is 0.3% (no attenuation). Therefore, the real-time callable power coefficient is 0.7. If another lithium battery has an SOC of 60% (basic coefficient 0.6), a temperature of 38℃ (exceeding the safe range, the coefficient is reduced to 0.6×0.8=0.48), and a cycle number of 1200 times (the coefficient is further reduced to 0.48×0.8=0.384), the final callable power coefficient is 0.384. Multiply the coefficient by the rated output power of the lithium battery to obtain the callable power of the lithium battery, and determine the maximum green electricity support capability that the lithium battery can provide to the air conditioner at present. After obtaining the refrigeration capacity increment boundary power value and the callable power of the lithium battery, the minimum intersection of the photovoltaic predicted power, the refrigeration capacity increment boundary power value, and the callable power of the lithium battery is taken. The photovoltaic predicted power refers to the output power of the photovoltaic system connected to the air conditioner in the next 1 hour obtained through the photovoltaic power prediction model. The logic of taking the minimum intersection is that:

[0034] The green electricity quota needs to meet the photovoltaic energy generation, lithium battery power supply and air conditioning simultaneously, so as to avoid resource waste or system overload caused by limitation of any link. For example, the photovoltaic predicted power is 1.2 kW, the refrigeration capacity increment boundary power value is 0.55 kW, and the lithium battery callable power is 1.92 kW. The minimum value of the three is 0.55 kW, so the preliminary quota upper limit is locked at 0.55 kW, which ensures that the green electricity supply of photovoltaic and lithium battery can cover the air conditioning demand, and avoids excessive consumption of green electricity by air conditioning, which leads to insufficient power supply of other equipment. The inverter efficiency compensation factor is superimposed to generate the air conditioning callable green electricity quota. The output power of photovoltaic and lithium battery is direct current, which needs to be converted into alternating current required by air conditioning through inverter. However, the inverter has energy loss, so the inverter efficiency compensation factor (compensation factor = 1 / inverter efficiency) needs to be introduced to correct the preliminary quota. The correction formula is: final green electricity quota = minimum value of the three × compensation factor. Taking the above 0.55 kW as an example, the final green electricity quota = 0.55 × 1.087 ≈ 0.6 kW. Through this compensation, the energy loss of the inverter is offset, and it is ensured that the green electricity actually obtained by the air conditioner can still meet the 0.55 kW refrigeration power demand. At the same time, considering that the photovoltaic power fluctuates greatly due to weather, the quota is refreshed every cycle during the photovoltaic fluctuation period, which is defined as the period when the difference between the photovoltaic predicted power and the actual output power exceeds 20%. The refresh cycle is set to 5 minutes. Each time the quota is refreshed, the photovoltaic predicted power, lithium battery callable power (real-time update of SOC and health status), and refrigeration capacity increment boundary power value are recalculated. If the temperature of the hot spot area changes, it is recalculated, and the above fusion steps are repeated to obtain a new green electricity quota, so as to ensure that the quota is always matched with the real-time supply and demand state, avoid the disconnection between green electricity supply and air conditioning demand caused by photovoltaic fluctuation, and make the whole AI optimization process smoothly transition from refrigeration demand analysis to green electricity resource allocation stage.

[0035] After completing the fusion calculation of the air conditioning callable green electricity quota, it needs to be real-time bound with the hot spot area positioned in the foregoing, and then a heat-electricity coupling factor is generated. This link is the core of realizing accurate matching between refrigeration demand and green electricity supply, which not only connects the polygon hot spot area positioned by the three-dimensional temperature field in the foregoing and the dynamically refreshed green electricity quota, but also ensures that the green electricity resource flows to the area most in need of refrigeration through the space mapping and dynamic occupation mechanism, avoids waste of quota or insufficient refrigeration of hot spot area, and provides quantitative coupling basis for subsequent linkage mode activation and cooperative decision making. The specific implementation of real-time binding of hot spot area and air conditioning callable green electricity quota and generation of heat-electricity coupling factor is as follows:

[0036] The geometric coordinates, temperature difference amplitude and heat source power of the hotspot area are respectively associated to the quota allocation weight by a space mapping matrix. The space mapping matrix is a matrix model for mathematically associating the space features of the hotspot area with the quota weight. The dimension of the matrix is consistent with the number of the hotspot areas and the dimension of the quota allocation. For example, if there are three hotspot areas in the computer room, the dimension of the matrix is set to 3*3 (3 areas*3 allocation dimensions). Each matrix element represents the weight contribution value of a certain hotspot area in a certain dimension. The specific association process needs to be performed in three steps.

[0037] Quantify the geometric coordinate weight, according to the distance between the hot spot area and the air outlet of the air conditioner, use the distance reciprocal weighting method to calculate, the formula is "geometric coordinate weight = 1 / (1 + distance / 100)" (distance unit is cm, 100 is the calibration coefficient), for example, hot spot 1 is 50 cm away from the air outlet, the weight = 1 / (1+50 / 100)=0.67. Hot spot 2 is 150 cm away from the air outlet, the weight = 1 / (1+150 / 100)=0.4, ensure that the hot spot close to the air outlet can get green electricity support in priority, improve the refrigeration efficiency, quantify the temperature difference amplitude weight, based on the difference between the hot spot area and the average temperature of the same height channel, set the difference weight linear mapping relationship (the weight increases by 0.2 for every 1℃ increase in difference, with a maximum of 1.0), for example, hot spot 1 temperature difference 6℃, weight = 6x0.2 = 1.0, hot spot 2 temperature difference 3℃, weight = 3x0.2 = 0.6, intuitively reflect the influence of temperature difference on quota priority, quantify the heat source power weight, the heat source power refers to the actual heat dissipation power of the equipment corresponding to the hot spot area, which is calculated by the heat flow of the hot spot area in the three-dimensional temperature field, the formula is heat source power = heat flux density x hot spot area", the heat flux density is obtained from the k-ε turbulence model solver, and the weight is calculated by power ratio weighting method, that is, heat source power weight = hot spot heat source power / total heat source power of all hot spots, for example, the heat source power of 3 hot spots is 800W, 500W, 300W, the total is 1600W, then the weight of hot spot 1 is 800 / 1600=0.5, hot spot 2=500 / 1600=0.3125, hot spot 3=300 / 1600=0.1875, ensure that the hot spot corresponding to the high heat dissipation equipment can obtain more green electricity quota. Finally, the weights of the three dimensions are summed by weighting, the weight proportions of geometric coordinates, temperature difference amplitude and heat source power are 0.3, 0.4 and 0.3 respectively, taking into account efficiency and urgency, the final quota allocation weight of each hot spot area is obtained, for example, the final weight of hot spot 1 = 0.67x0.3 + 1.0x0.4 + 0.5x0.3 = 0.201 + 0.4 + 0.15 = 0.751, hot spot 2 = 0.4x0.3 + 0.6x0.4 + 0.3125x0.3 = 0.12 + 0.24 + 0.09375 = 0.45375, hot spot 3 = (assuming the geometric coordinate weight is 0.3) 0.3x0.3 + (the temperature difference of 2℃ corresponds to the weight of 0.4) 0.4x0.4 + 0.1875x0.3 = 0.09 + 0.16 + 0.05625 = 0.30625, the higher the weight value, the higher the priority of the hot spot area in the allocation of green electricity quota, based on the final quota allocation weight, the high weight hot spot area automatically obtains the priority of green electricity quota allocation, the specific implementation logic is:

