Central air conditioner energy-saving management system based on data analysis

Abstract

The invention relates to the technical field of air conditioner energy-saving management, and discloses a central air conditioner energy-saving management system based on data analysis, and the system comprises a system modeling and data collection module which is used for building an air conditioner physical model and collecting room temperature, humidity and wind speed data; the online depth prediction module is used for calculating a future state and correcting a prediction error by using a deep neural network; the robust adaptive model predictive control module is used for constructing an optimization control problem with an energy consumption index and a state tracking index and solving control input; and the performance index and switching control module is used for designing control laws for different working modes and proving the stability of the system by a Lyapunov function. According to the method, the data analysis and online depth prediction technology is adopted, the air conditioner state is pre-judged in real time, the precise regulation and control effect is achieved, and compared with an existing single model prediction scheme, the problems of large prediction errors and response lag are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of air - conditioning energy management, and particularly to a central air - conditioning energy management system based on data analysis. Background Art

[0002] With the wide application of modern air - conditioning systems, how to efficiently and energy - savingly manage the operation of air - conditioners has become an urgent problem to be solved. Traditional air - conditioning energy management solutions mainly rely on simple temperature control or operation modes based on preset rules. In practical applications, traditional methods cannot adapt to environmental changes and fluctuations in air - conditioning loads in real time, resulting in high energy consumption and limited effects. Therefore, how to achieve precise and efficient central air - conditioning energy management has become an important direction in current technology research and development.

[0003] In the prior art, most air - conditioning systems use methods based on prediction models to adjust temperature and energy consumption, but these methods have deficiencies. Traditional model - predictive control techniques generally rely on fixed physical models and have weak capabilities to cope with external disturbances, and cannot dynamically adjust control strategies. Therefore, traditional methods cannot effectively respond to complex changes in the operating environment of air - conditioning systems, resulting in unsatisfactory energy - saving effects.

[0004] Most traditional air - conditioning energy management systems use static prediction models or rule - based adjustments based on historical data, and cannot effectively respond to real - time environmental changes and load fluctuations. Existing single - model - based prediction technologies are prone to large prediction errors in complex environments and have slow response speeds. Due to changes in environmental factors such as indoor temperature, humidity, and wind speed, the adjustment effects and real - time performance of traditional methods cannot meet actual requirements.

[0005] Existing static control methods usually lack sensitivity to external environmental changes, especially when dealing with external disturbances, they show poor response capabilities. Therefore, the prior art fails to effectively utilize real - time data and external disturbance information for dynamic optimization control, resulting in energy waste.

[0006] Many existing air - conditioning management systems rely on the adjustment of a single area and lack coordinated optimization control among multiple areas. For example, when different temperature and humidity requirements exist in multiple rooms or areas, traditional systems usually only perform independent control on a single area and cannot perform coordinated optimization according to overall requirements, resulting in insufficient energy utilization and limited energy - saving effects.

[0007] Therefore, the present invention proposes a central air - conditioning energy management system based on data analysis to solve the above - mentioned problems. Summary of the Invention

[0008] Aiming at the deficiencies of the prior art, the present invention provides a central air - conditioning energy management system based on data analysis to solve the problems raised in the above - mentioned background art.

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions: A central air-conditioning energy-saving management system based on data analysis, the central air-conditioning energy-saving management system includes:

[0010] The system modeling and data acquisition module is used to establish an air-conditioning physical model and collect room temperature, humidity, and wind speed data;

[0011] The online deep prediction module is used to calculate the future state and correct the prediction error by using a deep neural network;

[0012] The robust adaptive model predictive control module is used to construct an optimization control problem with energy consumption indicators and state tracking indicators and solve the control input;

[0013] The performance index and switching control module is used to design control laws for different working modes and prove the stability of the system with a Lyapunov function;

[0014] The distributed cooperative optimization design module is used to realize the optimization of multi-region local control laws and global coordination control;

[0015] The integrated controller implementation process module is used to integrate data transfer, control instruction generation, and feedback update of each module to form a closed-loop control.

