Indoor space environment control method based on optimization algorithm and related device

Abstract

The application provides an indoor space environment control method based on an optimization algorithm and related equipment, and the method comprises the following steps: constructing an indoor comprehensive environment model by combining historical equipment control parameters and historical space environment data; constructing a comprehensive target function; creating a state space of an ant colony optimization algorithm and generating constraint conditions of the ant colony optimization algorithm; taking the comprehensive target function as an algorithm optimization target, generating an initial equipment control strategy of an environment control device based on the indoor comprehensive environment model and using the ant colony optimization algorithm; taking an optimal parameter interval as a strategy initial solution of a model predictive control strategy, generating a strategy prediction result by the model predictive control strategy based on the indoor comprehensive environment model; and updating algorithm parameters of the ant colony optimization algorithm according to the strategy prediction result and generating an optimal equipment control strategy of the environment control device. The application has the effect of meeting the environmental experience comfort of users on the basis of the lowest energy consumption.

Description

Technical Field

[0001] This invention belongs to the field of environmental control optimization technology, specifically relating to an indoor space environmental control method and related equipment based on optimization algorithms. Background Technology

[0002] Faced with global climate change and the energy crisis, improving building energy efficiency has become an urgent task. Against this backdrop, applications such as intelligent building management systems, home automation control, and office environment optimization are gradually becoming more widespread. These applications aim to intelligently adjust indoor environmental parameters through advanced sensing technologies, network communication, and control algorithms to create comfortable, healthy, and efficient indoor environments.

[0003] However, the indoor environment is a highly complex and dynamically changing system involving multiple interacting factors. While traditional environmental control equipment can intelligently adjust based on environmental data, it often employs simple threshold control or fixed control logic, lacking the ability to accurately grasp and predict dynamic changes in the indoor environment. To meet user comfort requirements, the system tends to over-adjust, leading to excessive energy consumption; conversely, when switching to energy-saving mode, it often fails to achieve the ideal comfort level, negatively impacting the user experience. This contradiction between comfort and energy efficiency has become a major challenge for the industry. Summary of the Invention

[0004] This invention provides an indoor space environment control method and related equipment based on optimization algorithms to solve the problem that environmental control equipment is difficult to balance the contradiction between comfort and energy efficiency.

[0005] In a first aspect, the present invention provides an indoor space environment control method based on an optimization algorithm, the method comprising the following steps:

[0006] The historical equipment control parameters of the environmental control equipment and the historical spatial environment data corresponding to the historical equipment control parameters are obtained. The historical spatial environment data includes historical indoor environment data inside the indoor enclosed space where the environmental control equipment is deployed and historical outdoor environment data outside the indoor enclosed space.

[0007] An integrated indoor environment model of the enclosed indoor space is constructed by combining the historical equipment control parameters and the historical spatial environment data.

[0008] A comprehensive objective function is constructed based on the indoor environmental comfort level of the enclosed space and the energy consumption index of the environmental control equipment.

[0009] The state space of the ant colony optimization algorithm is created based on the list of device control parameters of the environmental control equipment, and the constraints of the ant colony optimization algorithm are generated according to the parameter limits of each device control parameter in the list of device control parameters.

[0010] Using the comprehensive objective function as the algorithm optimization objective, an initial equipment control strategy for the environmental control device is generated based on the indoor comprehensive environment model and the ant colony optimization algorithm. The initial equipment control strategy includes the optimal parameter range for each of the equipment control parameters.

[0011] The optimal parameter range is used as the initial solution of the model predictive control strategy, and the comprehensive objective function is used as the strategy optimization objective of the model predictive control strategy. Based on the indoor comprehensive environment model, the strategy prediction result is generated through the model predictive control strategy.

[0012] The algorithm parameters of the ant colony optimization algorithm are updated based on the prediction results of the strategy, and the optimal device control strategy of the environmental control device is generated using the updated ant colony optimization algorithm.

[0013] Optionally, constructing the comprehensive indoor environment model of the enclosed indoor space by combining the historical equipment control parameters and the historical spatial environment data includes the following steps:

[0014] A data association driving model between the historical equipment control parameters and the historical spatial environment data is constructed based on convolutional neural networks and long short-term memory networks.

[0015] An indoor physical model of the enclosed indoor space is constructed by combining the historical indoor environmental data and the historical outdoor environmental data. The indoor physical model includes a thermodynamic model, a humidity model, an air quality model, and a lighting model.

[0016] The gray box model method is used to merge the data association-driven model and the indoor space physical model into an integrated indoor environment model of the enclosed indoor space.

[0017] Optionally, the formula for the thermodynamic model is as follows:

[0018]

[0019] In the formula: T in The indoor ambient temperature of the enclosed space is represented by C, where C represents the heat capacity of the indoor air, and Q represents the indoor temperature. H Q represents the heat provided by the environmental control equipment. s Q represents outdoor solar radiation heat. in The heat generated by the indoor heat source is represented by UA, which represents the total thermal conductivity of the building containing the enclosed indoor space, and T is the total thermal conductivity of the building.out Indicates the outdoor ambient temperature;

[0020] The formula for the humidity model is as follows:

[0021]

[0022] In the formula: W in V represents the absolute humidity of the enclosed indoor space, and m represents the indoor volume of the enclosed indoor space. s W represents the air supply coverage of the environmental control equipment. s Indicates the absolute humidity of the supply air, m r The return air volume of the environmental control equipment is expressed in m. in W represents the amount of infiltrated air. out G represents the absolute outdoor humidity, and G represents the indoor humidity generation rate.

[0023] The formula for the air quality model is as follows:

[0024]

[0025] In the formula: C in G represents the CO2 concentration in the enclosed indoor space. CO2 Q represents the CO2 production rate of the enclosed indoor space, and C represents the ventilation volume of the enclosed indoor space. out Indicates outdoor CO2 concentration;

[0026] The formula for the lighting model is as follows:

[0027] E = k1L + k2I s +δ

[0028] In the formula: E represents the indoor illuminance of the enclosed indoor space, L represents the artificial lighting brightness controlled by the environmental control equipment, and I s The outdoor solar irradiance is represented by k1 and k2, which represent the equipment illumination influence coefficient and the natural illumination influence coefficient, respectively, and δ represents the adjustment coefficient.

[0029] Optionally, the step of using the gray box model method to fuse the data association-driven model and the indoor space physical model into a comprehensive indoor environment model of the enclosed indoor space includes the following steps:

[0030] For any of the indoor space physical models, a new model error term is introduced for the indoor space physical model;

[0031] By combining the data association-driven model and the indoor space physical model, an initial comprehensive indoor environment model of the enclosed indoor space is obtained;

[0032] After preprocessing the historical equipment control parameters and the historical spatial environment data, the historical equipment control parameters and the historical spatial environment data are integrated into a model training set;

[0033] The model training set is input into the initial indoor integrated environment model. Based on the historical equipment control parameters, the model outputs environmental correlation change results through the data correlation driving model. Combining the environmental correlation change results and the historical spatial environment data, the indoor spatial environment prediction data is output through the indoor spatial physical model.

[0034] The mean square error between the predicted indoor space environment data and the historical indoor environment data is used as the model loss function.

[0035] The model error term of the indoor space physical model and the model parameters of the data association-driven model are adjusted simultaneously according to the model loss function.

[0036] The initial indoor integrated environment model is continuously trained using the model training set until the model loss function reaches its minimum value, thus obtaining the trained indoor integrated environment model.

[0037] Optionally, after preprocessing the historical equipment control parameters and the historical spatial environment data, integrating the historical equipment control parameters and the historical spatial environment data into a model training set includes the following steps:

[0038] Unify the data sampling rate of the historical equipment control parameters and the historical spatial environment data;

[0039] The data of the historical equipment control parameters at each time point are aligned with the data of the historical spatial environment data at the previous time point;

[0040] The historical equipment control parameters and historical spatial environment data, after data alignment, are integrated into a model training set.

[0041] Optionally, the step of using the comprehensive objective function as the algorithm optimization objective, and generating the initial equipment control strategy for the environmental control device based on the indoor comprehensive environment model and the ant colony optimization algorithm, includes the following steps:

[0042] Initialize the algorithm parameters of the ant colony optimization algorithm;

[0043] Based on the constraints, multiple ant positions are randomly generated in the state space, and different ant positions represent different device control parameters.

[0044] For each ant location, the corresponding comprehensive objective function value is calculated using the indoor integrated environment model;

[0045] Update the pheromone concentration in the state space, and update the ant position by combining the comprehensive objective function value and the updated pheromone concentration;

[0046] Repeat the above steps of calculating the comprehensive objective function value and updating the ant position until the maximum number of iterations of the ant colony optimization algorithm is reached, and output the optimal ant path under the constraints.

[0047] The optimal parameter ranges of each control parameter of the environmental control device are determined based on the optimal ant path, and the optimal parameter ranges of each control parameter of the environmental control device are integrated into an initial device control strategy.

[0048] Optionally, the step of using the optimal parameter range as the initial solution of the model predictive control strategy, using the comprehensive objective function as the strategy optimization objective of the model predictive control strategy, and generating strategy prediction results based on the indoor comprehensive environment model and through the model predictive control strategy includes the following steps:

[0049] The optimal parameter range is used as the initial solution of the model predictive control strategy;

[0050] Based on the initial solution of the strategy, the prediction time domain and control time domain of the model predictive control strategy are determined through the indoor integrated environment model;

[0051] Based on the prediction time domain, the comprehensive objective function is used as the policy optimization objective of the model predictive control strategy. The control sequence in the control time domain is continuously optimized using the sequential quadratic programming method until the comprehensive objective function reaches its minimum value, thus obtaining the optimal control sequence. The optimal control sequence is then used as the policy prediction result generated by the model predictive control strategy.