[0038] First, according to the weight proportion of each hotspot area (the weight of a hotspot / the total weight of all hotspots), the basic quota is allocated, for example, the current air conditioner can call the green electricity quota of 0.6kW, the total weight of hotspots 1, 2 and 3 is 0.751+0.45375+0.30625≈1.511, then the basic quota of hotspot 1 is 0.6×(0.751 / 1.511)≈0.298kW, hotspot 2≈0.6×(0.45375 / 1.511)≈0.18kW, hotspot 3≈0.6×(0.30625 / 1.511)≈0.122kW, the system reserves 10% of the flexible quota for high-weight hotspots, for example, hotspot 1 is a high-weight hotspot, and an additional flexible quota of 0.6×10%=0.06kW is obtained, the final actual quota=0.298+0.06=0.358kW, ensuring that high-priority hotspots can obtain more sufficient green electricity support and quickly eliminate high-temperature hazards. On the basis of the priority allocation mechanism, a quota dynamic preemption mechanism is established, the core of the mechanism is to monitor the temperature changes and new hotspot generation of the hotspot area in real time, when the system detects that a new hotspot is generated through the three-dimensional temperature field, that is, the temperature difference between a certain grid and the average temperature of the same height channel suddenly exceeds 5℃, and a continuous threshold grid cluster is formed, the preemption process is triggered immediately, which is specifically reflected in the three-level preemption strategy:

[0039] New hotspot priority determination, calculate the quota allocation weight of new hotspot (according to the three-dimensional weighting method in the previous text), if the new hotspot weight is higher than all existing hotspots (such as the highest existing weight 0.751, the new hotspot weight 0.82), it is determined as a high-priority new hotspot, and a first-level preemption is started, if the weight is between the existing hotspots (such as 0.5), it is a medium-priority new hotspot, and a second-level preemption is started, if the weight is the lowest (such as 0.2), it is a low-priority new hotspot, and no preemption is performed, only waiting for the next round of quota refresh, quota reduction and redirection, in the first-level preemption, 30% of the quota of the existing low-weight hotspots (the last 50% in the weight ranking) is reduced, for example, the existing low-weight hotspots 2 and 3 have quotas of 0.18kW and 0.122kW respectively, after reducing 30%, they become 0.126kW and 0.085kW respectively, a total of 0.091kW is reduced, and all of them are redirected to the new hotspot, the final quota of the new hotspot = the basic quota allocated according to the weight + 0.091kW, in the second-level preemption, only 20% of the quota of the lowest-weight hotspot is reduced, and redirected to the new hotspot, to avoid excessive impact on the refrigeration of multiple existing hotspots, after preemption, verification, after the quota is redirected, the temperature change of the hotspots whose quotas are reduced is monitored in real time, if the temperature of the hotspots rises by more than 2℃ within 10 minutes, 50% of the reduced quota is automatically restored, to balance the refrigeration demand of new and old hotspots, for example, the temperature of hotspot 3 rises from 28℃ to 30.5℃ after being reduced, 0.122×30%×50%≈0.018kW of the quota is restored, to ensure that the original hotspots will not be deteriorated due to preemption, through the weight correlation of the space mapping matrix, high-weight priority allocation and dynamic preemption mechanism, the hotspot area and the green electricity quota are deeply bound, and the coupling relationship between the two is finally quantified as a "heat-electricity coupling factor", which is the average value of the actual allocation amount of all hotspot area quotas / demand amount, the demand amount is the equivalent electric power required for refrigeration of the hotspot area, for example, hotspot 1 requires 0.35kW, the actual allocation is 0.358kW, hotspot 2 requires 0.18kW, the actual allocation is 0.126kW, hotspot 3 requires 0.12kW, the actual allocation is 0.085kW, and the new hotspot requires 0.2kW, the actual allocation is 0.091kW, then the coupling factor = (0.358 / 0.35+0.126 / 0.18+0.085 / 0.12+0.091 / 0.2) / 4≈(1.023+0.7+0.708+0.455) / 4≈0.721, the closer the factor is to 1, the higher the heat-electricity matching degree, which provides a key quantitative basis for activating the linkage mode combined with the intensity of photovoltaic power fluctuation.

[0040] After generating the thermal-electric coupling factor, it needs to be combined with the photovoltaic power fluctuation intensity to activate the corresponding linkage mode, which is the key to connecting thermal-electric resource matching and collaborative decision execution. It not only connects the thermal-electric coupling factor generated by binding the hot spot area and green electricity quota, but also combines the real-time stability of photovoltaic output to dynamically select the adaptive linkage mode through a double threshold trigger mechanism, avoiding the complex scenarios of photovoltaic fluctuation and thermal-electric coupling changes that a single mode cannot handle. It provides a mode basis for the subsequent deep deterministic policy gradient model to execute the collaborative decision of the wind valve, water temperature, and lithium battery. The specific implementation is as follows:

[0041] The core parameter definition needs to be clarified. The photovoltaic power fluctuation intensity refers to the deviation amplitude of the actual output power of photovoltaic per unit time and the predicted power. It is calculated by the maximum difference of power in a 5-minute sliding window / average value of predicted power x 100%, which is used to quantify the stability of photovoltaic output. The high active interval of the thermal-electric coupling factor is calibrated according to historical tuning data, which refers to the interval where the factor value is greater than or equal to 0.6. The set threshold in the double threshold trigger mechanism is set to 20%, and the stable threshold (photovoltaic fluctuation duration threshold) is set to 10 minutes. Both of them together constitute the judgment standard for mode switching. When the photovoltaic power fluctuation intensity exceeds the set threshold (20%) and the thermal-electric coupling factor is in the high active interval (≥0.6), the system automatically activates the wind valve-water temperature linkage mode. The refinement and innovation of this mode lies in the dynamic proportional linkage + scenario-based constraint calibration mechanism. Specifically, after activation, the system first adjusts the linkage proportion of the wind valve and water temperature according to the specific value of the current thermal-electric coupling factor. If the coupling factor is between 0.6 and 0.8, the basic proportion of the wind valve opening increment: water temperature lifting amplitude = 1:0.5 is adopted to ensure the balance between cold supply and green electricity consumption. If the coupling factor is greater than or equal to 0.8 (high thermal-electric matching degree), the proportion is optimized to "1:0.7" (wind valve opening degree increases by 10%, water temperature increases by 7°C), which further reduces the air conditioning power consumption by increasing the water temperature lifting amplitude to adapt to the green electricity supply rhythm during photovoltaic fluctuation. At the same time, scenario-based constraint calibration is introduced:

[0042] For different types of equipment in different areas of the machine room, the linkage ratio is fine-tuned. The server cabinet area equipment is sensitive to temperature, and the upper limit of the water temperature increase amplitude is set to 8°C to avoid high temperature affecting equipment operation. The storage equipment area is more sensitive to humidity, and the upper limit of the air valve opening increase amplitude is set to 15% to prevent rapid air supply causing local humidity to drop suddenly, causing static risk. In addition, the system will collect the difference between the air supply temperature and the dew point temperature after the air valve adjustment. If the difference is less than 2°C (close to the risk of condensation), the water temperature increase amplitude will be automatically reduced by 20% to ensure that the linkage process always meets the anti-condensation constraint and avoids neglecting equipment safety in pursuit of green power consumption. When the photovoltaic fluctuation continues to exceed the stability threshold, it indicates that the photovoltaic output is unstable for a long time, and simply relying on the air valve-water temperature linkage cannot maintain thermal-electric balance. The system is forced to switch to the lithium battery-water temperature buffer mode, which focuses on the dual protection of air valve speed locking and staged lithium battery smoothing. In terms of locking the air valve adjustment speed, the system first sets the upper limit of the air valve adjustment speed according to the current compressor load state. The upper limit of the speed is set to 5% / minute for low load, 3% / minute for medium load, and 1% / minute for high load. By strictly locking the speed, the system avoids frequent air valve adjustment causing compressor load fluctuation, which in turn causes further imbalance between green power consumption and photovoltaic supply. The system will monitor the feedback signal of the air valve actuator in real time. If the actual adjustment speed exceeds the upper limit by 5%, a mechanical brake command is triggered immediately to force the speed back within the threshold, ensuring stable air valve state. In terms of enabling lithium battery smoothing strategy, a three-stage power compensation mechanism is adopted:

[0043] The first stage is basic compensation. When the actual photovoltaic power is 10%-20% lower than the predicted power, the lithium battery discharges at a ratio of 1:1 of the "predicted power-actual power" to supplement the green power gap. For example, the predicted power is 1.2kW, the actual power is 1.0kW, and the lithium battery discharges 0.2kW. The second stage is enhanced compensation. When the deviation exceeds 20%-30%, the lithium battery discharge ratio is increased to 1:1.2, and the standby capacity of the lithium battery is called to avoid insufficient compensation. The third stage is emergency compensation. When the deviation exceeds 30%, in addition to the lithium battery discharging at full power, a photovoltaic-mains switch signal is triggered simultaneously to ensure uninterrupted air conditioning power supply. In addition, the discharge depth of the lithium battery (current discharge amount / capacity) is calculated in real time during the smoothing process. If the depth exceeds 30%, the compensation ratio is automatically reduced to 1:0.8 to balance power supply stability and lithium battery health status, allowing the entire AI optimization process to naturally transition from mode activation to specific decision execution phase.

[0044] One of the core execution logics after the activation of the wind valve-water temperature linkage mode is to dynamically require the synchronous increase of the water temperature set value when increasing the opening degree of the hot spot area wind valve. This operation not only inherits the core goal of the linkage mode in the previous paragraph, which is to adapt to green power supply through the coordinated adjustment of the wind valve and water temperature, but also avoids the risk of condensation by presetting the proportional relationship and the nonlinear mapping table, ensuring the refrigeration effect of the hot spot area while avoiding equipment failure caused by excessively low supply air temperature due to single adjustment of the wind valve. It provides a safety constraint basis for the wind valve opening degree decision of the deep deterministic policy gradient model, and the specific implementation is as follows:

[0045] The definition of the explicit preset proportional relationship is not a fixed linear proportion, but refers to the minimum adaptive relationship that the wind valve opening degree increase amplitude and the water temperature increase amplitude need to meet under different wind valve opening degree increase amplitudes and different supply air humidity scenarios. The core purpose is to ensure that the water temperature is increased synchronously when the wind valve opening degree is increased to maintain the supply air temperature not lower than the dew point temperature. The essence is the collaborative anti-condensation constraint of air volume-temperature. For example, when the wind valve opening degree is increased by 10%, if the supply air humidity is 60%, the water temperature needs to be increased by at least 4°C, and if the humidity is 70%, the water temperature needs to be increased by at least 6°C, which reflects the logic that the higher the humidity, the more stringent the water temperature increase requirement. The key to achieving this preset proportional relationship is to establish a nonlinear mapping table of wind valve opening degree increase amplitude and water temperature increase amplitude. The construction of this mapping table needs to be based on historical condensation event data as the core training basis to ensure coverage of common operating scenarios in the machine room. The specific refinement process consists of three steps:

[0046] Data collection and labeling, collect the past 1 year in the computer room of the wind valve adjustment record (opening increment range 0-20%), water temperature change record (elevation amplitude 0-10℃), supply air humidity data (40%-80%) and condensation event record (labeling each time the opening increment, water temperature, humidity parameters when condensation occurs), form an effective sample, for example, a sample is labeled as wind valve opening increment 15%, water temperature elevation 3℃, humidity 75%, condensation occurs, feature association and model training, with wind valve opening increment (x axis), supply air humidity (z axis) as input features, water temperature elevation amplitude (y axis) as output target, using gradient boosting tree (GBT) model to train the nonlinear mapping relationship, the model will automatically learn that the higher the humidity, the larger the opening increment, the larger the required water temperature elevation amplitude, for example, when the opening increment is 5% and the humidity is 50%, the model outputs the minimum water temperature elevation amplitude of 2℃, when the opening increment is 15% and the humidity is 75%, the output minimum elevation amplitude is 8℃, mapping table generation and calibration, the nonlinear relationship obtained by model training is discretized into table form, the row dimension of the table is the wind valve opening increment (divided by 1% step, a total of 21 rows), the column dimension is the supply air humidity (divided by 5% step, a total of 9 columns), and the cell is the corresponding minimum water temperature elevation amplitude. At the same time, combine the rated water temperature range of air conditioning equipment to calibrate the table. If the model output water temperature elevation amplitude leads to the final water temperature exceeding 18℃, adjust the cell value to the maximum allowed elevation amplitude. If the current water temperature is 16℃, the maximum elevation amplitude is 2℃, to avoid water temperature exceeding the upper limit of equipment operation. When the wind valve opening is adjusted in actual execution, the minimum water temperature elevation requirement is obtained according to the current supply air humidity by looking up the table, and the specific refinement process needs to be closely linked with real-time data collection:

[0047] When the wind valve controller receives an instruction to increase the opening degree of the hot spot area wind valve, such as an increase from 40% to 50%, an increase of 10%, it immediately collects the current supply air humidity through the humidity sensor in the air duct, locates the corresponding row (opening degree increase of 10%) and column (humidity of 65%) in the non-linear mapping table, reads the minimum water temperature increase requirement in the cell, and sends the requirement to the water temperature controller. After receiving the water temperature controller, calculate the difference between the current water temperature and the target water temperature, such as the current water temperature is 13℃, the target water temperature needs to be ≥13+5=18℃, if the water temperature controller can normally increase the water temperature to 18℃, meet the minimum requirement, then feedback the water temperature ready signal, the wind valve controller executes the opening degree increase operation, if due to equipment failure or water temperature has reached the upper limit, such as the current water temperature is 17℃, can only increase 0.5℃, cannot meet the requirement of 5℃, then trigger the soft limit of the wind valve opening degree, which is different from the forced stop, that is, the wind valve controller recalculates the maximum allowed opening degree increase according to the actual water temperature increase, the formula is "allowed opening degree increase = instruction increase × (actual water temperature increase / minimum water temperature increase requirement)", for example, the instruction increase is 10%, the actual increase is 0.5℃, the minimum requirement is 5℃, then the allowed opening degree increase = 10% × (0.5 / 5) = 1%, that is, the wind valve is only increased from 40% to 41%, not the original 50%, at the same time, the system sends a warning signal of insufficient water temperature and wind valve limit to the digital twin platform, reminding the operation and maintenance personnel to check the water temperature equipment, avoiding dew due to insufficient water temperature, and trying to meet the cooling demand of the hot spot area through limited opening degree increase, balancing safety and functionality, so that the whole collaborative decision-making process not only has the scientific nature of data-driven, but also meets the operation safety specifications of the machine room equipment.

[0048] In the water temperature setting decision-making link, the lithium battery state is the key constraint condition for determining whether to increase the water temperature to absorb green electricity. This judgment not only inherits the logic of ensuring system stability through wind valve action constraints in the previous lithium battery charging and discharging decision-making, but also ensures that the water temperature increase operation has reliable energy storage support through the three judgment standards of the lithium battery ready mode, avoiding green electricity absorption interruption or battery damage due to poor lithium battery state, while providing a safety bottom for extreme scenarios through degradation strategy, making the water temperature setting decision-making more scientific and fault-tolerant. The specific implementation is as follows:

[0049] The lithium battery in the ready mode needs to be accurately defined, which is not a single state judgment, but also meets the three requirements of the state of charge (SOC) in the optimal life interval, the battery temperature in the thermal stability window and the state of health (SOH) higher than the decline threshold. The refinement and innovation of this judgment logic are reflected in the dynamic interval calibration + multi-dimensional coupling verification mechanism. The state of charge (SOC) refers to the percentage of the current remaining capacity of the lithium battery to the total capacity, and its optimal life interval is not a fixed value, but is dynamically adjusted according to the number of lithium battery cycles. The optimal interval of new batteries (cycle number <200 times) is set to 40%-80%, which can avoid lithium dendrite growth caused by full charge storage and meet the instantaneous discharge demand. The optimal interval of batteries with cycle number 200-800 times is adjusted to 35%-75%, which reduces battery degradation by narrowing the upper limit. The optimal interval of batteries with cycle number >800 times is further adjusted to 0%-70%, which balances capacity loss and power supply stability. The interval parameters are updated in real time by the battery management system (BMS) to ensure adaptation to the aging degree of the battery. The thermal stability window of the battery temperature is set according to the room temperature and the type of the battery. The thermal stability window of the ternary lithium battery is 15-35℃, and the thermal stability window of the iron phosphate lithium battery is 10-40℃. The system collects real-time temperature through NTC temperature sensors attached to the surface of the battery monomer. If the temperature is within the window for 3 consecutive times, and the temperature fluctuation amplitude is <2℃ / minute, the temperature is determined to be qualified, avoiding false judgment caused by instantaneous temperature fluctuation. The state of health (SOH) refers to the ratio of the current capacity of the battery to the rated capacity, reflecting the aging degree of the battery, and the decline threshold is set to 80% (i.e. SOH≥80% is determined to be not significantly degraded). The calculation of SOH needs to be combined with the charging curve and the discharge curve to make a comprehensive judgment. The BMS records the capacity change of each charge and discharge, and the average method of charge and discharge capacity is used to calculate SOH. For example, a lithium battery with a rated capacity of 50kWh can only store 42kWh when fully charged, so SOH=42 / 50×100%=84%, which is higher than the decline threshold. If it can only store 38kWh, SOH=76%, which is lower than the threshold. At the same time, the decline acceleration warning is introduced. When SOH decreases by more than 2% within 1 month, even if it is still higher than 80%, the pre-processing mechanism will be triggered to prevent rapid degradation of the battery in advance. Only when the above three requirements are met, the lithium battery is determined to be in the ready mode, and the system can normally execute the water temperature raising operation to absorb green electricity. If any requirement is not met, the lithium battery is determined to be in the non-ready mode, and the degradation strategy needs to be started immediately.

