Air conditioner fresh air control method and system of comprehensive novel energy-saving building

Abstract

The invention discloses an air conditioner fresh air control method and system for a comprehensive novel energy-saving building, and relates to the technical field of intelligent control, and the method comprises the following steps: collecting sensor data in real time, carrying out wave noise reduction processing by using Kalman filtering, and constructing a standardized data packet containing a timestamp, an equipment ID and a numerical value type in combination with smoothing processing; dynamically adjusting energy scheduling by using a block chain method in combination with real-time electricity price, calculating an optimal power purchasing scheme, and generating an energy scheduling instruction; dividing a building plane by adopting a bee colony algorithm, establishing a dynamic air volume distribution matrix, outputting an air volume control instruction and marking areas which do not meet requirements; and when the continuous unsatisfied demand area and meteorological sudden change are detected, simulating an extreme working condition through the generative adversarial network, generating an emergency control instruction set, and adjusting the rotating speed of a fan and the opening degree of an air valve. Through a global feature vector updating mechanism of federal learning, the method can adapt to building layout changes in real time, and the overall energy efficiency ratio is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control, and particularly to an air-conditioning fresh air control method and system for a comprehensive new energy-saving building. Background Art

[0002] At present, there are many deficiencies in the existing air-conditioning fresh air control systems. In the case of data noise interference, the existing systems rely on off-line trained models or preset rules, resulting in response lags and local optimization failures, and unable to achieve the dynamic balance between energy efficiency and comfort. Moreover, the existing technologies still lack the ability of adaptive adjustment for abnormal working conditions. When sudden environmental changes occur, the system often cannot quickly generate optimal control instructions, resulting in energy waste or substandard local comfort.

[0003] In the control technology of the current intelligent building air-conditioning fresh air system, a very big problem faced is the lack of dynamic response ability. Many traditional methods adopt static air volume distribution strategies or control algorithms based on fixed thresholds, showing obvious limitations in solving the highly dynamic changes of the building environment. Specifically, the multi-dimensional data such as temperature, humidity, CO 2 concentration, and particulate matter collected by the sensor network have great spatio-temporal heterogeneity. At the same time, the data processing methods often adopt simple moving average or low-pass filtering methods, which cannot efficiently eliminate the deviation caused by instantaneous interference signals. The response delay and adjustment accuracy decline of the control system caused by these problems are very obvious and are also important problems that need to be solved urgently. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an air-conditioning fresh air control method for a comprehensive new energy-saving building to solve the problems of control lag, energy waste, and response delay in extreme working conditions of the air-conditioning fresh air system caused by insufficient real-time fusion accuracy of multi-source heterogeneous data.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an air-conditioning fresh air control method for a comprehensive new energy-saving building, which includes: collecting sensor data in real time, performing wave noise reduction processing using Kalman filtering, and combining smoothing processing to construct a standardized data packet including a timestamp, a device ID, and a numerical type;

[0008] Using the blockchain method to dynamically adjust energy scheduling in combination with real-time electricity prices, calculating an optimal electricity procurement plan, and generating an energy scheduling instruction;

[0009] Using the bee colony algorithm, divide the building floor plan, establish a dynamic air volume distribution matrix, output air volume control instructions and mark the areas where the demand is not met;

[0010] When continuously unmet demand areas and meteorological mutations are detected, simulate extreme working conditions through a generative adversarial network and generate an emergency control instruction set to adjust the fan speed and damper opening;

[0011] Use federated learning to optimize global control and update the decisions of the bee colony algorithm and the generative adversarial network in real time.

[0012] As a preferred solution of the air supply control method for the integrated new energy-saving building described in the present invention, wherein: the sensor data includes indoor data and outdoor data;

[0013] Perform wave noise reduction processing using Kalman filtering, combined with smoothing processing, to construct a standardized data packet including a timestamp, a device ID, and a numerical type. The specific steps are as follows.

[0014] Dynamically process indoor data and outdoor data using Kalman filtering and output a stable data sequence after noise reduction;

[0015] Use a trusted data packet to encapsulate and integrate the timestamp, device ID, and numerical type of the data, and output a standardized data packet.

[0016] As a preferred solution of the air supply control method for the integrated new energy-saving building described in the present invention, wherein: use the blockchain method to dynamically adjust energy scheduling in combination with real-time electricity prices, calculate the optimal electricity procurement plan, and generate an energy scheduling instruction. The specific steps are as follows.

[0017] Verify the data integrity of the standard data packet through a smart contract and output a verified real-time status data packet;

[0018] Use stochastic differential game for dynamic optimization to calculate the optimal strategy plan;

[0019] Combine with the smart contract to trigger automatic execution, generate a scheduling instruction structure body, and output the execution status hash and energy scheduling instruction stored on the chain for certification.

