Artificial intelligence-based water air conditioning energy-saving speed regulation and dynamic power allocation method

Abstract

The application discloses an energy-saving speed regulation and dynamic power allocation method for water-air conditioning based on artificial intelligence, and comprises the following steps: S1, collecting system operation state data and constructing a multi-modal input; S2, using an improved DreamerV3 model to perform state prediction and action evaluation and outputting a control action; S3, optimizing a control strategy based on a constructed reward function and generating a target control parameter; S4, executing the control parameter and adjusting the equipment operation state; S5, adjusting power distribution in combination with a power grid load trend; S6, correcting the control parameter according to environmental changes and user demand; S7, identifying an abnormal state and triggering a fault-tolerant strategy; and S8, performing cooperative scheduling based on the equipment state and generating a coordinated control instruction. The application realizes energy-saving operation and intelligent regulation and control of a water-air conditioning system, and improves the coordination and energy efficiency of multiple devices.

Description

Technical Field

[0001] This invention relates to the field of air conditioning energy-saving operation technology, and in particular to an artificial intelligence-based method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioners. Background Technology

[0002] With the continuous advancement of energy conservation and environmental protection concepts, the operating efficiency of air conditioning systems, as one of the main sources of energy consumption in building operation, is receiving increasing attention. Among various air conditioning systems, water-cooled air conditioning is widely used in commercial buildings, industrial sites, and some large residential buildings due to its advantages such as high cooling efficiency and low operating costs. However, traditional water-cooled air conditioning control methods still mainly rely on fixed parameter settings or manual adjustments based on empirical rules, lacking the ability to flexibly respond to changes in actual operating conditions and the environment. This static control strategy often cannot adapt to the complex environment in which the air conditioning system operates, easily leading to problems such as insufficient temperature and humidity control accuracy, high energy consumption, and unreasonable distribution of operating load.

[0003] In recent years, research on energy-saving control of air conditioning has gradually introduced fuzzy control, PID algorithms, and model predictive control methods. These methods have improved the accuracy and efficiency of regulation to some extent, but due to their reliance on physical models and parameter settings, they still suffer from limitations such as poor adaptability and control lag for nonlinear, highly coupled water-cooled air conditioning systems. For example, model construction often struggles to accurately cover all equipment operating states and external environmental changes, leading to scenario-dependent control effects and insufficient generalization ability. Furthermore, when facing fluctuations in building power grid load, existing control methods struggle to make timely and reasonable adjustments to output power, failing to achieve a dynamic balance between energy consumption and load.

[0004] On the other hand, in actual operation, when critical equipment such as compressors, water pumps, or fans malfunction, existing systems mostly rely on manual inspections or post-incident handling, lacking timely and effective anomaly identification and automatic fault-tolerance mechanisms. This not only affects the continuous operation of the system but may also lead to energy waste or equipment damage. Regarding multi-device collaborative control, most current methods still focus on individual device control, ignoring the coupling relationship between the operating states of different devices, making it difficult to achieve coordinated optimization of power distribution and parameter scheduling for compressors, water pumps, and fans.

[0005] Therefore, how to provide an AI-based method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioners is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an artificial intelligence-based method for energy-saving speed regulation and dynamic power allocation in water-cooled air conditioning systems. This invention introduces an improved DreamerV3 reinforcement learning algorithm, and combines the operating status of the compressor, water pump, and fan in the water-cooled air conditioning system, the power grid load trend, and indoor and outdoor environmental parameters. It describes in detail the intelligent control process for energy-saving speed regulation and power allocation based on world model prediction and potential state space planning, which has the advantages of strong modeling ability, high regulation accuracy, and high efficiency of multi-device collaboration.

[0007] The energy-saving speed regulation and dynamic power allocation method for water-cooled air conditioners based on artificial intelligence according to embodiments of the present invention includes the following steps:

[0008] S1. Collect the operating status data of the water air conditioning system and construct a multimodal input vector;

[0009] S2. Input the multimodal input vector into the improved DreamerV3 reinforcement learning model, predict state transitions through the world model, generate a sequence of latent states, perform action selection and value evaluation based on the sequence of latent states, and output a sequence of control actions.

[0010] S3. Construct a reward function that includes power consumption, temperature and humidity deviation and equipment operation stability, optimize the control action sequence, and generate target control parameters for compressor, water pump and fan.

[0011] S4. Execute the target control parameters to adjust the operating frequency and output power of the compressor, water pump and fan;

[0012] S5. Based on the grid load trend information in the potential state sequence, adjust the target power of compressors, water pumps and fans, and implement power allocation;

[0013] S6. Based on the difference between changes in the indoor and outdoor environment and the user-set values, adjust the control parameters to keep the indoor temperature and humidity within the set range.