[0050] Check the status of the standby lithium battery group. The standby lithium battery group uses the same readiness mode judgment standard as the main lithium battery group. If the standby group is in the readiness mode, the relay will automatically switch to the standby group for power supply, ensuring uninterrupted green power supply, and the water temperature will still be raised according to the original request amplitude. If the standby group is also in the non-readiness mode, or the machine room is not configured with a standby group, the water temperature raising amplitude will be limited to the preset safety baseline. The safety baseline is dynamically calculated based on the current lithium battery status. The formula is safety baseline water temperature = current water temperature + (lithium battery actual status value / standard threshold) x original request raising amplitude. For example, the original request water temperature is raised from 14°C to 18°C (amplitude 4°C), and the current SOC = 32% (standard threshold lower limit 40%). The safety baseline water temperature = 14 + (32 / 40) x 4 = 17.2°C, i.e. the water temperature can only be raised by 3.2°C. This avoids the sudden increase in air conditioning load after the water temperature is raised due to insufficient lithium battery power supply, causing system voltage fluctuations. At the same time, the system will generate an alarm log on the digital twin platform for lithium battery non-readiness and water temperature amplitude limitation, recording the current lithium battery status parameters (SOC, temperature, SOH) and water temperature adjustment amplitude, which is convenient for subsequent tracing and battery maintenance for maintenance personnel. The entire AI optimization process always prioritizes equipment safety while pursuing green power consumption efficiency.

[0051] In the lithium battery charging and discharging decision, the discharging operation in the peak period of photovoltaic needs to be strictly restricted by the action of the air valve. This design not only implements the core logic of the previous multi-dimensional constraint to ensure the safety of the collaborative decision, but also, according to the characteristics of the peak period of photovoltaic, adjusts the speed of the air valve and the load of the compressor to avoid the impact of lithium battery over-limit discharging on the compressor. This provides a hardware-level safety guarantee for the charging and discharging decision of the deep deterministic policy gradient model. The specific implementation is as follows:

[0052] The core logic of restricting lithium battery discharging in the peak period of photovoltaic through air valve action is that the peak period of photovoltaic usually refers to the noon period (11:00-13:00). At this time, the predicted power of photovoltaic exceeds 30% of the regular power demand of air conditioning. The system will preferentially use the excess green power of lithium battery storage, and then discharge to supplement when the subsequent photovoltaic output decreases. However, if the lithium battery discharging power is too large, it will cause the voltage of the air conditioning compressor to rise sharply. If the adjustment speed of the air valve is too fast (the opening degree increases by more than 15% in a short time), it will further exacerbate the load fluctuation of the compressor, causing equipment abnormal noise, shortened service life, and other problems. Therefore, the lithium battery discharging power needs to be inversely restricted by the action of the air valve (adjustment speed) to form a linkage protection mechanism of the air valve-compressor-lithium battery. The key to achieving this restriction is to establish a correlation model between the adjustment speed of the air valve and the change rate of the compressor load. The adjustment speed of the air valve refers to the change amplitude of the opening degree of the air valve per unit time, and the change rate of the compressor load refers to the change amplitude of the ratio of the actual load of the compressor to the rated load per unit time. The model establishment process needs to complete data fitting and verification based on historical operation data:

[0053] The wind valve adjustment record, compressor load monitoring data and corresponding lithium battery discharge power data in the photovoltaic peak period within 1 year of the machine room are collected to form effective samples, such as a sample record of a wind valve adjustment rate of 8% / min, a compressor load change rate of 4.2% / min, and a lithium battery discharge power of 2.5kW. A multiple linear regression algorithm is used to construct a correlation model, with the wind valve adjustment rate (x) as the independent variable and the compressor load change rate (y) as the dependent variable. The regression equation "y=0.5x+0.2" is obtained by least squares fitting (fitting goodness R²≥0.92 to ensure model reliability). The equation shows that for every 1% / min increase in the wind valve adjustment rate, the compressor load change rate increases by an average of 0.5% / min, directly reflecting the positive correlation between the two. The compressor tolerance threshold is introduced to calibrate the model, and the tolerance threshold of the load change rate is determined according to the technical parameters of the compressor model. The tolerance threshold is substituted into the regression equation to back-calculate the corresponding wind valve adjustment rate threshold. When the wind valve adjustment rate exceeds 15% / min, the compressor load change rate will exceed the tolerance threshold, and the lithium battery discharge constraint needs to be started. When the wind valve adjustment rate exceeds the compressor tolerance threshold, the system automatically generates a lithium battery discharge power upper limit curve. The generation logic of this curve is based on the inverse suppression relationship between "load-power", and the specific process is as follows:

[0054] The current compressor load change rate corresponding to the associated model is obtained (such as the adjustment rate 18% / min, and the load change rate =0.5x18+0.2=9.2% / min is obtained by substituting the equation), the amplitude of the load change rate exceeding the threshold (9.2%-8%=1.2% / min) is calculated, which is negatively correlated with the upper limit of the lithium battery discharge power, the greater the amplitude, the more the discharge power needs to be limited, and an exponential decay function is used to generate the upper limit curve, the function expression is "P upper limit=P ratedx e^(-k x Ay)", wherein P rated is the rated discharge power of the lithium battery, k is the attenuation coefficient, and Ay is the load change rate exceeding amplitude (P upper limit=5x e^(-0.8x1.2)=5x e^(-0.96)≈5x0.382≈1.91kW is obtained by substituting), at the same time, the horizontal axis of the curve is the adjustment rate of the air valve, and the vertical axis is the upper limit of the discharge power of the lithium battery, along with the adjustment rate from 15% / min to 25% / min, the upper limit power gradually decreases from 5kW (when the adjustment rate=15% / min, Ay=0, P upper limit=5x e^0=5kW) to less than 0.5kW, forming a clear decay trend of higher rate and lower power limit, ensuring that the compressor load is always in a safe range, if the lithium battery management system detects that the current discharge power attempts to exceed the value corresponding to the upper limit curve, an air valve adjustment rate forced deceleration instruction is triggered immediately, the instruction is sent to the air valve actuator through the real-time control channel of the digital twin platform, and the actuator reduces the air valve adjustment rate from the current value (18% / min) to the safe speed (such as 16% / min, at this time, P upper limit=5x e^(-0.8x(0.5x16+0.2-8))=5x e^(-0.8x0.2)=5x0.852≈4.26kW, which can cover the current discharge power of 2.2kW) corresponding to the upper limit curve at a deceleration rate of 2% / s, at the same time, the system sends a power over-limit warning to the lithium battery management system, temporarily limits the lithium battery discharge power to the current upper limit value, until the air valve adjustment rate decreases to the safe range, forming an over-limit-deceleration-limited power closed loop control, which not only avoids damage to the compressor due to overload, but also ensures the continuous operation of the lithium battery discharge operation through dynamic adjustment. With the lithium battery readiness mode determination and water temperature setting constraints, a multi-dimensional safety protection network is formed to ensure the stability and safety of the system during the peak period of photovoltaic power generation.