[0020] As a preferred solution of the air supply control method for the integrated new energy-saving building described in the present invention, wherein: use the bee colony algorithm to divide the building floor plan, establish a dynamic air volume distribution matrix, output air volume control instructions and mark the areas where the demand is not met. The specific steps are as follows.

[0021] Dynamically divide the building floor plan, determine the air volume demand weights of different areas, and output an initial air volume demand matrix with weights;

[0022] Optimize by combining with the bee colony algorithm, construct a fitness function for air volume constraints, and output the optimal air volume distribution plan;

[0023] Combine the optimal air volume with real-time standard data, calculate the residual value using a sliding window, mark the abnormal areas that do not meet the requirements, and output the final air volume control instruction and the coordinates of the abnormal areas.

[0024] As a preferred solution of the air-conditioning fresh air control method for the comprehensive new energy-saving building described in the present invention, wherein: when continuously unsatisfied demand areas and meteorological mutations are detected, simulate extreme working conditions through a generative adversarial network and generate an emergency control instruction set to adjust the fan speed and the opening degree of the air valve. The specific steps are as follows.

[0025] Combine the abnormal coordinates with the meteorological mutation index for abnormal fusion detection, and output an abnormal event report form;

[0026] Generate a candidate instruction set through a generative adversarial network, and combine the upper and lower limits of the built-in fan speed and the air valve coupling constraint to output a control instruction;

[0027] Optimize the control instruction using reinforcement learning to adjust the fan speed and the opening degree of the air valve.

[0028] As a preferred solution of the air-conditioning fresh air control method for the comprehensive new energy-saving building described in the present invention, wherein: use federated learning to optimize global control, and update the decisions of the bee colony algorithm and the generative adversarial network in real time. The specific steps are as follows.

[0029] Combine feature-level federated learning with standardized data packets to output a global feature vector;

[0030] Use the decision tree algorithm to automatically generate control rules and output a control strategy table;

[0031] Use an adaptive adjustment algorithm combined with dynamic weight adjustment to update the decision in real time.

[0032] In a second aspect, the present invention provides an air-conditioning fresh air control system for a comprehensive new energy-saving building, including a data processing module, an energy scheduling module, an air volume distribution module, an emergency simulation module, and a global optimization module;

[0033] The data processing module is used to collect sensor data in real time, perform wave noise reduction processing using Kalman filtering, and combine with smoothing processing to construct a standardized data packet including a timestamp, a device ID, and a numerical type;

[0034] The energy scheduling module is used to dynamically adjust energy scheduling using the blockchain method combined with real-time electricity prices, calculate the optimal electricity procurement plan, and generate an energy scheduling instruction;

[0035] The air volume distribution module is used to divide the building plane by using the bee colony algorithm, establish a dynamic air volume distribution matrix, output the optimal air volume distribution plan and mark the areas where the demand is not met;

[0036] The emergency simulation module is used to simulate extreme working conditions and generate an emergency control instruction set through the generative adversarial network when it detects continuously unmet demand areas and meteorological mutations, and adjust the fan speed and damper opening;

[0037] The global optimization module is used to optimize the global control by using federated learning and update the decisions of the bee colony algorithm and the generative adversarial network in real time.

[0038] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the air-conditioning fresh air control method for the comprehensive new energy-saving building as described in the first aspect of the present invention is implemented.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the air-conditioning fresh air control method for the comprehensive new energy-saving building as described in the first aspect of the present invention is implemented.

[0040] The beneficial effects of the present invention are as follows: Through the collaborative application of the Kalman filter and the blockchain method, the signal-to-noise ratio of the sensor data is improved, and the influence of instantaneous interference on the control instructions is effectively eliminated. The time stamp and device ID binding mechanism of the standardized data packet greatly reduces the time alignment error control of multi-source data. At the energy scheduling level, based on the real-time electricity price response algorithm of stochastic differential game and combined with the automatic execution characteristics of the blockchain smart contract, the fluctuation range of the electricity purchase cost is compressed. The global feature vector update mechanism of federated learning enables the system to adapt to the building layout change in real time and improves the overall energy efficiency ratio. Description of the Drawings

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0042] Figure 1 It is a flowchart of the air-conditioning fresh air control method for the comprehensive new energy-saving building in Embodiment 1.

[0043] Figure 2 It is a schematic diagram of the air-conditioning fresh air control module for the comprehensive new energy-saving building in Embodiment 1.

[0044] Figure 3 It is the data preprocessing flowchart in Embodiment 1.