[0014] S7. Monitor the operating status of the compressor, water pump and fan, and when abnormal status is identified, invoke the fault-tolerant control strategy to adjust the operating parameters;

[0015] S8. Based on the status information of the compressor, water pump and fan, perform coordinated scheduling, generate coordinated control commands according to the joint status, and complete power allocation and operation parameter scheduling.

[0016] Optionally, the operating status data includes indoor temperature, indoor humidity, compressor power, fan speed, water pump frequency, historical power grid load sequence, current power grid load value, outdoor temperature and humidity, and user-set temperature and humidity values.

[0017] Optionally, S2 specifically includes:

[0018] S21. The multimodal input vector composed of the running state data is denoted as... Where t represents the current time step, n represents the dimension of the input vector, and the vector includes indoor temperature, indoor humidity, compressor power, fan speed, water pump frequency, historical load sequence of the power grid, current load value of the power grid, outdoor temperature and humidity, and user-set temperature and humidity values;

[0019] S22, Input the multimodal vector I t The improved DreamerV3 reinforcement learning model is input and mapped to a latent state vector through the encoding module. Where m is the dimension of the potential state;

[0020] S23, Based on the latent state vector z t Using a world model to predict state transitions, a potential state sequence {z} is generated. t+1 ,z t+2 ,…,z t+H}, where H is the number of prediction steps, z t+h Represents the potential state at time step t+h;

[0021] S24, Based on the latent state sequence {z t ,z t+1 ,…,z t+H The action generation module selects the control action for each time step to form a control action sequence {a}. t ,a t+1 ,…,a t+H},in, k represents the dimension of the control action;

[0022] S25. Evaluate the potential state sequence and its corresponding control action sequence, and determine the control action a for the current time step based on the cumulative value result. t This serves as the output of the control action sequence.

[0023] Optionally, S3 specifically includes:

[0024] S31. Set the control action sequence as {a t ,a t+1 ,…,a t+H}, where t represents the current time step and H represents the prediction step number. Let k represent the control action vector at time step t+h, where k represents the control action dimension.

[0025] S32. Construct the reward function R, including the energy consumption term R. P Temperature and humidity deviation item R E With control stationarity term R S :

[0026]

[0027] Where R represents the cumulative reward function, This represents the weighting coefficients for the three reward categories, satisfying w1+w2+w3=1;

[0028] S33, Electricity Consumption Item Represented as:

[0029]

[0030] in, This represents the energy consumption at time step t+h. Let represent the power values ​​of the compressor, fan, and water pump respectively at time step t+h. This represents the penalty coefficient for the corresponding power term;

[0031] S34, Temperature and Humidity Deviation Item Represented as:

[0032]

[0033] in, This represents the temperature and humidity deviation term at time step t+h. Indicates indoor temperature and humidity. This indicates that the user sets the temperature and humidity. The penalty coefficient representing the deviation between temperature and humidity;

[0034] S35, Controlling stationarity term Represented as:

[0035]

[0036] in, This represents the penalty value for control changes at time step t+h. This represents the control action vector between two adjacent time steps, and ||·|| represents the Euclidean norm. This represents the penalty coefficient for controlling changes;

[0037] S36. Input the reward function R into the strategy optimization module, and use a gradient-based method to optimize the control action sequence to generate compressor control parameters, water pump control parameters and fan control parameters.

[0038] Optionally, S4 specifically includes:

[0039] S41. Extract the control parameters of the current time step and assign the control parameters to compressor control parameters, water pump control parameters and fan control parameters;

[0040] S42. Based on the compressor control parameters, set the compressor operating frequency, calculate the compressor output power based on the operating frequency, and drive the compressor's frequency converter through control signals to achieve the specified operating state.

[0041] S43. Based on the pump control parameters, calculate the target flow requirement of the pump, determine the corresponding operating frequency and output power, and set the pump drive system through signal commands to achieve operating frequency adjustment.

[0042] S44. Based on the fan control parameters, generate air volume control requirements, calculate the target operating frequency and output power of the fan according to the control requirements, and send the control command to the fan speed control module to complete the frequency and power setting.

[0043] Optionally, S5 specifically includes:

[0044] S51. Extract historical load representations and load change trend variables from the potential state sequence to form a set of power grid load trend information;

[0045] S52. Input the grid load trend information and the control parameters of the current time step into the power allocation module to generate the compressor power allocation factor, water pump power allocation factor and fan power allocation factor;

[0046] S53. Based on the original target power of the compressor, water pump and fan respectively, scale them proportionally according to the corresponding power allocation factor to generate the target power of the compressor, water pump and fan, and write the adjusted power values ​​into the compressor control parameters, water pump control parameters and fan control parameters respectively.

[0047] Optionally, S6 specifically includes:

[0048] S61. Collect the indoor temperature and humidity at the current time step, obtain the temperature and humidity values ​​set by the user, and calculate the temperature difference and humidity difference;

[0049] S62. Collect the outdoor temperature and outdoor humidity at the current time step, and combine the temperature difference and humidity difference to construct an environmental deviation quantity, which is used to describe the degree of deviation between the current state and the user's expectations.