[0052] Optionally, updating the algorithm parameters of the ant colony optimization algorithm based on the strategy prediction result, and generating the optimal device control strategy for the environmental control device using the updated ant colony optimization algorithm, includes the following steps:

[0053] The pheromone distribution of the ant colony optimization algorithm is updated based on the prediction results of the strategy.

[0054] The target constraints of the ant colony optimization algorithm are generated based on the optimal parameter range.

[0055] Using the comprehensive objective function as the algorithm optimization objective, the optimal equipment control strategy for the environmental control device is generated by combining the state space and the objective constraints and employing the ant colony optimization algorithm.

[0056] In a second aspect, the present invention also provides an indoor space environment control system based on an optimization algorithm, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the indoor space environment control method based on the optimization algorithm as described in the first aspect.

[0057] Thirdly, the present invention also provides an environmental control device, including the indoor space environment control system based on the optimization algorithm described in the second aspect.

[0058] The beneficial effects of this invention are:

[0059] First, this invention can more accurately capture the dynamic changes in the indoor environment. By constructing a comprehensive environmental model, it fully considers the complex interactions between indoor and outdoor environmental factors, improving the accuracy and adaptability of environmental control. Second, by incorporating comfort and energy efficiency into a unified optimization framework, this invention effectively resolves the contradiction between comfort and energy saving in traditional methods, achieving a balanced optimization of both. This invention uses an ant colony optimization algorithm to generate an initial control strategy and combines it with a model-predictive control strategy for further optimization. This allows for continuous updates to the optimization algorithm parameters based on the strategy prediction results, enabling the control strategy to continuously improve over time. This multi-level optimization method significantly improves the efficiency and effectiveness of the control strategy. Moreover, the control strategy of this invention considers the limit constraints of equipment control parameters, ensuring that the generated control strategy is within the actual operable range, improving the practicality and reliability of the solution. This allows for minimizing energy consumption while ensuring user comfort, providing an efficient, accurate, and learning-capable environmental control solution for intelligent buildings and energy conservation and environmental protection. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating an indoor space environment control method based on an optimization algorithm in one embodiment of this application.

[0061] Figure 2 This is a schematic diagram of the system structure of the environmental control device in one embodiment of this application. Detailed Implementation

[0062] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0063] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0064] Figure 1 This is a flowchart illustrating an indoor space environment control method based on an optimization algorithm in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 1 As shown, the indoor space environment control method based on optimization algorithm disclosed in this invention specifically includes the following steps:

[0065] S101. Obtain the historical equipment control parameters of the environmental control equipment, as well as the historical spatial environmental data corresponding to the historical equipment control parameters.

[0066] Historical spatial environmental data includes historical indoor environmental data for enclosed spaces equipped with environmental control devices and historical outdoor environmental data for the exterior of these spaces. Specifically, historical control parameters of the environmental control devices are first collected, including temperature setpoints, humidity setpoints, wind speed setpoints, and lighting system brightness setpoints. Simultaneously, historical spatial environmental data corresponding to these control parameters also needs to be collected. This environmental data is divided into two parts: indoor environmental data and outdoor environmental data. Indoor environmental data includes parameters such as indoor temperature, humidity, CO2 concentration, and illuminance. This data is typically collected using various sensors installed indoors. For example, temperature and humidity sensors can monitor real-time changes in indoor temperature and humidity; CO2 sensors can detect indoor air quality; and illuminance sensors are used to measure indoor light intensity. These sensors typically record data at regular time intervals (e.g., every 5 minutes or every 10 minutes), forming a continuous time series.

[0067] Outdoor environmental data includes parameters such as outdoor temperature, humidity, wind speed, wind direction, and solar radiation intensity. This data can be obtained from weather stations installed outside buildings or from local meteorological departments. The collection frequency of outdoor environmental data is usually consistent with that of indoor environmental data to facilitate subsequent data analysis and modeling. In practice, this historical data can be stored in a database. The database structure includes the following main tables:

[0068] Equipment control parameter table: Records the set values ​​of each environmental control device at each point in time.

[0069] Indoor Environmental Data Sheet: Records indoor environmental parameters at each point in time.

[0070] Outdoor Environmental Data Table: Records outdoor environmental parameters at each point in time.

[0071] These tables are linked by timestamps to ensure data consistency and traceability. The data collection time span typically needs to cover at least one complete seasonal cycle, ideally a whole year, to capture environmental change patterns under different seasons and weather conditions. The data collection frequency needs to be high enough to capture the dynamic characteristics of environmental changes, but storage and processing capacity limitations must also be considered.

[0072] Data preprocessing is a crucial step in this process. First, the raw data needs to be cleaned to remove obvious outliers and missing values. Outliers may be caused by sensor malfunctions or temporary interference, and can be identified and handled by setting reasonable thresholds or using statistical methods (such as the 3σ rule). For missing values, interpolation methods can be used to impute them, such as linear interpolation or more complex time-series interpolation methods. Furthermore, the data needs to be standardized to unify data from different scales onto the same unit of measurement. Common standardization methods include min-max standardization and Z-score standardization. Through these processes, a clean, consistent historical dataset suitable for subsequent analysis can be obtained. This dataset will provide an important foundation for subsequent modeling and optimization, enabling environmental control equipment to make smarter and more efficient decisions and controls based on the patterns and regularities of historical data.

[0073] S102. Construct an integrated indoor environmental model of the enclosed space by combining historical equipment control parameters and historical spatial environment data.

[0074] In this step, the main task is to construct a comprehensive indoor environmental model of the enclosed space. The core objective of this model is to establish a data correlation between the control parameters of environmental control equipment and indoor and outdoor environmental data. This correlation will help understand and predict the impact of changes in control parameters on the indoor environment, providing a theoretical basis for subsequent optimized control. To construct the comprehensive indoor environmental model, historical data collected in step S101 is required. Various modeling methods can be employed. For example, the physical model method can estimate model parameters based on physical equations using optimization algorithms such as least squares. Data-driven methods, on the other hand, use machine learning algorithms such as multiple linear regression, support vector machine regression, random forests, or neural networks to directly learn the relationship between control parameters and environmental parameters from the data.

[0075] The constructed comprehensive indoor environment model will provide crucial support for subsequent optimized control. It can help predict indoor environmental changes under different control strategies, evaluate the effectiveness of these strategies, and provide a simulation platform for finding the optimal control strategy. Simultaneously, this model can also be used for real-time monitoring and anomaly detection, allowing for the timely identification of anomalies or performance degradation in the environmental control system by comparing model predictions with actual measurements.

[0076] S103. Construct a comprehensive objective function based on the indoor environmental comfort level of enclosed spaces and the energy consumption index of environmental control equipment.

[0077] This step involves constructing a comprehensive objective function that simultaneously considers indoor environmental comfort and the energy consumption index of environmental control equipment. This comprehensive objective function will serve as the common optimization objective for subsequent ant colony optimization algorithms and model predictive control strategies, aiming to minimize energy consumption while ensuring indoor environmental comfort. The process of constructing the comprehensive objective function typically includes the following key aspects:

[0078] 1. Indoor environmental comfort evaluation:

[0079] Indoor environmental comfort typically includes multiple aspects such as thermal comfort, humidity comfort, air quality comfort, and visual comfort. A comfort index can be defined for each aspect, and then these can be combined into a comprehensive comfort index.

[0080] a) Thermal comfort: This can be achieved using the PMV (Predicted Average Votes) index or an adaptive thermal comfort model. The formula for calculating the PMV model is as follows:

[0081] PMV = (0.303e -0.036M +0.028)L

[0082] Where M is the human metabolic rate and L is the heat load. A PMV value between -0.5 and +0.5 is considered comfortable.

[0083] b) Humidity comfort: This can be evaluated directly using relative humidity, or by combining it with dew point temperature. The ideal relative humidity range is typically between 40% and 60%.

[0084] c) Air quality comfort: CO2 concentration can be used as an indicator (or the concentration of other gases that affect indoor air quality can be used as an indicator). A CO2 concentration below 1000 ppm is generally considered good indoor air quality.

[0085] d) Visual comfort: This can be evaluated using indoor illuminance levels. Different scenarios have different recommended illuminance ranges; for example, the recommended illuminance range for offices is typically between 300 and 500 lux.

[0086] The overall comfort index can be expressed as:

[0087] P C =w1f(PMV)+w2g(RH)+w3h(CO2)+w4i(Lux)

[0088] Where f(), g(), h(), i() are the normalization functions of each comfort index, and w1, w2, w3, w4 are the weights of each index.

[0089] 2. Energy Consumption Index Evaluation:

[0090] The energy consumption index typically includes the energy consumption of air conditioning, lighting, and ventilation systems within environmental control equipment. Total energy consumption can be calculated using the following formula:

[0091] P E =E H +E L +E V

[0092] Among them, E H It refers to the energy consumption of the air conditioning system, E. L It is the energy consumption of the lighting system, E V It refers to the energy consumption of the ventilation system.

[0093] The energy consumption of each system can be estimated by the power of the equipment and its operating time, for example:

[0094]

[0095] Among them, P i t is the rated power of the i-th air conditioning unit. i It refers to runtime, COP. i It refers to the energy efficiency ratio.

[0096] 3. Construction of the comprehensive objective function:

[0097] The overall objective function needs to consider both comfort and energy consumption, and can be expressed as a weighted sum:

[0098] J = α·P C -β·P E

[0099] Here, α and β are weighting coefficients used to balance the importance of comfort and energy consumption. A negative sign is used because the goal is to maximize comfort while minimizing energy consumption.

[0100] 4. Definition of constraints:

[0101] While constructing the objective function, it is also necessary to define some constraints to ensure that the optimization results are within a practically feasible range. These constraints include:

[0102] a) Limitations on the range of equipment operating parameters, such as the range of temperature setpoints and the range of wind speed.

[0103] b) Minimum requirements for comfort indicators, such as an absolute value of PMV not exceeding 0.5.

[0104] c) The upper limit of energy consumption is determined by the building's energy quota or the capacity of the power system.