[0045] Figure 4 It is the exception handling flowchart in Embodiment 1. Specific implementation manners

[0046] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific implementation manners of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0047] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0048] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments.

[0049] Embodiment 1, referring to Figures 1 to 4 , this embodiment provides a fresh air control method for a comprehensive new energy-saving building, including the following steps:

[0050] S1. Collect sensor data in real time, perform wave noise reduction processing using Kalman filtering, and combine with smoothing processing to construct a standardized data packet including a timestamp, a device ID, and a numerical type.

[0051] Furthermore, use Kalman filtering to dynamically process data and output a stable data sequence after noise reduction;

[0052] Specifically, use Kalman filtering to process the received real-time data, so that the filtering algorithm starts running from a reasonable starting point; according to the state estimate at the previous moment, predict the state value and covariance matrix at the current moment to reduce the interference of sudden noise; combine with the current sensor observation value, calculate the Kalman gain, correct the predicted state, obtain the optimal estimate value, and update the covariance matrix, and output the optimal estimate state vector and the updated covariance matrix; use the current optimal estimate as the initial value for the next moment, and execute in a loop to output a stable data sequence after noise reduction.

[0053] Use a trusted data packet to encapsulate and integrate the timestamp, device ID, and numerical type of the data, and output a standardized data packet;

[0054] Specifically, timestamps, device IDs, and value types are extracted from the denoised stable data sequence, and a set of structured data fields is output to unify the data format and reduce parsing errors caused by structural differences in multi-source data. A hash value is calculated for the structured data to generate a digital signature, ensuring that the data has not been tampered with during transmission. A data block with the hash value and signature is output to enhance data credibility and prevent malicious tampering or transmission errors. The verified data fields, hash value, and signature are encapsulated into a standardized data packet according to the protocol, and a trusted data packet that meets the standard is output. The data packet supports efficient cross-platform and cross-system data exchange, facilitating subsequent blockchain storage or federated learning calls. The trusted data packet is stored in a distributed database, and a fast retrieval index based on timestamps and device IDs is established, outputting a standardized data packet.

[0055] Preferably, Kalman filter denoising provides highly reliable input for subsequent control algorithms, reduces misjudgments caused by sensor noise, dynamically adapts to environmental changes, and is more robust than the fixed threshold method. The trusted data packet encapsulates a standardized data format, simplifies multi-system integration, and the hash signature mechanism ensures the privacy and security of data in federated learning, preventing malicious nodes from injecting false data. The denoised data is encapsulated by the trusted data packet to form a closed-loop trusted chain from collection to application, supporting accurate decision-making for building energy conservation control.

[0056] S2. Use the blockchain method to dynamically adjust energy scheduling in combination with real-time electricity prices, calculate the optimal electricity procurement plan, and generate energy scheduling instructions.

[0057] Furthermore, the standard data packet is used to verify the data integrity through a smart contract, and a verified real-time status data packet is output.

[0058] Specifically, the standardized data packet is sent to the blockchain network, and the smart contract is used to perform validations on time validity, device legality, and data consistency. The verification results on the chain are pushed to the energy scheduling system and stored in the local cache database for quick query at the same time. The smart contract verifies the consistency of the hash value and time series, eliminates transmission interference or malicious tampering, reduces incorrect scheduling caused by sensor failures or cyberattacks, and outputs a verified real-time status data packet.

[0059] It should be noted that for time validity verification, the time difference between the data timestamp and the blockchain time does not exceed the estimated time. For device legality verification, it is verified whether the device is in the pre-registered whitelist. For data consistency verification, it is verified whether the data hash has been tampered with.

[0060] Preferably, the dual verification of Kalman filtering and blockchain smart contracts realizes end-to-end data trust from physical sensors to control systems. Tamper-proofing at the physical layer eliminates abnormal jump data caused by electromagnetic interference, mechanical vibration, etc. in sensors, ensuring the authenticity of the original signal. In the transmission layer, each data packet is accompanied by a blockchain transaction ID for traceability, and any data modification will leave an indelible record. Trusted device identity prevents unauthorized devices from accessing through the device ID whitelist mechanism, preventing malicious nodes from forging data to attack the system.

[0061] It should be noted that stochastic differential games are used for dynamic optimization to calculate the optimal strategy plan.

[0062] Furthermore, key feature parameters of standardized data packets and real-time electricity prices are extracted to output a standardized dynamic energy state vector.