[0016] Preferably, in the system modeling and data acquisition module, based on the dynamic characteristics of the central air-conditioning system operation state and the energy consumption optimization requirements, it further includes:

[0017] The physical modeling unit constructs a mathematical model describing the time variation of room temperature, humidity, and wind speed according to the thermodynamic characteristics and fluid dynamics characteristics of the central air-conditioning system, and defines system state variables, control inputs, disturbance terms, and uncertainties. The algorithm formula is:

[0018]

[0019] Among them, E1(t) is the system state vector, V1(t) is the control input vector, Φ1(·) is the system dynamic function, Γ1(·) is the control coupling matrix, Δ1(·) is the uncertainty term, and Ω1(t) is the external disturbance term;

[0020] The system state model output by the physical modeling unit provides variable definitions for the data acquisition unit and provides state evolution constraint conditions for the data preprocessing unit;

[0021] The data acquisition unit obtains the real-time operation data of the central air-conditioning system through sensor devices and converts the collected data into a format available for the mathematical model. Let the original data measured by the sensor be Y2(t). Therefore, the converted format of the system state variables is:

[0022] E2(t)=Ψ2(Y2(t))+ε2(t),

[0023] Among them, E2(t) is the converted system state variable, Y2(t) is the original sensor data, Ψ2(·) is the data conversion function, and ε2(t) is the measurement error;

[0024] The output data of the data acquisition unit enters the data preprocessing unit for filtering and abnormality detection.

[0025] Preferably, the system modeling and data acquisition module further includes, based on the dynamic characteristics of the central air-conditioning system operation state and the energy consumption optimization requirements:

[0026] The data preprocessing unit uses filtering, anomaly detection and missing data completion algorithms to improve data quality and provide data input to the subsequent online depth prediction module;

[0027] The data filtering uses weighted moving average filtering to process data noise:

[0028]

[0029] Among them, E3(t) is the state data after filtering, w 3,k is the weighting coefficient, N is the time window length;

[0030] The anomaly detection sets an anomaly detection threshold θ3, if the data satisfies:

[0031] ||E3(t)-E3(t-1)||>Θ3, the data is marked as abnormal data and removed,

[0032] Among them, Θ3 is the anomaly detection threshold;

[0033] The missing data is completed by using the nearest neighbor interpolation method:

[0034]

[0035] Among them, K is the number of nearest neighbor samples, t i is the most recent non-missing data time point;

[0036] The output E3(t) of the data preprocessing unit is used as input data of the online depth prediction module;

[0037] The state mapping unit combines the physical model with the preprocessed data to construct a standardized state variable to match the input format of the online depth prediction module. The state mapping function is:

[0038] E4(t)=M4(E3(t)),

[0039] Among them, E4(t) is the standardized state data, and M4(·) is the standardization mapping function;

[0040] The standardized data E4(t) output by the state mapping unit is directly used in the online depth prediction module to achieve future state prediction.

[0041] Preferably, in the online depth prediction module, based on the standardized state data provided by the system modeling and data acquisition module and the prediction accuracy optimization requirements, it further includes:

[0042] A data input unit that processes the standardized state data E4(t) provided by the system modeling and data acquisition module and converts it into the input format of the deep neural network model. The input format is converted to:

[0043] X5(t) = F5(E4(t)),

[0044] where X5(t) is the neural network input data vector, E4(t) is the standardized state data, and F5(·) is the data preprocessing function;

[0045] The output X5(t) of the data input unit is used as the input of the deep learning prediction unit;

[0046] A deep learning prediction unit that uses a deep neural network to predict the future state of the central air-conditioning system and calculates the state at the future τ moment based on the current state data X5(t) The prediction model is:

[0047]

[0048] Among them, is the predicted future state, N5(·) is the deep neural network prediction model, and θ5 is the parameter of the neural network;

[0049] The output of the deep learning prediction unit is used as the input of the prediction error correction unit;