[0050] S63. Generate a control correction value based on the relationship between the environmental deviation and the control parameters, and add it to the current control parameters to obtain the corrected control parameters;

[0051] S64. Map the corrected control parameters to the operating control inputs of the compressor, water pump and fan, and update the control commands.

[0052] Optionally, S7 specifically includes:

[0053] S71. Collect operating frequency, output power, voltage, current and temperature rise data of compressor, water pump and fan, and construct the operating state vector for each time step;

[0054] S72. Input the running state vector into the anomaly detection module, combine it with the prediction model to generate the predicted state, and calculate the state residual at the current time step:

[0055]

[0056] Among them, s t Indicates the actual operating status of the data. Represents the predicted state, r t Represents the state residual;

[0057] S73. Set the residual threshold θ, when the state residual r t If the value is greater than the threshold θ, it is determined that the current state is abnormal;

[0058] S74. In abnormal conditions, select the corresponding fault-tolerant control strategy based on the equipment type and the abnormality detection results, and generate alternative control parameters according to the selected strategy:

[0059] If the compressor output power is detected to be out of limit, the compressor target power will be set to 70% of the current value.

[0060] If the pump feedback frequency is detected as missing, the pump frequency will be set to the average of the last three cycles.

[0061] If the fan current value jumps beyond the safety threshold, the fan output power is set to the stable value of the previous cycle.

[0062] If the above exception types cannot be matched, keep the control parameters unchanged;

[0063] S75. Write the alternative control parameters into the control command channels of the compressor, water pump and fan respectively, to replace the original operating parameters.

[0064] Optionally, S8 specifically includes:

[0065] S81. Collect the operating status of the compressor, water pump, and fan at the current time step, construct the compressor state vector, water pump state vector, and fan state vector respectively, and concatenate them in sequence into a joint state vector. in, Represents the compressor state vector. Represents the pump state vector. S represents the wind turbine state vector. t Represents the joint state vector;

[0066] S82. Based on the joint state vector and the preset joint control strategy, calculate the compressor target power, water pump operating frequency and fan output power in a unified manner to form a coordinated control command.

[0067] S83. Assign the parameters in the coordinated control command to the control parameter structures of the compressor, water pump and fan respectively, and update the operating frequency and output power;

[0068] S84. Write the updated control parameters into the control input channel of the corresponding device to complete the operation parameter scheduling and power allocation.

[0069] The beneficial effects of this invention are:

[0070] This invention, by introducing an improved DreamerV3 reinforcement learning algorithm, breaks through the limitations of traditional water-cooled air conditioning systems that rely on static control rules and fixed model parameters, achieving dynamic modeling and control of complex nonlinear operating conditions and multi-device coupled states. By utilizing a world model mechanism to predict the operating states of the compressor, water pump, and fan, a potential state space is constructed, freeing the control strategy from dependence on precise physical modeling, thereby significantly improving the control system's adaptability to changes in the external environment.

[0071] This invention further integrates grid load variation trends, performing action planning and value assessment within the potential state space. It can dynamically adjust the operating frequency and output power of equipment based on load fluctuations, coordinating power allocation among the three types of equipment to achieve overall optimization of energy distribution. Furthermore, by constructing a reward function system, it comprehensively weighs power consumption, temperature and humidity deviations, and equipment operational stability, making the generated control strategy more aligned with energy conservation and comfort objectives.

[0072] In terms of anomaly identification and fault handling, this invention designs an anomaly detection mechanism based on state residual analysis and matches it with a preset fault-tolerant strategy to correct control parameters in real time, ensuring the system's continuous control capability under abnormal operating conditions. Through a collaborative scheduling method driven by multi-device joint state vectors, the compressor, water pump, and fan can collaboratively execute control commands according to their respective states, solving the problems of power distribution imbalance and mutual interference in traditional air conditioning systems, and improving the overall stability and efficiency of the system. Attached Figure Description

[0073] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0074] Figure 1 The flowchart shows the energy-saving speed regulation and dynamic power allocation method for water-cooled air conditioners based on artificial intelligence proposed in this invention.

[0075] Figure 2 This is a schematic diagram of the improved DreamerV3 reinforcement learning structure in the AI-based water air conditioning energy-saving speed regulation and dynamic power allocation method proposed in this invention.

[0076] Figure 3 This is a diagram showing the collaborative scheduling control logic structure of the compressor, water pump, and fan in the AI-based energy-saving speed regulation and dynamic power allocation method for water-cooled air conditioning proposed in this invention. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0078] refer to Figure 1-3 An AI-based method for energy-saving speed regulation and dynamic power allocation in water-cooled air conditioning systems includes the following steps:

[0079] S1. Collect the operating status data of the water air conditioning system and construct a multimodal input vector;

[0080] S2. Input the multimodal input vector into the improved DreamerV3 reinforcement learning model, predict state transitions through the world model, generate a sequence of latent states, perform action selection and value evaluation based on the sequence of latent states, and output a sequence of control actions.