[0063] Specifically, a high-dimensional feature vector covering temporal features and statistical features is extracted from the standardized data. Temporal features calculate the mean, variance, and fast Fourier transform frequency-domain energy through a sliding window. At the same time, a window sliding and feature combination strategy is used to extract electricity price features, and cross-feature analysis is used to analyze the dynamic correlation between electricity price and load, outputting a high-dimensional vector containing multi-dimensional features to characterize the real-time state of the energy system. Through an autoencoder for unsupervised feature compression, a standardized dynamic energy state vector is output.

[0064] Preferably, standardization and dimensionality reduction reduce data redundancy and improve algorithm efficiency. The sliding window temporal features can capture electricity price mutations and improve dynamic response capabilities. Through multi-modal feature engineering and adaptive dimensionality reduction, heterogeneous energy data is transformed into a state vector with high information density, providing a dynamic output for subsequent optimization.

[0065] Combining stochastic differential games and reinforcement learning algorithms, the optimal electricity purchase strategy is calculated in real time, and an optimal strategy plan including electricity purchase volume and energy storage scheduling is output.

[0066] Specifically, the dynamic energy state vector is used as the input and combined with stochastic differential games. Building energy and the electricity market are used as game participants to construct a cost value function. At the same time, it is optimized by combining deterministic policy gradients. The deterministic policy gradient algorithm is used to directly output continuous electricity purchase actions, and finally an optimal strategy plan is output.

[0067] Preferably, for the solution of the strategies of stochastic differential games, by combining deep reinforcement learning and game theory methods, efficient cooperation and competition among multiple agents in a dynamic and uncertain environment are achieved. The agents can learn the optimal strategies in the continuous action space. By embedding stochastic differential equations, the strategies can naturally adapt to environmental noise and state fluctuations, thus enhancing the real-time response ability to dynamic environments. By solving the game equilibrium of multiple agents, the local optimality or selfish behavior of unilateral strategies is avoided, promoting the overall stability and fairness of the system and enhancing the scalability of the algorithm and the adaptability to complex scenarios.

[0068] It should be noted that the cost value function expression is as follows:

[0069] Q(s,a) = r + γ·Q(s′,a′);

[0070] Among them, Q(s,a) represents the expected total revenue of the current power purchase strategy. When the environment is in state s, the agent selects action a and then always acts according to the current strategy, and finally can obtain an estimated value of the total revenue. s represents the current state, which is the current electricity price, load, and weather data. a is the current action, representing the electricity purchase quantity. r represents the immediate reward, which is the cost of this power purchase. γ is the discount factor, representing the importance weight of future rewards, 0 ≤ γ ≤ 1. s′ represents the next state, which is the electricity price and load at the next moment. a′ is the next action, representing the electricity purchase quantity at the next moment.

[0071] Combined with the trigger of the smart contract for automatic execution, a scheduling instruction structure is generated, and the execution status hash and energy scheduling instruction for on-chain storage and proof are output;

[0072] Specifically, by defining transaction rules through blockchain smart contracts, setting power purchase conditions and energy storage scheduling logic, setting automatic electricity bill settlement and default penalty clauses, the smart contract deployed on the blockchain contains transaction rules and fund management logic. The rule code is public, and all participants can audit it, reducing human intervention and being transparent and trustworthy. When the data meets the conditions of the smart contract, the contract execution and data on-chain are automatically triggered, and real-time power data that can be verified on the chain is output. Using the smart contract to purchase electricity and schedule energy storage by itself, settle funds in real time, complete power transactions and generate verifiable transaction records.

[0073] Better, through the coordinated optimization of blockchain smart contracts and random differential games, the dynamic, reliable and efficient energy scheduling is achieved. The smart contract verification mechanism ensures that all input data are verified by digital signatures to prevent malicious nodes from forging or tampering with data, and improves the system's anti-attack capabilities. Through the integration of blockchain, game theory and smart contracts, a highly reliable, adaptive and low-latency energy scheduling system is constructed, which solves the shortcomings of traditional methods in data authenticity, dynamic optimization, execution efficiency and other aspects. Blockchain smart contracts are fully automated, and no human intervention is required from data input to transaction settlement. All operations and records are uploaded to the chain to prevent data forgery, eliminate middlemen, reduce transaction friction costs and achieve high efficiency and low cost.

[0074] S3. Use the bee colony algorithm to divide the building plane, establish a dynamic air volume distribution matrix, output air volume control instructions and mark areas where demand is not met.