[0105] These constraints can be expressed as inequality constraints.

[0106] 5. Normalization of the objective function:

[0107] Because comfort and energy consumption indicators have different dimensions and orders of magnitude, directly adding them together can lead to one aspect dominating the optimization process. Therefore, it is necessary to normalize these indicators to ensure their numerical ranges are consistent. Common normalization methods include min-max normalization and Z-score normalization. Normalization ensures that comfort and energy consumption are considered in a balanced manner during optimization. In constructing the comprehensive objective function, optimization objectives at different time scales also need to be considered. Short-term objectives focus more on immediate comfort, while long-term objectives focus more on energy efficiency and equipment lifespan. A time-weighted factor can be introduced to balance short-term and long-term objectives, allowing the control strategy to achieve a balance between immediate comfort and long-term benefits.

[0108] Furthermore, the construction of the objective function should also consider the dynamic characteristics of the environmental control system. For example, air conditioning systems experience significant energy consumption fluctuations during startup and shutdown, and frequent switching can affect equipment lifespan. Therefore, a penalty term can be added to the objective function to reduce unnecessary frequent adjustments. In practical applications, the weights and constraints of various indicators need to be continuously adjusted based on system operating data and user feedback to achieve the best control effect. The most suitable form of the objective function for a specific environment and user needs should be determined through repeated trials and optimizations.

[0109] S104. Create the state space of the ant colony optimization algorithm based on the list of equipment control parameters of the environmental control equipment, and generate the constraint conditions of the ant colony optimization algorithm according to the parameter limits of each equipment control parameter in the list of equipment control parameters.

[0110] This step provides the search space and search boundary for the subsequent ant colony optimization algorithm, and is a crucial foundation for the optimization process. First, it's necessary to determine the list of control parameters for the environmental control equipment. This list typically includes temperature setpoints, humidity setpoints, and fan speed setpoints for air conditioning systems, brightness setpoints for lighting systems, and air exchange rate setpoints for ventilation systems. Each parameter represents a control dimension, and together they constitute a multi-dimensional control space. Creating the state space is essentially defining this multi-dimensional control space. Each control parameter corresponds to a dimension in the space, and the range of parameter values ​​defines the range of that dimension. For example, if the temperature setpoint range is 18℃ to 30℃, then in the temperature dimension, the state space extends from 18 to 30.

[0111] For discrete parameters, such as wind speed with low, medium, and high settings, they can be mapped to a numerical space, for example, represented by 1, 2, and 3. The purpose of this is to unify all parameters into a numerical space, facilitating subsequent optimization calculations. The state space is typically defined in vector form. Assuming there are n control parameters, a point in the state space can be represented as an n-dimensional vector, with each component corresponding to the value of a control parameter. This vector represents a control strategy of the environmental control system.

[0112] Next, constraints need to be generated based on the parameter limits of the device control parameters. These constraints define the legal search region in the state space. Constraints typically include the following categories:

[0113] 1. Parameter range constraints: Each control parameter has its reasonable range of values. Exceeding this range will cause the equipment to malfunction or create an uncomfortable environment. For example, the indoor temperature setpoint is limited to between 18℃ and 30℃.

[0114] 2. Inter-parameter constraints: Some parameters are mutually influential or correlated. For example, the setpoints for temperature and humidity need to be considered simultaneously to avoid excessively humid or dry conditions.

[0115] 3. Equipment Capacity Constraints: The actual capabilities of environmental control equipment also impose constraints on parameter settings. For example, the cooling and heating capacity of an air conditioning system is limited, which restricts the range of indoor temperatures that can be achieved under extreme external environments.

[0116] 4. Energy consumption constraints: It is necessary to set an upper limit on total energy consumption to meet energy conservation requirements or power system capacity limitations.

[0117] 5. Comfort Constraints: Minimum comfort requirements need to be set to ensure that optimization results do not sacrifice user experience.

[0118] These constraints play a crucial role in ant colony optimization algorithms. They limit the range of movement of ants in the state space, ensuring that every considered solution is feasible. In the algorithm implementation, these constraints are typically transformed into decision conditions or penalty functions to guide the ants' path selection and pheromone updates. This step defines a structured search space and explicit constraints for the ant colony optimization algorithm. This provides clear boundaries and guidance for the subsequent optimization process, helping the algorithm to find the optimal control strategy that satisfies all requirements more efficiently.

[0119] S105. Using the comprehensive objective function as the algorithm optimization objective, the initial equipment control strategy of the environmental control equipment is generated based on the indoor comprehensive environment model and the ant colony optimization algorithm.

[0120] Among them, the ant colony optimization algorithm is a swarm intelligence optimization algorithm inspired by the foraging behavior of ants. In this algorithm, when ants move in the state space, they choose paths based on pheromone concentration and heuristic information. Pheromones concentration reflects historical search experience, while heuristic information represents the estimated distance from the current state to the target. The specific implementation process of the algorithm is as follows:

[0121] A certain number of ants are randomly generated in the state space, with each ant representing a set of initial device control parameters. Simultaneously, the pheromone distribution is initialized, typically with the initial pheromone concentration on all paths set to the same small positive number.

[0122] Each ant chooses its next direction of movement based on the pheromone concentration at its current location and heuristic information. In environmental control problems, this is equivalent to adjusting the values ​​of various control parameters. The probability of selection is proportional to the weighted product of the pheromone concentration and the heuristic information. The pheromone concentration reflects historical search experience, and the heuristic information can be designed based on the estimated contribution of the current parameter settings to the objective function.

[0123] The solution (i.e., a set of control parameters) constructed by each ant is evaluated. The evaluation criterion is the comprehensive objective function defined in step S103, while also considering the constraints set in step S104. If a solution violates the constraints, a penalty function can be used to reduce its score.

[0124] The pheromone distribution in the state space is updated based on the evaluation results of each ant. Generally, paths with higher scores receive a greater increase in pheromones. At the same time, pheromones on all paths undergo a certain degree of evaporation to prevent the algorithm from prematurely converging to a local optimum.

[0125] Repeatedly update the ant's movement position, evaluate the solution constructed by the ant, and update the pheromone distribution until the preset number of iterations is reached or a certain convergence condition is met.

[0126] After the algorithm completes, the optimal solution is selected as the initial device control strategy. This strategy includes the optimal range for each control parameter.

[0127] In practical applications, multiple independent optimization runs are required to reduce the impact of random factors and improve the reliability of the results. Furthermore, elitist strategies and local search strategies can be introduced to enhance algorithm performance. Ant colony optimization can efficiently search for near-optimal control strategies in complex multidimensional state spaces. This initial control strategy provides a good starting point for subsequent model predictive control, contributing to improved efficiency and performance of the entire control system.

[0128] It's important to note that the ant colony optimization algorithm generates an initial control strategy, which is insufficient to cope with complex and dynamic changes in the environment. This is why model predictive control is needed in subsequent steps to achieve real-time response and adjustment to environmental changes.

[0129] S106. The optimal parameter range is used as the initial solution of the model predictive control strategy, and the integrated objective function is used as the strategy optimization objective of the model predictive control strategy. Based on the indoor integrated environment model, the strategy prediction result is generated through the model predictive control strategy.

[0130] In this step, the optimal parameter range obtained through the ant colony optimization algorithm in step S105 is used as the initial solution for the Model Predictive Control (MPC) strategy, and the comprehensive objective function defined in step S103 is used as the optimization objective of MPC. Based on the indoor integrated environment model constructed in step S2, the strategy prediction results are generated through the MPC strategy. This process combines the global search capability of ant colony optimization with the dynamic prediction and real-time optimization capability of MPC to achieve more accurate and adaptive environmental control.

[0131] Model predictive control (MPC) is an advanced control strategy that uses a system model to predict the system's behavior over a future period and optimizes the control sequence accordingly. The core idea of ​​MPC is to perform an optimization calculation in each control cycle, but only execute the first step of the optimization result, and then restart the optimization process in the next control cycle. This rolling optimization approach allows MPC to adapt to the dynamic changes and uncertainties of the system.

[0132] In practical applications, the integrated indoor environmental model needs to be updated regularly to adapt to slow changes in system characteristics (such as changes in building thermal characteristics due to seasonal variations). This can be achieved through online parameter estimation or periodic model recalibration. Considering the existence of model errors and environmental disturbances, the MPC strategy needs to have a certain degree of robustness. This can be achieved by considering uncertainties during the optimization process, such as using robust MPC or stochastic MPC methods. These methods can consider the uncertainty range of model parameters or the statistical characteristics of environmental disturbances during the optimization process, thereby generating a more robust control strategy. MPC needs to complete the optimization calculation within each control cycle, so computational efficiency is an important consideration. Fast MPC algorithms, such as explicit MPC or approximate dynamic programming, can be used to improve computational speed. In addition, the length of the control cycle can be adjusted according to the dynamic characteristics of the system to achieve a balance between control performance and computational load. During the optimization process, various constraints must be strictly observed, including physical limitations of equipment, comfort requirements, and energy consumption limits. Soft constraint methods can be used, that is, incorporating the degree of constraint violation into the objective function to balance control performance and constraint satisfaction.

[0133] S107. Update the algorithm parameters of the ant colony optimization algorithm based on the strategy prediction results, and use the updated ant colony optimization algorithm to generate the optimal equipment control strategy for the environmental control equipment.

[0134] First, the pheromone distribution of the ant colony optimization algorithm is updated based on the prediction results of the MPC strategy. This update process reflects the MPC's prediction information about future environmental changes, helping to guide the ant colony algorithm to search for more promising solutions. Specifically, for control sequences that perform well in the MPC prediction results (i.e., sequences that can maintain high comfort and low energy consumption during the prediction period), the pheromone concentration is increased in their corresponding parameter space region. Control sequences that better meet various constraints are given a higher pheromone increase. Sequences that exhibit small fluctuations and maintain stable control effects during the prediction period have their pheromone concentration increased. Control strategies that perform well under different prediction scenarios are given a higher pheromone reward.