[0081] S3. Construct a reward function that includes power consumption, temperature and humidity deviation and equipment operation stability, optimize the control action sequence, and generate target control parameters for compressor, water pump and fan.

[0082] S4. Execute the target control parameters to adjust the operating frequency and output power of the compressor, water pump and fan;

[0083] S5. Based on the grid load trend information in the potential state sequence, adjust the target power of compressors, water pumps and fans, and implement power allocation;

[0084] S6. Based on the difference between changes in the indoor and outdoor environment and the user-set values, adjust the control parameters to keep the indoor temperature and humidity within the set range.

[0085] S7. Monitor the operating status of the compressor, water pump and fan, and when abnormal status is identified, invoke the fault-tolerant control strategy to adjust the operating parameters;

[0086] S8. Based on the status information of the compressor, water pump and fan, perform coordinated scheduling, generate coordinated control commands according to the joint status, and complete power allocation and operation parameter scheduling.

[0087] This invention constructs a water-cooled air conditioning energy-saving speed regulation and dynamic power allocation method based on a reinforcement learning framework. This method enables continuous perception and feedback adjustment of the system's operating status, automatically completing frequency adjustment, power distribution, and comfort control under various environmental and load conditions. Applicable to complex air conditioning control scenarios, this method ensures system stability and energy efficiency even under multi-objective conditions, demonstrating high versatility and engineering potential.

[0088] In this embodiment, the operating status data includes indoor temperature, indoor humidity, compressor power, fan speed, water pump frequency, historical power grid load sequence, current power grid load value, outdoor temperature and humidity, and user-set temperature and humidity values.

[0089] This invention explicitly incorporates indoor and outdoor environmental parameters, electrical equipment status information, historical and real-time power grid load data, and user requirements into a unified input variable system, constructing a high-dimensional multimodal input vector. This effectively enhances the system's overall perception capability of the operating environment and task requirements. This multimodal perception structure improves the accuracy and adaptability of subsequent world model prediction and control action generation, providing a data foundation for robust control under complex operating conditions.

[0090] In this embodiment, S2 specifically includes:

[0091] S21. The multimodal input vector composed of the running state data is denoted as... Where t represents the current time step, n represents the dimension of the input vector, and the vector includes indoor temperature, indoor humidity, compressor power, fan speed, water pump frequency, historical load sequence of the power grid, current load value of the power grid, outdoor temperature and humidity, and user-set temperature and humidity values;

[0092] S22, Input the multimodal vector I t The improved DreamerV3 reinforcement learning model is input and mapped to a latent state vector through the encoding module. Where m is the dimension of the potential state;

[0093] S23, Based on the latent state vector z t Using a world model to predict state transitions, a potential state sequence {z} is generated. t+1 ,z t+2 ,…,z t+H}, where H is the number of prediction steps, z t+h Represents the potential state at time step t+h;

[0094] S24, Based on the latent state sequence {z t ,z t+1 ,…,z t+H The action generation module selects the control action for each time step to form a control action sequence {a}.t ,a t+1 ,…,a t+H},in, k represents the dimension of the control action;

[0095] S25. Evaluate the potential state sequence and its corresponding control action sequence, and determine the control action a for the current time step based on the cumulative value result. t This serves as the output of the control action sequence.

[0096] The improved DreamerV3 reinforcement learning model used in this invention achieves multi-step state prediction and action value evaluation in the latent state space. It can capture the dynamic change trend of the system and make forward-looking control decisions. Through the structural representation of the latent state sequence and the optimization of reinforcement learning strategies, this method improves the stability and computational efficiency of control actions and reduces misjudgment and frequent adjustment phenomena in long-term operation.

[0097] In this embodiment, S3 specifically includes:

[0098] S31. Set the control action sequence as {a t ,a t+1 ,…,a t+H}, where t represents the current time step and H represents the prediction step number. Let k represent the control action vector at time step t+h, where k represents the control action dimension.