[0075] Furthermore, the building plane is dynamically divided, the air volume demand weights of different areas are determined, and the weighted initial air volume demand matrix is ​​output;

[0076] Specifically, the CAD floor plan and real-time infrared thermal imaging data are combined to automatically divide the areas based on the building use, such as office areas, conference rooms, corridors, and equipment rooms. The K-means clustering algorithm is used to group the sensor network topology to ensure that each partition contains at least one group of environmental sensors. If the population density in a certain area changes suddenly, the partition reconstruction is triggered, and a building partition map with topological relationships is output; the real-time sensor data and regional function labels are dynamically weighted by the entropy weight method, and the index matrix is ​​constructed and then standardized and the information entropy is calculated, and the air volume demand weight vector of each area is output; the building partition map, air volume demand weight vector and design specifications are combined to calculate the benchmark value according to the regional area and population density, and the weight vector is combined with the benchmark value to output the weighted air volume demand matrix;

[0077] The clustering algorithm reduces the redundancy of sensor deployment, the entropy weight method avoids subjective weighting bias, dynamically responds to environmental mutations, the regional function label reflects the regional priority, the matrix form is compatible with the subsequent bee colony algorithm optimization, reduces the number of iterations, and adapts to changes in building functions to improve control flexibility. Differentiated weight allocation can optimize the balance between comfort and energy saving;

[0078] Combined with the bee colony algorithm for optimization, the air volume constraint fitness function is constructed to output the optimal air volume allocation plan;

[0079] Specifically, the air volume constraint fitness function is constructed by combining the weighted air volume demand matrix, duct network topology parameters and real-time electricity price data, and the fitness function expression and constraint conditions are output to optimize comfort, energy consumption, and system stability, and reduce the bias caused by single-objective optimization; the bee swarm algorithm is used for iterative optimization, and the optimization is divided into four stages.

[0080] Specifically, in the first stage, the "employed bee" stage, each solution is locally searched by the employed bees around the current optimal solution. The second stage is the "onlooker bee" stage, where solutions are selected for in-depth search according to the fitness probability. The third stage is the "scout bee" stage, where solutions trapped in local optima are randomly reset to enhance the global search ability. The fourth stage is the termination output, which outputs the optimal air volume distribution plan;

[0081] It should be noted that the fitness function expression of the bee colony algorithm is as follows:

[0082] F = α·C + β·E + λ·P;

[0083] Among them, F is the comprehensive fitness score. The smaller it is, the better the distribution plan. C represents the comfort deviation, with a value range of [0, 1]. E is the energy consumption cost, with a value range of [0, 1]. P represents the pipeline network pressure imbalance degree, with a value range of [0, 0.5]. α is the comfort weight, with a value range of (0.4, 0.6), and the typical value is 0.5, which is dynamically adjusted according to the regional function. β represents the energy consumption weight, with a value range of (0.3, 0.5), and the typical value is 0.4, which is positively correlated with the real-time electricity price. λ is the pressure weight, with a value range of (0.1, 0.2).

[0084] Preferably, the air volume distribution is optimized by the bee colony algorithm, considering the comfort deviation, energy consumption cost, and pipeline network pressure balance at the same time. On the premise of ensuring the comfort of personnel, the operation strategy of the fan is dynamically adjusted in combination with the real-time electricity price to reduce the energy consumption cost. The employed bees, onlooker bees, and scout bees work together to quickly approach the global optimal solution, avoid being trapped in local optima, converge efficiently, and adapt to complex scenarios;

[0085] Combining the optimal air volume with the real-time standard data, the residual value is calculated using a sliding window, the abnormal areas that do not meet the requirements are marked, and the final air volume control instruction and the coordinates of the abnormal areas are output;

[0086] Specifically, the real-time sensor data is matched with the theoretical optimal air volume through the time stamp to make the data have spatio-temporal consistency. The missing data within the sensor sampling interval is filled by linear interpolation, and the dimensionality influence is eliminated by normalization processing. A standardized data sequence with time alignment is output to solve the problem of asynchronous time of multi-source data and improve the accuracy of residual calculation. The short-term anomalies are marked and a trigger instruction is set. When the trigger instruction is triggered, it becomes a continuous anomaly. At the same time, the optimal air volume instruction given by the bee colony algorithm is maintained in the normal area, and a new air volume distribution plan is output for the continuously abnormal area, and the final air volume control instruction and the coordinates of the abnormal area are output;

[0087] Preferably, by accurately identifying abnormal areas, the system reliability is improved. The dynamic residual calculation can quickly locate areas where requirements are not met. The hierarchical response mechanism distinguishes short-term fluctuations from continuous anomalies, reduces misoperations, realizes closed-loop control optimization, enhances comfort and energy efficiency, dynamically adjusts the air volume in abnormal areas, while maintaining the optimal distribution in normal areas to avoid global energy efficiency losses. The abnormal coordinate marking can assist manual troubleshooting quickly.