[0135] Next, new objective constraints for the ant colony optimization algorithm are generated based on the optimal parameter range obtained from MPC. Based on the MPC prediction results, a maximum allowable rate of change for the control parameters can be set to avoid drastic fluctuations in the control strategy. According to the comfort change trend predicted by MPC, a more stringent lower limit constraint for comfort is set. Energy consumption constraints need to be adjusted based on MPC-based energy consumption prediction to ensure long-term energy efficiency. Based on MPC predictions, some dynamic constraints need to be added, such as special requirements within specific time periods.

[0136] Then, the ant colony optimization algorithm is run again, with the comprehensive objective function as the optimization objective, searching within the updated state space while adhering to the new constraints. This optimization process is similar to step S105, as follows:

[0137] The optimal control sequence obtained by MPC is used as the initial position of some ants to accelerate convergence.

[0138] The calculation method of heuristic information can be adjusted based on the prediction results of MPC to better reflect the long-term control effect.

[0139] After each iteration, MPC can be used to perform local optimization on the current optimal solution, further improving the quality of the solution.

[0140] Considering the multiple performance metrics of MPC prediction, a multi-objective optimization method, such as Pareto optimization, is required.

[0141] During the optimization process, algorithm parameters such as pheromone update rate and evaporation rate can be dynamically adjusted based on the prediction of MPC.

[0142] Through this optimization process, the optimal control strategy for the environmental control equipment is ultimately obtained. This strategy considers not only the immediate control effect but also the prediction of future environmental changes, thus achieving better long-term control performance. During actual operation, the system continuously monitors environmental parameters and energy consumption data and periodically evaluates the control effect. If a significant deviation is found between the actual effect and the expectation, or if major changes occur in environmental conditions, the entire optimization process needs to be re-executed to maintain the optimality of the control strategy. By combining ant colony optimization algorithms, MPC, and dynamic feedback, an intelligent, efficient, and adaptable indoor environmental control system can be achieved, ensuring both user comfort and efficient energy utilization.

[0143] In one implementation, constructing an integrated indoor environmental model of an enclosed space by combining historical equipment control parameters and historical spatial environment data includes the following steps:

[0144] A data-driven model for the relationship between historical equipment control parameters and historical spatial environment data is constructed based on convolutional neural networks and long short-term memory networks.

[0145] An indoor physical model of an enclosed space is constructed by combining historical indoor and outdoor environmental data. The indoor physical model includes a thermodynamic model, a humidity model, an air quality model, and a lighting model.

[0146] The gray box model method is used to integrate the data association-driven model and the indoor space physical model into a comprehensive indoor environmental model of an enclosed indoor space.

[0147] In this implementation, a data-driven model is constructed based on Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs) to establish the relationship between historical equipment control parameters and historical spatial environment data. The core of this step is to utilize deep learning techniques to capture the complex nonlinear relationship between equipment control parameters and indoor environmental conditions. First, the architecture of a CNN-LSTM hybrid model is designed, with the CNN part extracting spatial features and the LSTM part capturing time-series characteristics. The input to the CNN is a multi-channel image representing the indoor environmental conditions, with each channel corresponding to a specific environmental parameter. A typical CNN structure includes multiple convolutional layers, activation functions (such as ReLU), and pooling layers. For example, a 3x3 convolutional kernel with a stride of 1 and padding of 1 can be used to maintain the spatial size of the feature map. The pooling layers can use 2x2 max pooling with a stride of 2 to reduce the size of the feature map.

[0148] The specific CNN structure is as follows:

[0149] Convolutional layer 1: 32 3x3 convolutional kernels, ReLU activation;

[0150] Max pooling layer 1: 2x2 pooling kernel, step size 2;

[0151] Convolutional layer 2: 64 3x3 convolutional kernels, ReLU activation;

[0152] Max pooling layer 2: 2x2 pooling kernel, step size 2;

[0153] Convolutional layer 3: 128 3x3 convolutional kernels, ReLU activation;

[0154] Global average pooling layer.

[0155] The feature map output by the CNN is flattened and fed into the LSTM network. The LSTM network is used to process time series data; its core is the LSTM unit, which contains an input gate, a forget gate, and an output gate. The mathematical expression of LSTM is as follows:

[0156] Input gate: i t =σ(W i ·[h t -1,x t ]+b i )

[0157] Forgotten Gate: f t =σ(W f ·[h t -1,x t ]+b f )

[0158] Output gate: o t =σ(W o ·[h t -1,x t ]+b o )

[0159] Candidate memory unit: C t =tanh(W C ·[h t -1,x t ]+b C )

[0160] Memory unit update: C t =f t *C t -1+i t *C t

[0161] Hidden state output: h t =o t *tanh(C t )

[0162] Here, σ is the sigmoid function, * denotes element-wise multiplication, and W and b are learnable parameters. The output of the LSTM is mapped to the final prediction result through a fully connected layer. The entire model is trained using the backpropagation algorithm, and the loss function can be either mean squared error (MSE) or mean absolute error (MAE). The optimizer can use the Adam algorithm, with a learning rate of 0.001 and a learning rate decay strategy. To prevent overfitting, a dropout layer can be added before the fully connected layer, with a dropout rate of 0.5. To improve the model's generalization ability, k-fold cross-validation can be used, typically with k set to 5 or 10. Furthermore, early stopping can be used to prevent overfitting, i.e., training stops when the performance on the validation set no longer improves. The advantage of this CNN-LSTM hybrid model is its ability to capture features in both spatial and temporal dimensions. CNNs effectively extract spatial distribution features, while LSTMs capture long-term dependencies in time series. This structure is particularly suitable for handling indoor environmental control problems because indoor environmental parameters have both spatial distribution characteristics (such as temperature gradients) and obvious time series characteristics (such as daily and weekly variations).

[0163] The data-driven model constructed using this method can learn complex nonlinear relationships and predict indoor environmental conditions under different control parameters. The model's output can be a prediction of indoor environmental parameters over a future period, such as temperature, humidity, and CO2 concentration. The model's performance can be measured by the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 The evaluation is based on metrics such as [insert metrics here]. The advantage of this data-driven approach is that it does not require explicit modeling of the physical processes and can automatically learn complex system dynamics.

[0164] Next, combining historical indoor and outdoor environmental data, an indoor physical model of the enclosed space is constructed. This physical model comprises four main components: a thermodynamic model, a humidity model, an air quality model, and a lighting model. This physics-based model provides a deep understanding of the mechanisms of indoor environmental change and offers a reliable description even when data is insufficient or new situations arise.

[0165] First, the thermodynamic model, based on the principle of energy balance, describes the distribution and changes in indoor temperature. This model considers multiple heat transfer pathways, including heat conduction through walls, radiant heat through windows, cooling / heating from the air conditioning system, and indoor heat sources (such as people and appliances). The core equation of the thermodynamic model can be expressed as:

[0166]

[0167] In the formula: T inThe temperature (C) represents the indoor ambient temperature of an enclosed space, and the heat capacity (Q) represents the indoor air. H Q represents the heat supplied by environmental control equipment. s Q represents outdoor solar radiation heat. in T represents the heat generated by indoor heat sources, UA represents the total thermal conductivity of the building containing the enclosed indoor space, and T represents the total thermal conductivity of the building. out This represents the outdoor ambient temperature. This equation needs to be solved numerically (such as the Runge-Kutta method). To improve the model's accuracy, the heat capacity effect of the wall can also be considered, dividing the wall into multiple layers, each with its own heat balance equation.

[0168] Humidity models describe changes in the moisture content of indoor air. This model needs to consider the dehumidification / humidification effect of the air conditioning system, moisture produced by human respiration and perspiration, and moisture brought in by outdoor air infiltration. The basic equation of the humidity model can be expressed as:

[0169]

[0170] In the formula: W in V represents the absolute humidity of an enclosed indoor space, and m represents the volume of the enclosed indoor space. s W represents the air supply coverage of environmental control equipment. s Indicates the absolute humidity of the supply air, m r This indicates the return air volume of environmental control equipment, in meters (m). in W represents the amount of infiltrated air. out Here, represents the absolute outdoor humidity, and G represents the indoor humidity generation rate. This model needs to be combined with a psychohumidity map to handle the conversion between relative and absolute humidity.

[0171] Air quality models primarily focus on changes in indoor CO2 concentration, as CO2 concentration is commonly used as an indicator of indoor air quality. This model needs to consider CO2 produced by human respiration, CO2 removed by ventilation systems, and CO2 infiltrated from outdoor air. The basic equation of an air quality model can be expressed as:

[0172]

[0173] In the formula: C in G represents the CO2 concentration in an enclosed indoor space. CO2 The value of C represents the CO2 production rate of an enclosed indoor space, where Q represents the ventilation volume of the enclosed indoor space. out This represents the outdoor CO2 concentration. This model can be extended to handle other air pollutants, such as particulate matter and volatile organic compounds.

[0174] Lighting models describe the distribution and changes in indoor illuminance levels, taking into account the combined effects of natural and artificial light. A basic lighting model can be represented as:

[0175] E = k1L + k2I s +δ

[0176] In the formula: E represents the indoor illuminance of the enclosed space, L represents the artificial lighting brightness controlled by the environmental control equipment, and I... s The outdoor solar irradiance is represented by k1 and k2, which represent the equipment illumination influence coefficient and the natural illumination influence coefficient, respectively, and δ represents the adjustment coefficient. The equipment illumination influence coefficient, the natural illumination influence coefficient, and the adjustment coefficient need to take into account the orientation and size of the windows, as well as the effects of shading facilities.