[0099] S32. Construct the reward function R, including the energy consumption term R. P Temperature and humidity deviation item R E With control stationarity term R S :

[0100]

[0101] Where R represents the cumulative reward function, This represents the weighting coefficients for the three reward categories, satisfying w1+w2+w3=1;

[0102] S33, Electricity Consumption Item Represented as:

[0103]

[0104] in, This represents the energy consumption at time step t+h. Let represent the power values ​​of the compressor, fan, and water pump respectively at time step t+h. This represents the penalty coefficient for the corresponding power term;

[0105] S34, Temperature and Humidity Deviation Item Represented as:

[0106]

[0107] in, This represents the temperature and humidity deviation term at time step t+h. Indicates indoor temperature and humidity. This indicates that the user sets the temperature and humidity. The penalty coefficient representing the deviation between temperature and humidity;

[0108] S35, Controlling stationarity term Represented as:

[0109]

[0110] in, This represents the penalty value for control changes at time step t+h. This represents the control action vector between two adjacent time steps, and ||·|| represents the Euclidean norm. This represents the penalty coefficient for controlling changes;

[0111] S36. Input the reward function R into the strategy optimization module, and use a gradient-based method to optimize the control action sequence to generate compressor control parameters, water pump control parameters and fan control parameters.

[0112] This invention maps the optimization results of action strategies into specific equipment control parameter outputs, clearly defining the target control frequency and power configuration of key equipment such as compressors, water pumps, and fans. The control parameter structure is clear and the boundaries are reasonable, which can accurately guide the execution layer to complete the target operating state and effectively avoid equipment load fluctuations and system efficiency decline caused by strategy ambiguity or instruction conflict.

[0113] In this embodiment, S4 specifically includes:

[0114] S41. Extract the control parameters of the current time step and assign the control parameters to compressor control parameters, water pump control parameters and fan control parameters;

[0115] S42. Based on the compressor control parameters, set the compressor operating frequency, calculate the compressor output power based on the operating frequency, and drive the compressor's frequency converter through control signals to achieve the specified operating state.

[0116] S43. Based on the pump control parameters, calculate the target flow requirement of the pump, determine the corresponding operating frequency and output power, and set the pump drive system through signal commands to achieve operating frequency adjustment.

[0117] S44. Based on the fan control parameters, generate air volume control requirements, calculate the target operating frequency and output power of the fan according to the control requirements, and send the control command to the fan speed control module to complete the frequency and power setting.

[0118] This invention introduces grid load trend information into dynamic power allocation and establishes a target power correction mechanism driven by prediction. This enables the air conditioning system to proactively respond to load reduction demands during peak power periods, avoiding local overload or energy waste. This mechanism improves the air conditioning system's adaptability to grid fluctuations and promotes the coordinated matching between energy consumption behavior and dynamic power supply and demand.

[0119] In this embodiment, S5 specifically includes:

[0120] S51. Extract historical load representations and load change trend variables from the potential state sequence to form a set of power grid load trend information;

[0121] S52. Input the grid load trend information and the control parameters of the current time step into the power allocation module to generate the compressor power allocation factor, water pump power allocation factor and fan power allocation factor;

[0122] S53. Based on the original target power of the compressor, water pump and fan respectively, scale them proportionally according to the corresponding power allocation factor to generate the target power of the compressor, water pump and fan, and write the adjusted power values ​​into the compressor control parameters, water pump control parameters and fan control parameters respectively.

[0123] This invention combines real-time changes in indoor and outdoor temperature and humidity with differences in user-set values ​​to construct a dynamic feedback control path. It updates control variables in the potential state space to achieve real-time and precise control of temperature and humidity. This correction mechanism enables the control strategy to maintain good responsiveness under different climate and load change scenarios, thereby ensuring the long-term stable achievement of comfort control goals.

[0124] In this embodiment, S6 specifically includes:

[0125] S61. Collect the indoor temperature and humidity at the current time step, obtain the temperature and humidity values ​​set by the user, and calculate the temperature difference and humidity difference;

[0126] S62. Collect the outdoor temperature and outdoor humidity at the current time step, and combine the temperature difference and humidity difference to construct an environmental deviation quantity, which is used to describe the degree of deviation between the current state and the user's expectations.

[0127] S63. Generate a control correction value based on the relationship between the environmental deviation and the control parameters, and add it to the current control parameters to obtain the corrected control parameters;

[0128] S64. Map the corrected control parameters to the operating control inputs of the compressor, water pump and fan, and update the control commands.

[0129] This invention introduces an abnormal state monitoring mechanism. Through residual detection of operating status and multi-channel monitoring mechanism, it can quickly identify abnormal states of compressor, water pump and fan during operation, and link fault-tolerant control strategy to generate backup parameters or activate safety strategy to ensure that the air conditioning system can still operate stably in the event of interference, signal interruption or abnormal parameters, thereby reducing the risk of equipment damage and energy efficiency reduction.

[0130] In this embodiment, S7 specifically includes:

[0131] S71. Collect operating frequency, output power, voltage, current and temperature rise data of compressor, water pump and fan, and construct the operating state vector for each time step;

[0132] S72. Input the running state vector into the anomaly detection module, combine it with the prediction model to generate the predicted state, and calculate the state residual at the current time step:

[0133]

[0134] Among them, s t Indicates the actual operating status of the data. Represents the predicted state, r t Represents the state residual;

[0135] S73. Set the residual threshold θ, when the state residual r t If the value is greater than the threshold θ, it is determined that the current state is abnormal;

[0136] S74. In abnormal conditions, select the corresponding fault-tolerant control strategy based on the equipment type and the abnormality detection results, and generate alternative control parameters according to the selected strategy:

[0137] If the compressor output power is detected to be out of limit, the compressor target power will be set to 70% of the current value.