[0088] S4. When continuously unmet demand areas and meteorological mutations are detected, the adversarial generative network is used to simulate extreme working conditions and generate an emergency control instruction set to adjust the fan speed and damper opening.

[0089] Furthermore, abnormal fusion detection is carried out by combining abnormal coordinates and the meteorological mutation index, and an abnormal event report form is output;

[0090] Specifically, by combining abnormal coordinates and the meteorological mutation index, the sliding window mean filter is used to smooth the sensor data to eliminate instantaneous noise. The wavelet transform is used to extract mutation features from the meteorological data to detect abnormal events, and local abnormal vectors and meteorological mutation features are output; a spatio-temporal weight matrix is constructed, and the abnormal and meteorological data are fused through the tensor product to output an abnormal feature matrix with spatio-temporal labels. The isolation forest is used to detect the abnormal type, and then the entropy weight method is used to calculate the abnormal priority, and an abnormal event report form is output;

[0091] Preferably, multi-source data fusion improves the detection accuracy. The dynamic weight is more adaptable to environmental changes. The dynamic weight mechanism can automatically adjust the influence degree of different environmental parameters, adapt to various extreme conditions such as sandstorms, high temperatures, and high humidity. Through spatio-temporal tagging, the area and time of abnormal occurrence are clarified, which is convenient for targeted regulation. Abnormal classification and priority ranking help the system quickly distinguish equipment failures, meteorological mutations, or compound anomalies, and optimize the response strategy;

[0092] A candidate instruction set is generated through the adversarial generative network, and combined with the built-in upper and lower limits of the fan speed and the damper coupling constraint, control instructions are output;

[0093] Specifically, the abnormal feature matrix, the physical constraints of the fan and the damper, and the environmental state parameters are input. The abnormal feature matrix is dimensionally reduced to extract key abnormal patterns, and the physical constraints are encoded as conditional vectors; a candidate instruction set is output through the adversarial generative network instruction generation. The instruction set is constraint-verified, and the instructions that violate the coupling relationship are eliminated, and the instructions that meet the conditions are retained, and control instructions are output;

[0094] Reinforcement learning is used to optimize the control instructions to adjust the fan speed and damper opening;

[0095] Specifically, a reinforcement learning strategy is used to generate control instructions. The proximal policy optimization method is adopted to make the policy update stable. At the same time, hard constraints and soft constraint conditions are embedded to output optimized instructions. Check whether the output instructions meet the safe operating range of the equipment. If the instructions violate the constraint conditions, they will be automatically corrected to the nearest safe value, and the final instructions will be sent to the fan and the damper actuator to adjust the operating state of the equipment.

[0096] Preferably, it has strong dynamic adaptability and can respond to environmental changes in real time. The equipment safety and energy consumption are balanced through hard constraints and soft constraints. The hard constraints prevent the fan from overspeed and the damper from jamming, and the soft constraints reduce ineffective regulation and extend the equipment life.

[0097] It should be noted that the hard constraints are as follows: for example, the fan speed shall not be lower than 300 revolutions per minute or higher than 1500 revolutions per minute, and the damper opening shall not be lower than 20% or higher than 90%. The soft constraints are as follows: if the adjustment amplitude is too large (such as the speed change exceeds 50 revolutions per minute), a penalty will be imposed.

[0098] S5. Use federated learning to optimize global control and update the decisions of the bee colony algorithm and the generative adversarial network in real time.

[0099] Furthermore, combined with feature-level federated learning and standardized data packets, a global feature vector is output.

[0100] Specifically, for the data in each region, the time dimension and the space dimension are extracted to construct a regional feature matrix. The spatial weight allocation is used to calculate the weights according to the regional area and the personnel density. Through principal component analysis, the features are dimensionally reduced to generate an initial global feature vector. The residual feedback mechanism is used to compare the predicted value of the global feature vector with the actual sensor readings, calculate the residuals, and use the Kalman filter to dynamically adjust the feature weights in the global feature vector to output the calibrated global feature vector.

[0101] Use the decision tree algorithm to automatically generate control rules and output a control strategy table.

[0102] Specifically, combined with the global feature vector and the historical control records, the random forest algorithm is used to calculate the importance scores of each feature in the global feature vector, screen out the key features, perform equal-frequency binning on the continuous features, and convert them into discrete values that can be processed by the decision tree to output an optimized feature subset. The C4.5 algorithm is used to generate a decision tree, with the features of the global feature vector as the input and the control instructions as the output, to output an initial control strategy table. Based on pessimistic error estimation pruning, the subtrees with little improvement in accuracy (which can be set by oneself, such as less than 2%) are removed. For the rules that produce different outputs under the same input conditions, weighted voting is adopted to output an optimized control strategy table.