[0177] The parameters of these physical models (such as thermal conductivity and humidity generation rate) need to be determined through experimental measurements or data fitting. Optimization algorithms such as least squares can be used to estimate these parameters based on historical data. During the modeling process, the interactions between different models also need to be considered. For example, temperature changes affect relative humidity, humidity changes affect thermal comfort, and CO2 concentration affects ventilation strategies. These interactions can be handled through coupled equations or iterative solutions. The advantage of physical models lies in their strong interpretability, their ability to reflect the intrinsic mechanisms of the system, and their ability to provide a reasonable description even when data is scarce. For example, even in new buildings without historical data, a preliminary physical model can be constructed as long as the physical properties of the building materials and the spatial geometry are known. Furthermore, physical models can be used to simulate extreme cases or rare events, which is difficult to achieve with purely data-driven methods.

[0178] Next, a gray-box modeling approach is adopted to integrate the data-driven model and the indoor space physical model into a comprehensive indoor environmental model for enclosed spaces. The gray-box model is a modeling method that combines the advantages of the white-box model (a model based on physical principles) and the black-box model (a purely data-driven model). This fusion approach aims to fully utilize the theoretical foundation of the physical model and the learning capabilities of the data-driven model, while simultaneously saving computational resources and improving computational efficiency. The core idea of ​​the gray-box model is to use a data-driven approach to compensate for or correct uncertainties and simplifying assumptions in the physical model. Specifically, the data-driven model can be used to predict the residuals (i.e., prediction errors) of the physical model. This method allows the physical model to capture major trends, while the data-driven model handles subtle adjustments and unmodeled effects.

[0179] The advantage of this gray-box model approach lies in its ability to more accurately describe complex indoor environmental dynamics by combining the theoretical foundation of a physical model with the learning capabilities of a data-driven model. The physical model provides an understanding of the system's fundamental behavior, enabling it to make reasonable predictions even in the face of new situations. The introduction of a physical model reduces reliance on large amounts of training data, especially in extreme or rare scenarios. The physical components of the model provide clear causal relationships, aiding in the understanding and interpretation of predictions. Through reasonable structural design and optimization strategies, the gray-box model can significantly improve computational efficiency while maintaining high accuracy. In practical applications, this gray-box model can predict changes in indoor environmental parameters based on the current state and future control strategies, providing a basis for optimized control. Furthermore, by comparing model predictions with actual measurements, it can detect system anomalies or equipment failures. It can also simulate energy consumption under different control strategies to find the optimal energy-saving solution, and by combining parameters such as temperature, humidity, and air quality, it can evaluate and optimize indoor comfort.

[0180] In another implementation, a data-driven model with physical constraints can be used for model fusion. Specifically, physical constraints are introduced during the training of the data-driven model to ensure that its output conforms to basic physical laws. For example, a physical constraint term can be added to the loss function: L_total = L_data + λ * L_physics, where L_data is the loss based on data (e.g., mean squared error), L_physics is the loss based on physical constraints (e.g., the degree of violation of the energy balance equation), and λ is the balance coefficient. In practice, the following steps can be taken: a) Initialization: Use the structure of the physical model as the initial architecture of the neural network and initialize the network weights with the parameters of the physical model. This ensures that the model has a reasonable physical basis in the early stages of training. b) Data preparation: Combine historical equipment control parameters, indoor environmental data, and outdoor environmental data into a training dataset. Ensure that the dataset covers various operating conditions and seasonal variations. c) Model training: Train the gray-box model using the prepared dataset. The training process can be divided into two stages: first, fix the physical model part and train only the data-driven part; then perform end-to-end fine-tuning, allowing the parameters of the physical model to be adjusted slightly. d) Online Learning: Design an online learning mechanism that allows the model to be continuously updated and optimized during actual operation. This can be achieved by periodically collecting new data and performing incremental learning. e) Model Evaluation: Evaluate the model performance using an independent test set. In addition to commonly used statistical metrics (such as RMSE, MAE), the model output should be checked to ensure it conforms to physical laws, such as energy conservation and mass conservation.

[0181] In one implementation, the gray-box model method is used to integrate the data association-driven model and the indoor space physical model into a comprehensive indoor environmental model of an enclosed indoor space, including the following steps:

[0182] For any indoor space physical model, introduce a new model error term into the indoor space physical model;

[0183] By combining the data association-driven model and the indoor space physical model, an initial comprehensive indoor environment model of the enclosed indoor space is obtained;

[0184] After preprocessing the historical equipment control parameters and historical spatial environment data, the historical equipment control parameters and historical spatial environment data are integrated into the model training set;

[0185] The model training set is input into the initial indoor integrated environment model. Based on historical equipment control parameters and through data association, the model outputs environmental correlation change results. Combining the environmental correlation change results and historical spatial environment data, the indoor spatial environment prediction data is output through the indoor spatial physical model.

[0186] The mean square error between the predicted indoor space environment data and the historical indoor environment data is used as the model loss function.

[0187] The model error term of the indoor space physics model and the model parameters of the data association-driven model are adjusted simultaneously based on the model loss function.

[0188] The initial indoor integrated environment model is continuously trained using the model training set until the model loss function reaches its minimum value, thus obtaining the trained indoor integrated environment model.

[0189] In this embodiment, for any indoor space physics model, a new model error term ε(t) is introduced to improve the model's flexibility and adaptability. These error terms ε(t) can be designed as functions of time, for example, using Fourier series expansion to represent periodic variations:

[0190] ε(t)=a0+∑(a n *cos(nωt)+b n *sin(nωt))

[0191] Where a0, a n b n It is a learnable parameter, and ω is the fundamental frequency (such as the cycle of a day). This representation allows the error term to capture changes at different time scales.

[0192] Introducing error terms increases the flexibility of the physical model, enabling it to adapt to complex real-world conditions. Secondly, by learning these error terms, potential flaws in the physical model can be identified, providing guidance for further improvements. Finally, this method preserves the interpretability of the physical model while improving its accuracy. In practical applications, analyzing the learned error terms can identify unknown factors or overlooked physical processes affecting the indoor environment, thereby continuously refining our understanding of indoor environmental dynamics.

[0193] The next step is to combine the flexibility of data-driven approaches with the theoretical foundation of physical models. Specifically, this involves creating a unified framework in which data-driven models and physical models can complement each other and work together. In practice, this can be achieved by using the following method: First, the output of the data-driven model (such as a CNN-LSTM network) can be used as input or parameters for the physical model. For example, the data-driven model can predict the dynamic changes in the thermal conductivity coefficient UA, and then this prediction can be input into the thermal balance equation of the physical model. Simultaneously, the intermediate states and outputs of the physical model can serve as additional inputs to the data-driven model, forming a closed-loop system.

[0194] This simultaneous equation can be represented as a set of coupled equations:

[0195] Y p =f p (X, θ) p Y NN )

[0196] Y NN =f NN (X, θ) NN Y p )

[0197] Y f =g(Y p Y NN )

[0198] Where X is the input variable (such as equipment control parameters, environmental conditions, etc.), Y p It is the output of the physical model, Y NN It is the output of the neural network model, θ p and θ NN These are the parameters of the physical model and the neural network model, respectively, f p and f NNLet represent the functions of the physical model and the neural network model, respectively, and g be the function that combines the outputs of the two models. A key advantage of this simultaneous approach is that it allows the model to flexibly rely on either physical principles or data-driven predictions in different situations. For example, in common cases where data is abundant, the model relies more on the data-driven part; while in cases where data is sparse or facing new situations, the model can rely more on physical principles. This adaptability greatly enhances the model's generalization performance and robustness.

[0199] Next, a high-quality, information-rich dataset needs to be created based on historical data to comprehensively reflect the dynamic characteristics of the indoor environment and the impact of control parameters. In practice, this involves first identifying and handling outliers, missing values, and noisy data. Outliers can be detected using statistical methods (such as Z-sCore or IQR), and then a decision can be made on whether to delete or replace them. Missing values ​​can be filled using interpolation methods (such as linear interpolation or more complex time-series interpolation). Then, ensure all data are on the same time scale and timestamps are aligned. This involves resampling or interpolating data at different sampling frequencies. Create new features or transform existing features to enhance the model's learning ability. For example, time-related features (such as time of day, day of the week, season, etc.) can be added, or lag features can be created to capture time dependencies. Transform features at different scales to the same scale so that the model can learn the importance of each feature fairly. Divide the processed dataset into training, validation, and test sets. A typical ratio is 70% / 15% / 15%, or a time-series splitting method can be used to simulate real-world application scenarios. Appropriate augmentation of the training data can improve the model's generalization ability. For example, small random perturbations can be added to simulate sensor noise.

[0200] The integrated model training set should include the following components:

[0201] Input features include pre-processed equipment control parameters (such as air conditioning set temperature, fan speed, etc.) and environmental parameters (such as outdoor temperature, humidity, solar radiation intensity, etc.).

[0202] Target variable: Preprocessed indoor environmental parameters (such as indoor temperature, humidity, CO2 concentration, etc.).

[0203] Time information: used to capture time-dependent and periodic patterns.

[0204] Additional engineering features: hysteresis features, statistical features, etc.

[0205] Preprocessing and integration improve data quality, eliminating noise and anomalies that can hinder model learning. Feature engineering and selection enhance the expressive power of the data, enabling the model to more easily capture underlying patterns and relationships. Standardization ensures that different features are treated fairly in the model. A reasonable data splitting strategy provides a reliable foundation for model training, validation, and testing.