[0055] In the collaborative decision-making process of the deep deterministic policy gradient model for the opening degree of the air valve, the water temperature setting, and the charging and discharging of lithium batteries, the three are not independent operations, but form a mutual constraint and dynamic calibration relationship through a closed-loop verification chain. This mechanism not only incorporates the single-dimensional safety rules established for the air valve (anti-condensation constraint), water temperature (lithium battery state constraint), and lithium battery (air valve action constraint) in the previous text, but also integrates the dispersed constraints into a systematic protection through the trigger-check-feedback cycle logic, avoiding the risk of global optimization caused by the failure of single-link verification, and ensuring that the air conditioning terminal machine room always maintains a balance between green power consumption and equipment safety. The specific implementation of the closed-loop verification chain formed by the constraints of the air valve opening degree, the water temperature setting, and the lithium battery charging and discharging is as follows:

[0056] The start of the closed-loop verification chain begins with the wind valve opening increment triggering a water temperature compensation request. When the system generates a wind valve opening increase instruction based on the cooling demand of the hot spot area, the wind valve controller immediately sends a water temperature compensation request to the water temperature controller, which includes key parameters such as the current wind valve opening increment and the supply air humidity. The core purpose is to obtain the minimum water temperature increase requirement that matches the opening increment based on the nonlinear mapping table established in the previous section. At this time, the water temperature compensation request becomes the first constraint connecting the wind valve and the water temperature, ensuring that the water temperature can be synchronized to increase when the wind valve opening increases to avoid the risk of condensation. If the water temperature controller does not receive the request or the request parameters are missing, the wind valve opening adjustment will automatically pause until the request is transmitted normally. Then, the water temperature increase demand triggers lithium state verification. After receiving the compensation request, the water temperature controller does not directly execute the water temperature increase operation, but first sends a lithium state verification instruction to the lithium battery management system (BMS). The verification content strictly follows the three requirements of the lithium battery readiness mode defined in the previous section, i.e., whether the state of charge (SOC) is in the optimal life interval, whether the battery temperature is in the thermal stability window, and whether the state of health (SOH) is higher than the degradation threshold (80%). After the lithium battery management system completes the verification through real-time battery parameters, it feeds back a "ready" or "not ready" signal to the water temperature controller: if the feedback is "ready", it means that the lithium battery has the required discharge capacity to support the water temperature increase (green electricity supply is sufficient), and the water temperature controller can execute the increase operation as requested. If the feedback is "not ready", it triggers the preliminary restriction of water temperature regulation. At this time, the water temperature controller can only increase the water temperature to the pre-set safety baseline, and simultaneously synchronizes the verification result to the wind valve controller to provide a basis for subsequent wind valve speed adjustment. This step forms the second constraint between water temperature and lithium battery, avoiding the forced increase of water temperature when the lithium battery state is poor, which leads to the interruption of green electricity supply. Subsequently, the lithium action is fed back to the wind valve speed regulation. Regardless of the lithium state verification result, the lithium battery executes the charge and discharge action, which produces real-time power data. These data are fed back to the wind valve controller through the closed-loop channel. The wind valve controller combines the wind valve regulation speed-power compressor load change rate correlation model established in the previous section to determine whether the current lithium battery discharge power will cause the compressor load to exceed the tolerance threshold.For example, when the lithium battery discharges power reaches 3kW, the corresponding compressor load rate may rise to 7% / minute. If the air valve adjustment rate is 14% / minute at this time, the air valve controller will automatically reduce the adjustment rate to 12% / minute to prevent the load from further rising. If the lithium battery reduces the discharge to 1kW due to the non-ready mode, the compressor load pressure decreases, and the air valve controller can appropriately relax the rate limit. This feedback process constitutes the third constraint between the lithium battery and the air valve, achieving dynamic adaptation of the three states. In the entire closed-loop verification chain, any link verification failure will trigger global strategy reorganization. If the water temperature controller fails to respond due to equipment failure after the air valve triggers a water temperature compensation request (first link failure), or the lithium battery verification triggered by the water temperature increase demand continues to feedback non-ready (second link failure), or the air valve rate cannot match the load requirement after the lithium battery feedback (third link failure), the system will immediately call the global strategy reorganization module. The reorganization operation is not simply to terminate the current decision, but to seek a new dynamic balance through multi-dimensional adjustment.

[0057] The air valve adjustment target is reduced, such as the original target opening of 50% to 45%, reducing the demand for water temperature compensation. Secondly, the water temperature increase range is reduced, such as the original request to increase 5°C to 3°C, reducing the dependence on lithium battery discharge. Finally, the working mode is switched according to the lithium battery state, such as switching from "discharge support" mode to "mains auxiliary" mode by introducing mains to supplement the green power gap. During the reorganization process, the system will re-execute the closed-loop verification every 30 seconds until the air valve opening, water temperature setting, and lithium battery charging and discharging all meet the constraint conditions, and finally a new dynamic balance is achieved. The AI optimization process can still continue to advance when local links fail, ensuring fault tolerance of system operation and continuing the optimization logic of safety first and collaborative adaptation described above.

[0058] After the air valve opening, water temperature setting, and lithium battery charging and discharging reach a dynamic balance through the closed-loop verification chain, the overall effect of collaborative decision-making needs to be evaluated through a joint reward function. This link is the key to the end and feedback of the photovoltaic lithium battery air conditioner terminal machine room AI optimization process based on digital twinning. It not only receives the three control instructions output by the deep deterministic policy gradient model described above, but also ensures the completeness and accuracy of the evaluation through the control quantity binding submission mechanism, avoiding decision bias caused by single instruction evaluation. At the same time, it provides quantitative basis for subsequent optimization strategy iteration. The specific implementation of the joint reward function overall evaluation using the control quantity binding submission mechanism is as follows:

[0059] The core logic of this mechanism is the coordination of three control instructions: wind valve opening degree (adjusting the distribution of cold energy in the hot spot area), water temperature setting (adapting to green power consumption and preventing dew formation), and lithium battery charging and discharging (balancing photovoltaic fluctuations and stable power supply). Any missing or delayed instruction will result in an evaluation result that cannot reflect the true optimization effect, so it is necessary to synchronize and bind the evaluation of the three instructions. Specifically, the system sets a unified instruction sending period for the wind valve controller, water temperature controller, and lithium battery management system (1 minute, consistent with the photovoltaic quota refresh period). Within each period, the three control instructions must be transmitted to the joint reward function evaluator within the same time window. If the evaluator only receives the wind valve opening degree and water temperature setting instructions within 2 seconds, or the time difference between the received instructions exceeds 2 seconds, it is determined that the instructions have not arrived synchronously. In this case, the evaluator only calculates the local reward value corresponding to the received instructions (only calculates the contribution of the wind valve opening degree to hot spot elimination and the control effect of the water temperature setting on dew risk), and returns an error code (such as "E001 - lithium instruction missing" or "E002 - instruction out of sync") to the digital twin platform, prompting the operator to check the instruction transmission link. No joint control vector is generated in this period to avoid making incorrect judgments based on incomplete data. When the three control instructions arrive synchronously at the evaluator, the evaluation process enters the next stage. First, it checks the consistency of the instruction timestamps. The evaluator extracts the timestamps of the three instructions and calculates the difference between the maximum and minimum timestamps. If the difference is ≤500 milliseconds, the timestamps are considered consistent, and the instructions are considered to be the coordinated output of the same decision-making period. If the difference is >500 milliseconds, even if the instructions arrive in the same time window, they are considered to be inconsistent. In this case, the error code "E003 - inconsistent timestamps" is returned, and the instruction retransmission mechanism is triggered to ensure that the evaluation is based on valid instructions from the same decision-making period. After the timestamp verification passes, the weighted fusion layer calculates the comprehensive scores of photovoltaic consumption rate, hot spot elimination rate, lithium battery loss index, and dew risk coefficient. These four indicators are the core dimensions for measuring optimization effect, and different weights are assigned according to the operating priority of the machine room (photovoltaic consumption rate weight 0.3, hot spot elimination rate weight 0.3, lithium battery loss index weight 0.2, and dew risk coefficient weight 0.2, with a total weight of 1).The photovoltaic consumption rate refers to the proportion of green electricity actually consumed by the air conditioner (including photovoltaic direct power supply and lithium battery discharge) in the total predicted photovoltaic power generation, and the calculation formula is: (photovoltaic direct power supply + lithium battery discharge) / photovoltaic prediction amount x 100%, for example, if the photovoltaic prediction amount is 10 kWh, the direct power supply amount is 6 kWh, and the lithium battery discharge amount is 2 kWh, then the consumption rate = 8 / 10 x 100% = 80%, the hotspot elimination rate refers to the proportion of the number of hot spot areas whose temperature drops below the safety threshold in the total number of hot spot areas in the evaluation period, for example, if 3 out of the original 5 hot spot areas meet the temperature standard, then the elimination rate = 3 / 5 x 100% = 60%, the lithium battery loss index is an index for measuring the life loss in the charging and discharging process of lithium battery, with a value of 0-1, 0 representing no loss and 1 representing serious loss, which is calculated by (actual discharge depth / rated safe discharge depth) x 0.5 + (actual temperature - optimal temperature) / (thermal stability window upper limit - optimal temperature) x 0.5, for example, if the actual discharge depth is 40%, the rated safe depth is 60%, the actual temperature is 30℃, the optimal temperature is 25℃, and the window upper limit is 35℃, then the loss index = (40 / 60) x 0.5 + (30-25) / (35-25) x 0.5 ≈ 0.33 + 0.25 = 0.58, the condensation risk coefficient is a safety index for evaluating the difference between the supply air temperature and the dew point temperature (with a value of 0-1, 0 representing no risk and 1 representing high risk), and the calculation formula is: 1 - (supply air temperature - dew point temperature) / safety difference threshold value" (the safety difference threshold value is set to 5℃), for example, if the supply air temperature is 22℃, the dew point temperature is 15℃, and the difference is 7℃, then the risk coefficient = 1 - 7 / 5 = -0.4 (take 0, representing no risk), if the difference is only 2℃, then the risk coefficient = 1 - 2 / 5 = 0.6.The weighted fusion layer first normalizes the four indicators to 0-100 points, then calculates the comprehensive score according to the weight, finally outputs the joint control vector and marks the cooperative effectiveness level, the joint control vector is a numerical vector integrated by three control instructions according to the preset format, which is convenient for digital twin platform storage and subsequent calling, the cooperative effectiveness level is divided according to the comprehensive score, 90-100 points is "A level (excellent cooperation)", 70-89 points is "B level (good cooperation)", 50-69 points is "C level (basic cooperation)", and 50 points or less is "D level (cooperation failure)", the system will append the level label in the joint control vector and push it to the control terminal and cloud management platform of the air conditioner terminal machine room: A level and B level vector is directly executed, C level vector needs to be executed after artificial confirmation, and D level vector triggers global strategy reorganization until the cooperative effectiveness level is improved to C level and above, through the control amount binding submission mechanism and multi-dimensional comprehensive evaluation, the joint reward function not only ensures the integrity and objectivity of the evaluation result, but also provides clear guidance for decision execution through the cooperative effectiveness level, echoing the logic of the closed-loop verification chain in the foregoing to ensure decision safety, so that the entire AI optimization process forms a complete closed loop from decision generation, constraint verification to effect evaluation, and finally realizes the multi-objective balance of efficient consumption of photovoltaic green electricity, accurate control of machine room hot spots, safe operation of lithium batteries and prevention of dewing risk.

[0060] The present application generates a three-dimensional temperature field to locate the hot spot area by deploying a distributed fiber temperature sensor array, combines a k-ε turbulence model to calculate the maximum refrigeration capacity increment of the air conditioner, fuses the increment, the predicted power of photovoltaic and the state of charge of lithium battery, superimposes the inverter efficiency compensation factor to obtain the callable green electricity quota of the air conditioner, binds the hot spot and the quota to generate a thermal-electric coupling factor, activates the linkage mode combined with the photovoltaic fluctuation intensity, triggers the deep deterministic policy gradient model to execute the collaborative decision of the air valve opening degree, water temperature setting and lithium battery charging and discharging, binds the three control amounts through the joint reward function, and outputs the joint control vector, which improves the green electricity consumption rate and temperature control accuracy, ensures the safety of equipment, and is suitable for efficient operation of the machine room.