[0103] Preferably, the decision tree algorithm automatically generates control rules, improving the interpretability and transparency of the control strategy. Through feature screening and rule pruning, it reduces the computational resource requirements, adapts to edge devices, dynamically adapts to environmental changes, and enhances robustness.

[0104] Use the adaptive adjustment algorithm combined with dynamic weight adjustment to update the decision in real time.

[0105] Specifically, assign initial weights to each rule by combining the control strategy table, real-time environmental data, and historical execution records. Calculate the execution residuals of each rule based on a time window. The sliding window reduces instantaneous noise interference and improves weight stability. Calculate the normalized residuals for each rule, add constraint conditions through the projected gradient method, and output the updated weight vector. Dynamically allocate the weights of each objective using the entropy weight method, use the non-dominated solution set to screen and eliminate the dominated rules, retain the non-inferior solution set, and output the optimized control instructions that simultaneously meet the energy consumption and comfort requirements. If the residuals of all rules exceed the threshold, trigger the emergency simulation in step four, and finally achieve real-time decision update.

[0106] Preferably, the real-time optimization of control decisions through the adaptive weight adjustment algorithm realizes high robustness, multi-objective balance, and energy efficiency maximization in a dynamic environment. Combining the entropy weight method and non-dominated solution set screening, it automatically balances conflicting objectives such as energy consumption, comfort, and equipment life. The projected gradient method is used to constrain the weight range to avoid algorithm divergence.

[0107] This embodiment also provides a fresh air control system for an integrated new energy-saving building, including: a data processing module for real-time collecting sensor data, performing wave noise reduction processing using Kalman filtering, and combining smoothing processing to construct a standardized data packet containing timestamps, device IDs, and numerical types; an energy scheduling module for dynamically adjusting energy scheduling using the blockchain method combined with real-time electricity prices, calculating the optimal electricity procurement plan, and generating energy scheduling instructions; a air volume distribution module for dividing the building plane using the bee colony algorithm, establishing a dynamic air volume distribution matrix, outputting the optimal air volume distribution plan, and marking the areas where the demand is not met; an emergency simulation module for simulating extreme working conditions and generating an emergency control instruction set by using a generative adversarial network when detecting continuously unmet demand areas and meteorological mutations, and adjusting the fan speed and damper opening; a global optimization module for optimizing global control using federated learning and updating the decisions of the bee colony algorithm and the generative adversarial network in real time.

[0108] This embodiment also provides a computer device applicable to the fresh air control method for an integrated new energy-saving building, including: a memory and a processor; the memory is used for storing computer-executable instructions, and the processor is used for executing the computer-executable instructions to implement the fresh air control method for an integrated new energy-saving building as proposed in the above embodiment.

[0109] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, trackball, or touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0110] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the air-conditioning fresh air control method for realizing an integrated new energy-saving building as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0111] In summary, the present invention, through: the collaborative application of the Kalman filter and the blockchain method, improves the signal-to-noise ratio of sensor data and effectively eliminates the influence of instantaneous interference on control instructions. The timestamp and device ID binding mechanism of the standardized data packet greatly reduces the time alignment error control of multi-source data. At the energy scheduling level, based on the real-time electricity price response algorithm of stochastic differential game and combined with the automatic execution characteristics of the blockchain smart contract, the fluctuation range of the electricity purchase cost is compressed. The global feature vector update mechanism of federated learning enables the system to adapt to building layout changes in real time and improves the overall energy efficiency ratio.

[0112] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling fresh air in air conditioning of a comprehensive new energy-saving building, characterized in that: include, Collect sensor data in real time, use Kalman filtering to reduce noise, combine with smoothing, and build a standardized data packet containing timestamp, device ID and value type; Use blockchain methods combined with real-time electricity prices to dynamically adjust energy dispatch, calculate the optimal electricity procurement plan, and generate energy dispatch instructions; The bee colony algorithm is used to divide the building plane, establish a dynamic air volume distribution matrix, output air volume control instructions and mark areas where demand is not met; When areas with continuous unmet demand and sudden meteorological changes are detected, the adversarial generation network simulates extreme conditions and generates emergency control instruction sets to adjust the fan speed and air valve opening; Federated learning is used to optimize global control and update the decisions of the swarm algorithm and the generative adversarial network in real time.