[0206] Next, the model training set is input into the initial integrated indoor environment model, and calculations are performed using data association to drive the model and the indoor space physical model. This step realizes the forward propagation of the model, generates prediction results, and lays the foundation for subsequent error calculation and parameter optimization. The specific implementation process is as follows:

[0207] The preprocessed model training set is input into the initial indoor integrated environment model in batches. Each batch contains data at a series of time steps, for example, 24 hours can be selected as a sequence length. The input data X includes historical equipment control parameters (such as air conditioning set temperature, fan speed, etc.) and historical spatial environment data (such as outdoor temperature, humidity, etc.). The historical equipment control parameters are processed by a data association-driven model (such as a CNN-LSTM network), and the output is the result of environmental association changes. This process can be represented as: Y NN =f NN (X c θ N N). Among them, X c It is a device control parameter, Y NN This is the predicted outcome of environmental association changes. Then, the environmental association change outcome Y... NN Historical spatial environment data X enυ This is combined and used as input to the physical model of the interior space. This combination process can be represented as: X p =g(Y NN X enυ Here, g is a combination function, which can be a simple concatenation or a more complex fusion method. Then, using the prepared input data, the indoor space environment prediction data is calculated through the indoor space physical model. This process can be represented as: Y p =f p (X p θ p , ε). Where, θ p These are the parameters of the physical model, and ε is the model error term. Finally, the outputs of the data association-driven model and the indoor space physical model are fused to obtain the final indoor space environment prediction data. The fusion method can be a simple weighted average or a more complex nonlinear combination.

[0208] For each time step and for each predictor variable (such as temperature, humidity, CO2 concentration, etc.), calculate the difference between the predicted value and the actual observed value. Assume Y... pred It is the model prediction value, Y true If the actual observed value is used, then the error can be expressed as: e = Y pred -Y trye The mean squared error between the predicted and actual values ​​is obtained by squaring the errors at all time steps and for all predicted variables, and then averaging the results. To prevent overfitting, a regularization term can be added to the loss function. To ensure that the model's predictions conform to physical laws, a physical constraint term can be added to the loss function. Taking all these factors into account, the final loss function can be constructed.

[0209] Next, we need to calculate the gradient of the loss function L with respect to each parameter. This includes the gradient of the physical model error term ε and the gradient of the data association-driven model parameters θ. NN The gradient of . Using the chain rule, we can obtain:

[0210]

[0211] Then, gradient descent is used to update the parameters. The update rules for the physical model error term and the data-driven model parameters are as follows:

[0212]

[0213] Where, η ε and η NN This refers to the learning rate, which can be a fixed value or adaptive. ε1 and These are the updated physics model error term and the data-driven model parameters, respectively. To improve training efficiency and stability, adaptive learning rate methods, such as the Adam optimizer, can be used. When updating the physics model error term ε, it is necessary to ensure that the updated physics model still satisfies the basic physical laws.

[0214] An initial integrated indoor environment model is continuously trained using the model training set. By simultaneously adjusting the error term of the physical model and the parameters of the data-driven model, co-optimization of the physical model and the data-driven model is achieved. When the model loss function reaches its minimum value, the trained integrated indoor environment model is obtained. This method preserves the interpretability of the physical model while utilizing the flexibility and learning ability of the data-driven approach, greatly improving the overall efficiency of model training.

[0215] In one implementation, after preprocessing the historical equipment control parameters and historical spatial environment data, integrating the historical equipment control parameters and historical spatial environment data into a model training set includes the following steps:

[0216] Unify the data sampling rate of historical equipment control parameters and historical spatial environment data;

[0217] Align the historical equipment control parameter data at each time point with the historical spatial environment data at the previous time point;

[0218] The historical equipment control parameters and historical spatial environment data, after data alignment, are integrated into the model training set.

[0219] In this implementation, standardizing the sampling rate of historical equipment control parameters and historical spatial environment data is a crucial step in data preprocessing. Specifically, a target sampling rate needs to be determined first. This sampling rate selection should consider the characteristics of indoor environmental changes and the response time of the control system. Generally, a sampling interval of 5 to 15 minutes is suitable for building environment control. Let's assume a 10-minute sampling interval is chosen as the standard. For data with a sampling rate higher than the target sampling rate, downsampling is required using methods such as the average method, median method, and nearest neighbor method. For data with a sampling rate lower than the target sampling rate, interpolation is required. Commonly used interpolation methods include linear interpolation, spline interpolation, and nearest neighbor interpolation. Furthermore, it is necessary to handle any missing data. For short-term missing data (such as a single time point), interpolation methods can be used to fill in the gaps. For long-term missing data, estimation using historical data from the same period or statistical models can be considered. This step ensures the consistency of all data across time scales, which is crucial for subsequent data analysis and model training. A standardized sampling rate allows for direct comparison and correlation analysis of different types of data.

[0220] Next, the historical device control parameter data at each time point will be aligned with the historical spatial environment data at the previous time point. In practice, a new data structure needs to be created, where each data point contains the device control parameters at the current moment and the spatial environment data from the previous moment. Assume the following data sequence:

[0221]

[0222] The aligned data structure should be as follows:

[0223]

[0224] The specific steps for implementing this alignment process are as follows:

[0225] Create a new data frame containing the following: timestamps, all control parameters, and all environmental parameters (suffixed with "_prev"). For each time point t(i), copy all control parameter values ​​at time t(i) to the new data frame, copy all environmental parameter values ​​at time t(i-1) to the new data frame, add the "_prev" suffix, and set the timestamp to t(i). For the first time point, since there is no environmental data from the previous time point, the environmental data from the first time point can be used as the initial value. Data alignment establishes the temporal relationship between control operations and environmental changes, enabling the model to learn the impact of control operations on the environment. This alignment method also considers the lag in indoor environmental changes, which is more consistent with actual physical processes. For example, after adjusting the air conditioner's set temperature, the indoor temperature does not change immediately, but the effect will only be apparent after a certain period of time.

[0226] Next, the aligned historical equipment control parameters and historical spatial environment data are integrated into the model training set. In practice, the first step is to design a data structure suitable for the machine learning model input. Typically, this structure can be a two-dimensional table, where each row represents a time point, and columns include a timestamp, all equipment control parameters at the current moment (e.g., air conditioning set temperature, fan speed, fresh air volume), all spatial environment data from the previous moment (e.g., indoor temperature, humidity, CO2 concentration), outdoor environment data at the current moment (e.g., outdoor temperature, humidity, solar radiation intensity), and other relevant features (e.g., time features, building features).

[0227] Then, features at different scales are transformed to the same scale so that the model can fairly learn the importance of each feature. The integrated dataset is then divided into training, validation, and test sets. Commonly used split ratios are 60%:20%:20% or 70%:15%:15%. For time series data, a time-order split is typically used to simulate real-world prediction scenarios. The processed data is then converted to a format directly usable by machine learning frameworks (such as PyTorch and TensorFlow), usually in tensor format.

[0228] In one implementation, using a comprehensive objective function as the algorithm optimization objective, the initial equipment control strategy for the environmental control equipment is generated based on an indoor comprehensive environment model and an ant colony optimization algorithm, including the following steps:

[0229] Initialize the algorithm parameters of the ant colony optimization algorithm;

[0230] Based on constraints, the positions of multiple ants are randomly generated in the state space, and different ant positions represent different device control parameters.

[0231] For each ant's location, the corresponding comprehensive objective function value is calculated using an indoor integrated environment model;

[0232] Update the pheromone concentration in the state space, and update the ant position by combining the comprehensive objective function value and the updated pheromone concentration;

[0233] Repeat the above steps of calculating the comprehensive objective function value and updating the ant position until the maximum number of iterations of the ant colony optimization algorithm is reached, and output the optimal ant path under the constraints.

[0234] The optimal parameter range of each control parameter of the environmental control equipment is determined based on the optimal ant path, and the optimal parameter range of each control parameter of the environmental control equipment is integrated into the initial equipment control strategy.

[0235] In this implementation, the ant colony optimization algorithm needs to initialize the following key parameters before execution:

[0236] Number of ants (m): This determines the number of ants exploring the solution space simultaneously in each iteration. It is typically set between 20 and 100, adjusted according to the problem size and complexity.

[0237] Maximum number of iterations (Max_Iter): Sets one of the conditions for the algorithm to stop. It is usually set between 100 and 1000, depending on the complexity of the problem and the expected accuracy.

[0238] Pheromone importance factor (α): controls the degree of influence of pheromones on ants' path selection, and is usually set between 1 and 5.

[0239] Heuristic factor importance (β): controls the degree of influence of heuristic information on the ant's path selection, usually set between 1 and 5.

[0240] Pheromone evaporation coefficient (ρ): Controls the evaporation rate of pheromones, typically set between 0.1 and 0.9. Smaller values ​​accelerate convergence but may lead to local optima; larger values ​​aid global search but may slow convergence.

[0241] Initial pheromone concentration (τ0): Sets the initial pheromone concentration in the state space, usually set to a small positive number, such as 0.1.

[0242] Global optimal solution update weight (Q): The amount of pheromone increase used when updating the global optimal solution, usually set to 1.

[0243] The initialization of these parameters can be expressed as:

[0244] params={'m':50,'Max_Iter':500,'alpha':2,'beta':3,'rho':0.5,'tau0':0.1,'Q':1}

[0245] The specific selection of parameters needs to be adjusted according to the characteristics of the problem. For example, for multidimensional optimization problems such as indoor environmental control, a larger number of ants and more iterations may be needed to ensure sufficient exploration of the solution space. In addition, it is necessary to initialize problem-specific parameters, such as the upper and lower bounds of control variables and constraints. For example, for air conditioning temperature setting: temp_bounds = [18, 28] # Temperature setting range (degrees Celsius).

[0246] Then, based on the constraints, multiple ant positions are randomly generated in the state space. Specifically, different ant positions represent different device control parameters. The dimensions and range of the state space are determined according to the characteristics of the problem. For indoor environmental control, the state space may include multiple dimensions such as temperature setting, humidity setting, wind speed, and fresh air volume. Physical limitations of the device and environment are considered, such as temperature setting range, humidity control capability, and energy consumption limitations. These constraints will limit the feasible range of ant positions. Under the condition that the constraints are met, a random number generator is used to generate an initial position for each ant.