[0138] If the pump feedback frequency is detected as missing, the pump frequency will be set to the average of the last three cycles.

[0139] If the fan current value jumps beyond the safety threshold, the fan output power is set to the stable value of the previous cycle.

[0140] If the above exception types cannot be matched, keep the control parameters unchanged;

[0141] S75. Write the alternative control parameters into the control command channels of the compressor, water pump and fan respectively, to replace the original operating parameters.

[0142] This invention constructs a joint state vector by fusing the operating status information of compressors, water pumps, and fans, and generates coordinated control commands in a reinforcement learning decision structure. This enables the operating status of each device to be uniformly adjusted based on the overall system operating conditions. The coordinated scheduling strategy ensures the reasonable allocation and dynamic optimization of energy among different devices, effectively alleviating energy waste and system instability caused by operational conflicts between devices.

[0143] In this embodiment, S8 specifically includes:

[0144] S81. Collect the operating status of the compressor, water pump, and fan at the current time step, construct the compressor state vector, water pump state vector, and fan state vector respectively, and concatenate them in sequence into a joint state vector. in, Represents the compressor state vector. Represents the pump state vector. S represents the wind turbine state vector. t Represents the joint state vector;

[0145] S82. Based on the joint state vector and the preset joint control strategy, calculate the compressor target power, water pump operating frequency and fan output power in a unified manner to form a coordinated control command.

[0146] S83. Assign the parameters in the coordinated control command to the control parameter structures of the compressor, water pump and fan respectively, and update the operating frequency and output power;

[0147] S84. Write the updated control parameters into the control input channel of the corresponding device to complete the operation parameter scheduling and power allocation.

[0148] This invention designs a joint state mapping and instruction writing mechanism in collaborative control. Control instructions are uniformly generated and distributed based on the joint state of multiple devices, achieving synchronous coordination of the operating parameters of compressors, water pumps, and fans. This mechanism reduces communication conflicts and execution deviations during multi-device control, enhances the overall consistency and collaborative efficiency of the control system, and is suitable for centralized intelligent management and control of medium and large-scale building air conditioning systems.

[0149] Example 1:

[0150] To verify the feasibility of this invention in practice, it was applied to the central air conditioning system of a typical five-story office building in Changzhou, Jiangsu Province. The building has a floor area of ​​approximately 9,600 square meters and accommodates about 280 people daily. Its internal structure includes offices, meeting rooms, open-plan office areas, a lounge area, and two medium-sized server rooms. The existing air conditioning system used water-cooled units equipped with a 45kW inverter compressor, an 18kW inverter water pump, and two parallel centrifugal fans. The control methods primarily involved preset operating time periods and manual adjustment at fixed frequencies. The system's response was sluggish, resulting in low energy efficiency. Especially during the summer months, when climate changes are frequent and electricity prices fluctuate wildly, the air conditioning load was difficult to adjust adaptively, leading to high operating costs.

[0151] This invention's implementation deploys an AI-based water-cooled air conditioning energy-saving speed regulation and dynamic power allocation system within the building. The hardware architecture comprises an edge computing server, a PLC controller, temperature and humidity sensors, a current / voltage acquisition module, and a wireless communication module, with an improved DreamerV3 reinforcement learning algorithm embedded as the core decision-making module. Every 5 minutes, the system collects indoor and outdoor environmental data, compressor power, fan speed, water pump frequency, historical and trend data of the power grid load, and user-defined target temperature and humidity, constructing a multimodal input vector that is input to the intelligent control model. The model uses a world model mechanism to predict state transitions and performs action selection and value evaluation in the potential state space, generating a sequence of control actions. The strategy optimization module further constructs a reward function based on energy consumption, environmental errors, and equipment stability, outputting target control parameters for the compressor, water pump, and fan. These parameters are sent to each equipment execution unit in real time, forming a closed-loop regulation.

[0152] In actual operation, the system introduces a power grid load forecasting mechanism to identify peak electricity demand in advance. In the power allocation logic, it dynamically reduces the compressor load while simultaneously increasing the output of the water pump and fan, enabling the air conditioning system to meet temperature and humidity control requirements while avoiding peak load pressure. Furthermore, the system monitors the operating status of the compressor and fan, and can quickly switch to a fault-tolerant control strategy in case of abnormal temperature rise, voltage fluctuations, or sensor failure, avoiding system malfunctions or human intervention.