2. The air conditioning fresh air control method for the comprehensive new energy-saving building according to claim 1 is characterized in that: The sensor data includes indoor data and outdoor data; The Kalman filter is used to perform wave noise reduction processing, combined with smoothing processing, to construct a standardized data packet containing a timestamp, a device ID, and a numerical type. The specific steps are as follows: Use Kalman filtering to dynamically process indoor and outdoor data and output a stable data sequence after noise reduction; Trusted data packets are used to encapsulate and integrate the timestamp, device ID, and value type of the data, and output standardized data packets.

3. The air conditioning fresh air control method for the comprehensive new energy-saving building according to claim 2 is characterized in that: The specific steps of using the blockchain method in combination with real-time electricity prices to dynamically adjust energy scheduling, calculate the optimal electricity procurement plan, and generate energy scheduling instructions are as follows: Verify the data integrity of the standard data packet through the smart contract and output the verified real-time status data packet; Use stochastic differential games for dynamic optimization and calculate the optimal strategy solution; Combined with smart contracts to trigger automatic execution, generate a scheduling instruction structure, and output the execution status hash and energy scheduling instructions for on-chain storage.

4. The air conditioning fresh air control method for the comprehensive new energy-saving building as claimed in claim 3 is characterized by: The bee colony algorithm is used to divide the building plane, establish a dynamic air volume distribution matrix, output air volume control instructions and mark the areas that do not meet the demand. The specific steps are as follows: Dynamically divide the building plane, determine the air volume demand weights of different areas, and output the initial air volume demand matrix with weights; Combined with the bee colony algorithm for optimization, the air volume constraint fitness function is constructed to output the optimal air volume allocation plan; Combine the optimal air volume with real-time standard data, use a sliding window to calculate the residual value, mark the abnormal areas that do not meet the requirements, and output the final air volume control instructions and the coordinates of the abnormal areas.

5. The air conditioning fresh air control method for the comprehensive new energy-saving building as claimed in claim 4 is characterized in that: When the continuous unmet demand area and the sudden change of weather are detected, the extreme working conditions are simulated by the adversarial generation network and an emergency control instruction set is generated to adjust the fan speed and the air valve opening. The specific steps are as follows: Combine abnormal coordinates with meteorological mutation index to perform abnormal fusion detection and output abnormal event report table; Generate candidate instruction sets through adversarial generative networks, combine built-in upper and lower limits of fan speed and air valve coupling constraints, and output control instructions; Reinforcement learning is used to optimize control instructions and adjust fan speed and air valve opening.

6. The air conditioning fresh air control method for the comprehensive new energy-saving building as claimed in claim 5 is characterized by: The specific steps of using federated learning to optimize global control and update the decisions of the swarm algorithm and the adversarial generative network in real time are as follows: Combine feature-level federated learning with standardized data packets to output a global feature vector; Automatically generate control rules using decision tree algorithm and output control strategy table; Use an adaptive adjustment algorithm combined with dynamic weight adjustment to update decisions in real time.

7. The air conditioning fresh air control method for the comprehensive new energy-saving building according to claim 6 is characterized in that: The random differential game is used for dynamic optimization to calculate the optimal strategy. The specific steps are as follows: Extract characteristic parameters of standardized data packets and real-time electricity prices, and output standardized dynamic energy state vectors; Combining stochastic differential game and reinforcement learning algorithm, the optimal power purchase strategy is calculated in real time, and the optimal strategy plan including power purchase amount and energy storage scheduling is output.

8. An air conditioning fresh air control system for a comprehensive new energy-saving building, based on the air conditioning fresh air control method for a comprehensive new energy-saving building according to any one of claims 1 to 7, characterized in that: Including data processing module, energy scheduling module, air volume distribution module, emergency simulation module and global optimization module; The data processing module is used to collect sensor data in real time, use Kalman filtering to perform wave noise reduction processing, and combine smoothing processing to construct a standardized data packet containing a timestamp, device ID and value type; The energy dispatch module is used to dynamically adjust energy dispatch using blockchain methods combined with real-time electricity prices, calculate the optimal electricity procurement plan, and generate energy dispatch instructions; The air volume distribution module is used to divide the building plane using a bee colony algorithm, establish a dynamic air volume distribution matrix, output the optimal air volume distribution plan and mark the area where the demand is not met; The emergency simulation module is used to simulate extreme working conditions and generate emergency control instruction sets through the adversarial generation network to adjust the fan speed and air valve opening when the continuous unmet demand area and weather mutation are detected; The global optimization module is used to optimize global control by using federated learning and to update the decisions of the swarm algorithm and the adversarial generation network in real time.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the air conditioning fresh air control method for the comprehensive new energy-saving building according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the air conditioning fresh air control method for a comprehensive new energy-saving building according to any one of claims 1 to 7 are implemented.

Images