[0247] The position of each ant (i.e., a specific set of equipment control parameters) is input into a pre-trained indoor integrated environment model to calculate the corresponding environmental state and performance indicators. These indicators are then integrated into a single evaluation value through a comprehensive objective function. Specifically, for each ant i, its position is represented by a set of control parameters X_i = [T_i, RH_i, V_i, Q_i, ...], and the calculation process is as follows:

[0248] Predicting environmental conditions using an integrated indoor environmental model:

[0249] [T_indoor,RH_indoor,CO2]=IndoorModel(X_i,Outdoor_Conditions)

[0250] Calculate the various performance indicators:

[0251] PMV=f_PMV(T_indoor, RH_indoor, V_i,...)

[0252] PPD = f_PPD(PMV)

[0253] Energy=f_Energy(X_i, T_indoor, RH_indoor,...)

[0254] IAQ = f_IAQ(CO2)

[0255] Calculate the value of the integrated objective function:

[0256] O=w1*abs(PMV)+w2*PPD+w3*Energy+w4*IAQ

[0257] Among them, w1, w2, w3, and w4 are the weights of each indicator, which can be adjusted according to specific needs.

[0258] By using a comprehensive environmental model, the indoor environmental state under different control parameters can be accurately predicted, avoiding the cost and time consumption of actual experiments. This greatly improves the efficiency and feasibility of the optimization process. The use of a comprehensive objective function allows for trade-offs among multiple objectives such as comfort, energy consumption, and air quality. By adjusting the weights, it can flexibly adapt to different optimization needs. This step provides the necessary foundation for subsequent pheromone updates and ant location updates.

[0259] Next, the pheromone concentration in the state space needs to be updated, and the ant positions are updated by combining the comprehensive objective function value and the updated pheromone concentration. Specifically, the pheromone concentration is updated based on the path quality of each ant (i.e., the comprehensive objective function value). A better path quality results in a greater increase in pheromone. During the update process, the pheromone volatilization phenomenon in nature needs to be simulated to prevent the algorithm from prematurely converging to a local optimum. Finally, based on the updated pheromone concentration and heuristic information, new ant positions are generated. The steps of calculating the comprehensive objective function value and updating the ant positions are repeated to continuously improve the quality of the solution until the maximum number of iterations of the ant colony optimization algorithm is reached, converging to a near-optimal solution. The optimal ant path under the constraints is then output, thus finding the best combination of control parameters that balances comfort, energy consumption, and air quality.

[0260] A detailed analysis of the control parameter combinations represented by the optimal ant path is conducted. The influence of each control parameter on the objective function is evaluated. Based on the optimal solution and sensitivity analysis results, a suitable interval is defined for each control parameter. It is ensured that the parameter interval satisfies all necessary constraints. This ultimately transforms the optimization algorithm results into practical control parameter intervals, bridging the gap between theoretical optimization and practical applications.

[0261] In one implementation, the optimal parameter range is used as the initial solution of the model predictive control strategy, and the integrated objective function is used as the strategy optimization objective of the model predictive control strategy. The generation of strategy prediction results based on the indoor integrated environment model and through the model predictive control strategy includes the following steps:

[0262] The optimal parameter range is used as the initial solution for the model predictive control strategy.

[0263] Based on the initial solution of the strategy and through the indoor integrated environment model, the prediction time domain and control time domain of the model predictive control strategy are determined;

[0264] Based on the prediction time domain, the comprehensive objective function is used as the policy optimization objective of the model predictive control strategy. The control sequence in the control time domain is continuously optimized using the sequential quadratic programming method until the comprehensive objective function reaches its minimum value, thus obtaining the optimal control sequence. The optimal control sequence is then used as the policy prediction result generated by the model predictive control strategy.

[0265] In this implementation, the optimal parameter range is obtained from the initial device control strategy generated by the preceding ant colony optimization algorithm. Specifically, for each control parameter of the environmental control device, there is a corresponding optimal parameter range. For example, for air conditioning temperature setting, the optimal parameter range might be [22℃, 24℃]; for humidity control, it might be [40%, 60%]; and for lighting brightness, it might be [300 lux, 500 lux]. Using these parameter ranges as initial solutions provides a near-optimal starting point for the model predictive control strategy. In practical implementation, initial values ​​also need to be selected from the initial solution of the strategy. The simplest method is to select the median value of each parameter range as the initial value. For multiple parameters, they can be represented in vector form. In practical applications, weighting factors can also be introduced to adjust the importance of different parameters.

[0266] Next, based on the initial solution of the strategy and using the indoor integrated environment model, it is necessary to determine the prediction horizon and control horizon of the model predictive control strategy. The prediction horizon refers to the length of time the model predicts forward, while the control horizon refers to the length of time during which the control input is actually optimized. The process of determining these two horizons first involves simulating the system response under different control inputs using the indoor integrated environment model. Based on the initial solution, multiple simulations are performed to observe the time required for the system to reach steady state; this time can serve as a reference for the prediction horizon. Generally, the prediction horizon should be long enough to capture the dynamic characteristics of the system, but it should not be too long to avoid excessive computational burden. The control horizon is usually shorter than the prediction horizon, and can be set to 1 / 3 to 1 / 2 of the prediction horizon. The selection of the control horizon requires a balance between control flexibility and computational complexity. A longer control horizon can provide more optimization freedom but also increases the computational burden; a shorter control horizon has less computation but may result in less smooth control. Specifically, the following formula can be used: T p =k*τ,T c =m*τ. Where T p It is the prediction time domain, T cThis is the control time domain, where τ is the characteristic time constant of the system (which can be obtained through step response experiments), and k and m are adjustment coefficients, usually k > m.

[0267] For indoor environmental control, τ can range from 10 to 30 minutes, depending on factors such as room size and insulation performance. k can be chosen between 3 and 5, and m can be chosen between 1 and 2. For example, if τ = 20 minutes, k = 4, and m = 1.5, then T_p = 80 minutes and T_c = 30 minutes. In practical applications, adaptive methods can also be used to dynamically adjust the time domain. For example, the values ​​of k and m can be adjusted based on changes in system response speed and external disturbances. When the system changes slowly, the values ​​of k and m can be increased to reduce the calculation frequency; when the system changes drastically, the values ​​of k and m can be decreased to improve control accuracy.

[0268] Next, based on the prediction time domain, the comprehensive objective function is used as the policy optimization objective of the model predictive control strategy, and the control sequence in the control time domain is optimized using the sequential quadratic programming method. The sequential quadratic programming (SQP) method transforms the nonlinear optimization problem into a series of quadratic programming subproblems. In each control cycle, the SQP algorithm first linearizes the system model, and then solves the quadratic programming problem to obtain the optimal control sequence. This process is repeated until the convergence condition or the maximum number of iterations is reached. In implementation, the following iterative formula can be used: u k +1=u k +α k *d k Among them, u k It is the control sequence for the k-th iteration, d k It is the search direction, α k It's the step size. After each iteration, the system state is updated, and the optimization window is scrolled forward one step.

[0269] The specific process is as follows: Using the initial solution as the starting point, the indoor integrated environment model is linearized near the current working point. Based on the linearized model and the integrated objective function, a quadratic programming problem is constructed. The optimal control increment d is solved using a quadratic programming solver. kThe control sequence is updated using the iterative formula described above. The process stops if the change in the objective function value is less than a preset threshold or the maximum number of iterations is reached. The first control action in the optimized control sequence is applied to the system. The time window is moved forward one step to obtain the new system state, achieving rolling optimization. This rolling optimization method can adapt to dynamic changes in the system and external disturbances, improving the robustness of control. For example, if a sudden increase in outdoor temperature is detected during optimization, the algorithm can quickly adjust the control strategy in the next optimization cycle, increasing cooling output in advance to maintain indoor comfort. The final optimal control sequence is the output of the model predictive control strategy, which considers both the current state and anticipates possible future changes, thus achieving more intelligent and efficient environmental control.

[0270] In one implementation, updating the algorithm parameters of the ant colony optimization algorithm based on the strategy prediction results, and using the updated ant colony optimization algorithm to generate the optimal device control strategy for the environmental control device includes the following steps:

[0271] Update the pheromone distribution of the ant colony optimization algorithm based on the strategy prediction results;

[0272] Generate the target constraints for the ant colony optimization algorithm based on the optimal parameter range;

[0273] Using the comprehensive objective function as the algorithm's optimization objective, and combining the state space and objective constraints, the optimal equipment control strategy for the environmental control equipment is generated using the ant colony optimization algorithm.

[0274] In this implementation, pheromones play a crucial role in the ant colony algorithm, representing the attractiveness of a path. Updating the pheromone distribution essentially involves adjusting the importance of different paths in the search space. Specifically, the policy prediction results are first mapped to the ant colony algorithm's search space. This can be achieved by establishing a correspondence between the prediction results and the ant colony search space. For example, if the prediction indicates a better temperature range, the pheromone concentration on the path corresponding to that temperature range is increased. Secondly, the pheromone concentration is adjusted using the pheromone update formula: τij(t+1)=(1-ρ)τij(t)+Δτij, where τij is the pheromone concentration on path (i,j), ρ is the pheromone evaporation coefficient (typically between 0.1 and 0.5), and Δτij is the pheromone increment. Δτij can be set according to the quality of the prediction results; for example, it can be set as a function of the improvement in the prediction results: Δτij=f(ΔJ), where ΔJ is the improvement in the objective function. Finally, to prevent the algorithm from prematurely converging to a local optimum, upper and lower limits for the pheromone can be set: τmin ≤ τij ≤ τmax. This process integrates advanced model predictive control knowledge into the ant colony algorithm, accelerating convergence and improving search efficiency. It also enhances the algorithm's adaptability, enabling it to better cope with dynamically changing environmental control requirements.