[0153] The operation test in this embodiment was conducted from July 20th to August 20th, 2024. The air conditioner operated daily from 6:00 AM to 8:00 PM, covering 32 days including weekdays and weekends. The ambient temperature ranged from 29.5℃ to 38.1℃, and the outdoor relative humidity ranged from 42% to 82%. Data from 10 consecutive days were selected for quantitative comparison. The specific energy-saving effects are shown in Table 1.

[0154] Table 1 Comparison of Energy-Saving Effects of Air Conditioning Systems

[0155]

[0156]

[0157] The data in the table show that, under similar operating loads and external weather conditions, the air conditioning system using the method of this invention achieves an average daily energy saving rate of 15.1%, with the most outstanding energy saving performance during peak electricity price periods (13:00-15:30), reaching a peak power reduction of 18.7% in some periods. The problems of large fluctuations and lag in temperature and humidity control in the original system have also been effectively solved. The indoor temperature control error has decreased from ±1.8℃ to ±0.7℃, and the relative humidity control accuracy has improved from ±7% to ±4%, significantly enhancing comfort in all areas.

[0158] Throughout the testing period, the system identified and handled four abnormal events, including two sensor signal failures, one abnormal compressor temperature rise, and one instantaneous voltage drop in the power grid. In all cases, the system automatically entered fault-tolerant control mode within 60 seconds, maintaining equipment stability through parameter rollback or limited operation. No personnel intervention was required during the test, and no air conditioning system interruptions or alarms occurred. Monitoring records showed a 21% reduction in equipment start-up frequency, a 27% improvement in power regulation stability, and a 14% decrease in fan vibration. In the long term, this will effectively reduce equipment maintenance frequency and component wear.

[0159] In summary, this embodiment fully verifies the adaptability and engineering feasibility of the present invention in complex building environments. It can realize energy-saving operation, load prediction and dynamic power allocation of air conditioning systems, and solve the problems of untimely response, high energy consumption and weak fault tolerance mechanism of traditional control strategies. It has high stability and practical promotion value.

[0160] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioners based on artificial intelligence, characterized in that: Includes the following steps: S1. Collect the operating status data of the water air conditioning system and construct a multimodal input vector; S2. Input the multimodal input vector into the improved DreamerV3 reinforcement learning model, predict state transitions through the world model, generate a sequence of latent states, perform action selection and value evaluation based on the sequence of latent states, and output a sequence of control actions. S3. Construct a reward function that includes power consumption, temperature and humidity deviation and equipment operation stability, optimize the control action sequence, and generate target control parameters for compressor, water pump and fan. S4. Execute the target control parameters to adjust the operating frequency and output power of the compressor, water pump and fan; S5. Based on the grid load trend information in the potential state sequence, adjust the target power of compressors, water pumps and fans, and implement power allocation; S6. Based on the difference between changes in the indoor and outdoor environment and the user-set values, adjust the control parameters to keep the indoor temperature and humidity within the set range. S7. Monitor the operating status of the compressor, water pump and fan, and when abnormal status is identified, invoke the fault-tolerant control strategy to adjust the operating parameters; S8. Based on the status information of the compressor, water pump and fan, perform coordinated scheduling, generate coordinated control commands according to the joint status, and complete power allocation and operation parameter scheduling.

2. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, The operating status data includes indoor temperature, indoor humidity, compressor power, fan speed, water pump frequency, historical power grid load sequence, current power grid load value, outdoor temperature and humidity, and user-set temperature and humidity values.

3. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, S2 specifically includes: S21. The multimodal input vector composed of the running state data is denoted as... Where t represents the current time step, n represents the dimension of the input vector, and the vector includes indoor temperature, indoor humidity, compressor power, fan speed, water pump frequency, historical load sequence of the power grid, current load value of the power grid, outdoor temperature and humidity, and user-set temperature and humidity values; S22, Input the multimodal vector I t The improved DreamerV3 reinforcement learning model is input and mapped to a latent state vector through the encoding module. Where m is the dimension of the potential state; S23, Based on the latent state vector z t Using a world model to predict state transitions, a potential state sequence {z} is generated. t+1 ,z t+2 ,…,z t+H }, where H is the number of prediction steps, z t+h Represents the potential state at time step t+h; S24, Based on the latent state sequence {z t ,z t+1 ,…,z t+H The action generation module selects the control action for each time step to form a control action sequence {a}. t ,a t+1 ,…,a t+H },in, k represents the dimension of the control action; S25. Evaluate the potential state sequence and its corresponding control action sequence, and determine the control action a for the current time step based on the cumulative value result. t This serves as the output of the control action sequence.

4. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, S3 specifically includes: S31. Set the control action sequence as {a t ,a t+1 ,…,a t+H }, where t represents the current time step and H represents the prediction step number. Let k represent the control action vector at time step t+h, where k represents the control action dimension. S32. Construct the reward function R, including the energy consumption term R. P Temperature and humidity deviation item R E With control stationarity term R S : Where R represents the cumulative reward function, This represents the weighting coefficients for the three reward categories, satisfying w1+w2+w3=1; S33, Electricity Consumption Item Represented as: in, This represents the energy consumption at time step t+h. Let represent the power values ​​of the compressor, fan, and water pump respectively at time step t+h. This represents the penalty coefficient for the corresponding power term; S34, Temperature and Humidity Deviation Item Represented as: in, This represents the temperature and humidity deviation term at time step t+h. Indicates indoor temperature and humidity, T set , This indicates that the user sets the temperature and humidity. The penalty coefficient representing the deviation between temperature and humidity; S35, Controlling stationarity term Represented as: in, This represents the penalty value for control changes at time step t+h. This represents the control action vector between two adjacent time steps, and ||·|| represents the Euclidean norm. This represents the penalty coefficient for controlling changes; S36. Input the reward function R into the strategy optimization module, and use a gradient-based method to optimize the control action sequence to generate compressor control parameters, water pump control parameters and fan control parameters.

5. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, S4 specifically includes: S41. Extract the control parameters of the current time step and assign the control parameters to compressor control parameters, water pump control parameters and fan control parameters; S42. Based on the compressor control parameters, set the compressor operating frequency, calculate the compressor output power based on the operating frequency, and drive the compressor's frequency converter through control signals to achieve the specified operating state. S43. Based on the pump control parameters, calculate the target flow requirement of the pump, determine the corresponding operating frequency and output power, and set the pump drive system through signal commands to achieve operating frequency adjustment. S44. Based on the fan control parameters, generate air volume control requirements, calculate the target operating frequency and output power of the fan according to the control requirements, and send the control command to the fan speed control module to complete the frequency and power setting.

6. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, S5 specifically includes: S51. Extract historical load representations and load change trend variables from the potential state sequence to form a set of power grid load trend information; S52. Input the grid load trend information and the control parameters of the current time step into the power allocation module to generate the compressor power allocation factor, water pump power allocation factor and fan power allocation factor; S53. Based on the original target power of the compressor, water pump and fan respectively, scale them proportionally according to the corresponding power allocation factor to generate the target power of the compressor, water pump and fan, and write the adjusted power values ​​into the compressor control parameters, water pump control parameters and fan control parameters respectively.

7. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, S6 specifically includes: S61. Collect the indoor temperature and humidity at the current time step, obtain the temperature and humidity values ​​set by the user, and calculate the temperature difference and humidity difference; S62. Collect the outdoor temperature and outdoor humidity at the current time step, and combine the temperature difference and humidity difference to construct an environmental deviation quantity, which is used to describe the degree of deviation between the current state and the user's expectations. S63. Generate a control correction value based on the relationship between the environmental deviation and the control parameters, and add it to the current control parameters to obtain the corrected control parameters; S64. Map the corrected control parameters to the operating control inputs of the compressor, water pump and fan, and update the control commands.

8. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, Specifically, S7 includes: S71. Collect operating frequency, output power, voltage, current and temperature rise data of compressor, water pump and fan, and construct the operating state vector for each time step; S72. Input the running state vector into the anomaly detection module, combine it with the prediction model to generate the predicted state, and calculate the state residual r at the current time step. t ; S73. Set the residual threshold θ, when the state residual r t If the value is greater than the threshold θ, it is determined that the current state is abnormal; S74. In abnormal conditions, select the corresponding fault-tolerant control strategy based on the equipment type and the abnormality detection results, and generate alternative control parameters according to the selected strategy: If the compressor output power is detected to be out of limit, the compressor target power will be set to 70% of the current value. If the pump feedback frequency is detected as missing, the pump frequency will be set to the average of the last three cycles. If the fan current value jumps beyond the safety threshold, the fan output power is set to the stable value of the previous cycle. If the above exception types cannot be matched, keep the control parameters unchanged; S75. Write the alternative control parameters into the control command channels of the compressor, water pump and fan respectively, to replace the original operating parameters.

9. The method for energy-saving speed regulation and dynamic power allocation of water-cooled air conditioning based on artificial intelligence according to claim 1, characterized in that, S8 specifically includes: S81. Collect the operating status of the compressor, water pump, and fan at the current time step, construct the compressor state vector, water pump state vector, and fan state vector respectively, and concatenate them in sequence into a joint state vector. in, Represents the compressor state vector. Represents the pump state vector. S represents the wind turbine state vector. t Represents the joint state vector; S82. Based on the joint state vector and the preset joint control strategy, calculate the compressor target power, water pump operating frequency and fan output power in a unified manner to form a coordinated control command. S83. Assign the parameters in the coordinated control command to the control parameter structures of the compressor, water pump and fan respectively, and update the operating frequency and output power; S84. Write the updated control parameters into the control input channel of the corresponding device to complete the operation parameter scheduling and power allocation.

Images