[0061] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited to the above examples, the above examples and descriptions in the specification are only preferred examples of the present application, and are not intended to limit the present application, various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning, characterized by: The method comprises the following steps: According to the three-dimensional temperature field, the hot spot area is located and the maximum refrigerating capacity increment of the air conditioning system is calculated, and the maximum refrigerating capacity increment, the photovoltaic predicted power and the lithium battery charge state are fused to calculate the green electricity quota available for the air conditioner; According to the three-dimensional temperature field, the hot spot area is located and the maximum refrigerating capacity increment of the air conditioning system is calculated, and the maximum refrigerating capacity increment, the photovoltaic predicted power and the lithium battery charge state are fused to calculate the green electricity quota available for the air conditioner; The distributed fiber-optic temperature sensor array is deployed on the surface of the cabinet at a non-uniform density, wherein the sensor deployment density in the high heat density area is doubled, the synchronously collected temperature data is input into a fluid dynamics solver based on the k-ε turbulent flow model, a three-dimensional temperature field cloud map is generated and a polygon hot spot area with a temperature difference exceeding a threshold is identified, and a dew point temperature safety boundary on the air supply path is calculated based on the temperature field, the difference between the current air supply temperature and the dew point temperature is multiplied by the maximum equivalent air supply capacity of the air conditioning system to obtain an upper limit value of the refrigerating capacity increment under the anti-condensation constraint; The maximum refrigerating capacity increment is converted into an equivalent electric power demand boundary, the lithium battery charge state is converted into a real-time available power coefficient through a health state evaluation model, and the minimum intersection of the photovoltaic predicted power, the refrigerating capacity increment boundary power value and the lithium battery available power is taken, and a compensation factor for inverter efficiency is superimposed to generate the green electricity quota available for the air conditioner, which is refreshed every cycle during the photovoltaic fluctuation period; The hot spot area and the green electricity quota available for the air conditioner are bound in real time to generate a heat-electricity coupling factor, and the heat-electricity coupling factor and the photovoltaic predicted power fluctuation intensity are combined to activate a linkage mode, triggering a deep deterministic policy gradient model to execute a cooperative decision, including: Air valve opening degree decision: through water temperature set value constraint, when the air valve opening degree in the hot spot area is increased, the water temperature set value is required to be increased synchronously according to the air supply dew point risk, and the water temperature increase amplitude and the air valve opening degree increase amplitude satisfy a preset proportional relationship; Water temperature setting decision: through lithium battery charge state constraint, when the water temperature is increased to absorb green electricity, the lithium battery is in a ready mode, including the charge state and the temperature safety range, otherwise the water temperature increase amplitude is limited; Lithium battery charging and discharging decision: through air valve action constraint, when discharging operation is performed during the photovoltaic output peak period, the air valve adjustment rate is lower than a safety threshold to prevent compressor impact; Finally, the air valve opening degree, the water temperature setting and the lithium battery charging and discharging decision are overall evaluated through a joint reward function, and only when the three control quantities are submitted synchronously, the complete reward value is calculated, and a joint control vector is output, for air conditioner end room AI optimization. The hot spot area and the green electricity quota available for the air conditioner are bound in real time to generate a heat-electricity coupling factor, and the heat-electricity coupling factor and the photovoltaic predicted power fluctuation intensity are combined to activate a linkage mode, triggering a deep deterministic policy gradient model to execute a cooperative decision, including:

2. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning according to claim 1, characterized in that: The geometric coordinates, temperature difference amplitude and heat source power of the hot spot area are respectively associated to the quota allocation weight through a space mapping matrix, wherein the high-weight hot spot area automatically obtains the priority allocation right of the green electricity quota, and a quota dynamic preemption mechanism is established, when a new hot spot is generated, the quota of the low-weight area is automatically reduced and redirected to the high-priority hot spot. The heat-electricity coupling factor and the photovoltaic power fluctuation intensity are combined to activate the linkage mode, and a double-threshold triggering mechanism is introduced, including the following steps:

3. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning according to claim 2, characterized in that: ​ When the photovoltaic power fluctuation intensity exceeds the set threshold and the thermal-electric coupling factor is in the high active interval, the wind valve-water temperature linkage mode is activated, and when the photovoltaic fluctuation continues to exceed the stability threshold, the lithium battery-water temperature buffer mode is forced to switch to, which locks the wind valve adjustment rate and enables the lithium battery smoothing strategy.

4. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning of claim 1, wherein: When the wind valve opening degree in the hot spot area is increased, the water temperature set value is required to be increased synchronously, and a preset proportional relationship is met, that is, a nonlinear mapping table of the wind valve opening degree increase and the water temperature increase amplitude is established, the mapping table is generated based on historical dewing event data training, when the wind valve opening degree is increased, the minimum water temperature increase requirement is obtained according to the current supply air humidity, and if the actual water temperature increase is insufficient, the wind valve opening degree soft amplitude limiting is triggered.

5. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning according to claim 1, characterized in that: The lithium battery in the ready mode is defined as meeting the three requirements of state of charge in the life optimal interval, battery temperature in the thermal stability window and health state higher than the decline threshold, when the system requests to increase the water temperature, if the lithium battery is in the non-ready mode, the degradation strategy is started, that is, the standby lithium battery group or the water temperature increase amplitude is limited to the preset safety baseline is called.

6. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning according to claim 1, characterized in that: The photovoltaic peak period lithium battery discharge is limited by the wind valve action constraint, and a correlation model of the wind valve adjustment rate and the compressor load change rate is established, when the wind valve adjustment rate exceeds the compressor tolerance threshold, the upper limit curve of the lithium battery discharge power is automatically generated, the curve decreases with the increase of the adjustment rate, and if the lithium battery tries to discharge over-limit, the wind valve adjustment rate forced deceleration instruction is triggered.

7. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning according to claim 6, characterized in that: The constraints of the wind valve opening degree, the water temperature set and the lithium battery charging and discharging form a closed loop verification chain, which specifically includes the following steps: The wind valve opening degree increase triggers the water temperature compensation request, the water temperature increase demand triggers the lithium battery state verification, and the lithium battery action feeds back to the wind valve rate control, and any link verification failure will start the global strategy reorganization, including reducing the wind valve adjustment target, shrinking the water temperature increase amplitude and switching the lithium battery working mode, until the dynamic balance of the three is reached again.

8. The photovoltaic lithium battery air conditioner terminal machine room AI tuning method based on digital twinning of claim 1, wherein: The overall evaluation of the joint reward function adopts a control quantity binding submission mechanism, which specifically includes the following steps: If the wind valve opening degree, the water temperature set and the lithium battery charging and discharging instruction do not arrive at the evaluator synchronously, only the local reward value is calculated and an error code is returned, when the three arrive synchronously, the instruction time stamp consistency is verified first, and then the comprehensive score of the photovoltaic consumption rate, the hot spot elimination rate, the lithium battery loss index and the dewing risk coefficient is calculated through the weighted fusion layer, and finally the joint control vector is output and the cooperation effectiveness level is marked.

Images