[0275] Next, the target constraints for the ant colony optimization algorithm are generated based on the optimal parameter interval. In practice, for each control parameter i, the constraint is defined as: Li ≤ xi ≤ Ui, where Li and Ui are the lower and upper limits of the optimal parameter interval corresponding to parameter i, respectively. These constraints are then combined into a constraint vector: C = [L1, U1, L2, U2, ..., Ln, Un]. In the implementation of the ant colony algorithm, whenever a new solution is generated, it is necessary to check whether these constraints are satisfied. A penalty function can be defined to handle cases where the constraints are not satisfied: P(x) = ∑max(0, Li-xi, xi-Ui), where x is the current solution. This penalty function is added to the objective function: J′(x) = J(x) + λP(x), where J(x) is the original objective function and λ is the penalty factor. By adjusting the value of λ, the strictness of the constraints can be controlled. Finally, using the comprehensive objective function as the algorithm's optimization objective, and combining the state space and objective constraints, the optimal equipment control strategy for the environmental control device is generated using the ant colony optimization algorithm. The execution process of the ant colony optimization algorithm in this step can be referenced from the aforementioned implementation method. The final generated optimal equipment control strategy includes the optimal control parameters for all equipment control parameters within the optimal parameter range.

[0276] The present invention also discloses an indoor space environment control system based on an optimization algorithm, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the indoor space environment control method based on the optimization algorithm as described in any of the above embodiments.

[0277] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it in this regard.

[0278] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0279] This invention also discloses an environmental control device, which includes the indoor space environmental control system, air conditioning unit, lighting unit, ventilation unit, etc., disclosed in the above embodiments. (See also...) Figure 2 The environmental control unit (i.e., environmental control equipment) includes an environmental control controller (i.e., an indoor space environmental control system). The environmental control equipment connects to the distribution box (input power DC 600V) and the mains power (input power AC 380V) via two interfaces, ST1X1 and ST1X2, respectively. The socket models used for these two interfaces are XC158 / 24F4ZTP40 and XC158 / 22FSZTP40. This dual-power supply design is likely to ensure stable system operation and reliable power supply. The environmental control equipment communicates with the host computer via two CAN interfaces, ST1X3 and ST1X4 (socket model XC158 / 22F10KTP40).

[0280] In addition, the environmental control equipment is equipped with a connecting cable interface ST1X6 (socket model XCE18F10KTP40) and a one-way valve interface ST1X7 (socket model XCE14F4KTP40). The ST1X6 cable interface allows connection to different unit equipment; for example, the ST1X6 cable interface can connect to the fan unit of an air conditioning unit. The environmental control equipment also includes a series of sensor terminals (STW1 to STW1-5), through which sensors connected to the terminals can detect data such as temperature, humidity, air pressure, and air quality.

[0281] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0282] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for controlling an indoor space environment based on an optimization algorithm, characterized by, The method comprises the following steps: obtaining historical device control parameters of an environment control device and historical space environment data corresponding to the historical device control parameters, the historical space environment data comprising historical indoor environment data inside an indoor enclosed space in which the environment control device is deployed and historical outdoor environment data outside the indoor enclosed space; constructing a data correlation driven model between the historical device control parameters and the historical space environment data based on a convolutional neural network and a long short-term memory network; constructing an indoor space physical model of the indoor enclosed space in combination with the historical indoor environment data and the historical outdoor environment data, the indoor space physical model comprising a thermal dynamics model, a humidity model, an air quality model and a lighting model; fusing the data correlation driven model and the indoor space physical model into an indoor comprehensive environment model of the indoor enclosed space by using a grey-box model method; constructing a comprehensive objective function according to an indoor environment comfort degree of the indoor enclosed space and a device energy consumption index of the environment control device; creating a state space of an ant colony optimization algorithm based on a device control parameter list of the environment control device, and generating constraint conditions of the ant colony optimization algorithm according to parameter limit values of each device control parameter in the device control parameter list; generating an initial device control strategy of the environment control device based on the indoor comprehensive environment model and by using the ant colony optimization algorithm, the initial device control strategy comprising an optimal parameter interval of each device control parameter, with the comprehensive objective function serving as an algorithm optimization target; taking the optimal parameter interval as an initial solution of a model predictive control strategy, taking the comprehensive objective function as a strategy optimization target of the model predictive control strategy, and generating a strategy prediction result based on the indoor comprehensive environment model and by using the model predictive control strategy; updating algorithm parameters of the ant colony optimization algorithm according to the strategy prediction result, and generating an optimal device control strategy of the environment control device by using the updated ant colony optimization algorithm.

2. The optimization algorithm based indoor space environment control method according to claim 1, wherein, The formula of the thermal dynamics model is as follows: ， wherein: represents the indoor ambient temperature of the indoor enclosed space, represents the heat capacity of the indoor air, represents the heat provided by the environmental control device, represents the outdoor solar radiation heat, represents the heat generated by the indoor heat source, represents the overall heat transfer coefficient of the building in which the indoor enclosed space is located, represents the outdoor ambient temperature; The formula of the humidity model is as follows: ， wherein: represents an indoor absolute humidity of the indoor enclosed space, represents an indoor volume of the indoor enclosed space, represents a supply air coverage of the environmental control device, represents a supply air absolute humidity, represents a return air volume of the environmental control device, represents a permeation air volume, represents an outdoor absolute humidity, represents an indoor humidity generation rate; The formula of the air quality model is as follows: ， wherein: represents the CO2 concentration of the indoor enclosed space, represents the CO2 generation rate of the indoor enclosed space, represents the ventilation rate of the indoor enclosed space, represents the outdoor CO2 concentration; The formula of the lighting model is as follows: ， wherein: E represents the indoor illuminance of the indoor enclosed space, L represents the artificial lighting intensity controlled by the environmental control device, represents the outdoor solar irradiance, and respectively represent the equipment light influence coefficient and the natural light influence coefficient, represents the adjustment coefficient.

3. The optimization algorithm based indoor space environment control method according to claim 2, wherein, The step of fusing the data correlation driven model and the indoor space physical model into the indoor comprehensive environment model of the indoor enclosed space by using the grey-box model method comprises the following steps: For any one of the indoor space physical models, a new model error term is introduced for the indoor space physical model; the data correlation driven model and the indoor space physical model are solved to obtain an initial indoor comprehensive environment model of the indoor enclosed space; after preprocessing the historical device control parameters and the historical space environment data, the historical device control parameters and the historical space environment data are integrated into a model training set; inputting the model training set into the initial indoor comprehensive environment model, outputting environment correlation change results based on the historical device control parameters and by driving the data correlation model, combining the environment correlation change results and the historical space environment data and outputting indoor space environment prediction data by the indoor space physical model; taking the mean square error between the indoor space environment prediction data and the historical indoor environment data as a model loss function; simultaneously adjusting the model error term of the indoor space physical model and the model parameters of the data correlation driven model according to the model loss function; continuously training the initial indoor comprehensive environment model by using the model training set until the model loss function reaches a minimum value, to obtain a trained indoor comprehensive environment model.

4. The optimization algorithm based indoor space environment control method according to claim 3, wherein, After preprocessing the historical device control parameters and the historical space environment data, integrating the historical device control parameters and the historical space environment data into a model training set includes the following steps: unifying the data sampling rates of the historical device control parameters and the historical space environment data; aligning the data of the historical device control parameters in each time node with the data of the historical space environment data in the previous time node; integrating the historical device control parameters and the historical space environment data after data alignment into a model training set.

5. The optimization algorithm based indoor space environment control method of claim 1, wherein, Taking the comprehensive target function as an algorithm optimization target, generating an initial device control strategy of the environment control device based on the indoor comprehensive environment model and using the ant colony optimization algorithm includes the following steps: initializing algorithm parameters of the ant colony optimization algorithm; randomly generating ant positions of multiple ants in the state space based on the constraint conditions, different ant positions representing different device control parameters; for each ant position, calculating a corresponding comprehensive target function value by using the indoor comprehensive environment model; updating pheromone concentration in the state space, and updating the ant position in combination with the comprehensive target function value and the updated pheromone concentration; repeating the calculation step of the comprehensive target function value and the updating step of the ant position until a maximum iteration number of the ant colony optimization algorithm is reached, and outputting an optimal ant path under the constraint condition; determining optimal parameter intervals of each device control parameter of the environment control device according to the optimal ant path, and integrating the optimal parameter intervals of each device control parameter of the environment control device into an initial device control strategy.

6. The optimization algorithm based indoor space environment control method according to claim 5, wherein, Taking the optimal parameter intervals as a strategy initial solution of the model predictive control strategy, taking the comprehensive target function as a strategy optimization target of the model predictive control strategy, and generating a strategy prediction result by the model predictive control strategy based on the indoor comprehensive environment model include the following steps: taking the optimal parameter intervals as a strategy initial solution of the model predictive control strategy; determining a prediction time domain and a control time domain of the model predictive control strategy based on the strategy initial solution and by the indoor comprehensive environment model; On the basis of the prediction time domain, the integrated objective function is taken as a strategy optimization target of the model predictive control strategy, a sequence quadratic programming method is used to rollingly optimize a control sequence in the control time domain until the integrated objective function takes a minimum value, an optimal control sequence is obtained, and the optimal control sequence is taken as a strategy prediction result generated by the model predictive control strategy.

7. The optimization algorithm based indoor space environment control method according to claim 6, wherein, The step of updating the algorithm parameters of the ant colony optimization algorithm according to the strategy prediction result and generating the optimal device control strategy of the environment control device by using the updated ant colony optimization algorithm comprises the following steps: updating pheromone distribution of the ant colony optimization algorithm according to the strategy prediction result; generating a target constraint condition of the ant colony optimization algorithm according to the optimal parameter interval; taking the integrated objective function as an algorithm optimization target, combining the state space and the target constraint condition, and generating the optimal device control strategy of the environment control device by using the ant colony optimization algorithm.

8. An indoor space environment control system based on optimization algorithm, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the indoor space environment control method based on an optimization algorithm as claimed in any one of claims 1 to 7 when executing the computer program.

9. An environmental control device, characterized by The indoor space environment control system based on an optimization algorithm as claimed in claim 8. The indoor space environment control system based on an optimization algorithm as claimed in claim 8.

Images