[0050] A prediction error correction unit that calculates the prediction error and adjusts the neural network model parameters through an adaptive update mechanism. The error calculation formula is:

[0051]

[0052] where E6(t + τ) is the prediction error, X5(t + τ) is the true state, is the predicted future state;

[0053] For parameter update, online gradient adjustment is performed using an adaptive learning rate η6 to update the network parameter θ5:

[0054]

[0055] Among them, L(E6) is the loss function, is the gradient of the loss function with respect to the neural network parameters, and η6 is the adaptive learning rate;

[0056] The updated parameter θ5 output by the prediction error correction unit is fed back to the deep learning prediction unit to form an adaptive prediction optimization closed loop.

[0057] Preferably, in the robust adaptive model predictive control module, based on the future state prediction result output by the online deep prediction module and the system energy consumption and state tracking requirements, it further includes:

[0058] An optimization problem construction unit, according to the future state prediction result provided by the online deep prediction module and the expected state X ref (t + τ), constructs an optimal control problem including energy consumption index and state tracking index, establishes system dynamic constraints and control input constraints, and constructs the optimization objective:

[0059]

[0060] Among them, J is the control optimization objective, N p is the prediction horizon length, X(t + k + 1) is the system state at time t + k + 1, X ref (t + k + 1) is the desired state, u(t + k) is the control input, λ1 is the weight of the state tracking index, and λ2 is the weight of the energy consumption index;

[0061] The optimization problem construction unit outputs the optimization objective J and related constraint conditions for the control solution unit to solve the control input.

[0062] Preferably, in the robust adaptive model predictive control module, based on the future state prediction result output by the online deep prediction module and the system energy consumption and state tracking requirements, it further includes:

[0063] A control solution unit, based on the optimization objective and constraint conditions output by the optimization problem construction unit, uses an online solution algorithm to solve the optimal control sequence u * (t) in real time, and feeds back the first control input to the execution unit, and the dynamic constraint description:

[0064] X(t + k + 1) = f(X(t + k), u(t + k)) + δ(t + k),

[0065] Among them, f(·) is the dynamic function constructed by the physical modeling unit, and δ(t + k) is the modeling error and external disturbance;

[0066] The control solution unit updates the control input using the iterative gradient descent algorithm, and the calculation formula is as follows:

[0067]

[0068] where η is the online update learning rate, is the gradient of the control input corresponding to the optimization objective J;

[0069] The control solution unit outputs the first element of the optimal control sequence u * (t) as the current control input, which is fed back to the system execution unit to achieve actual control;

[0070] The robustness adjustment unit adjusts the dynamic constraint parameters and the control input update rate through error analysis based on the optimal control input solved by the control solution unit and the actual operation feedback of the system, so as to achieve control robustness and adaptive performance. The robustness compensation formula is:

[0071]

[0072] where Δu(t) is the compensation control input, K is the compensation coefficient, X(t) is the actual system state, is the predicted state output by the online depth prediction module;

[0073] The robustness adjustment unit combines the compensation control input Δu(t) with the optimal control input u * (t) to form the final control input u(t): u(t) = u * (t) + Δu(t), ensuring that the system can operate stably under the action of disturbances and model errors.

[0074] Preferably, in the performance index and the switching control module, based on the control input, the state tracking error, and the system stability requirements output by the robust adaptive model predictive control module, it further includes:

[0075] The performance index construction unit constructs a performance index function J ref according to the state tracking error e(t) = X(t) - X p (t) and the control input u(t) output by the robust adaptive model predictive control module. The construction formula is:

[0076] J p = e(t) T Qe(t) + u(t) T Ru(t),

[0077] Among them, e(t) is the state tracking error, Q is the state tracking weight matrix, u(t) is the output control input of the robust adaptive model predictive control module, R is the control energy consumption weight matrix, X(t) is the actual system state provided by the physical modeling unit, and X ref (t) is the reference state output by the online depth prediction module after data normalization, and J p is the performance index function;

[0078] The performance index construction unit outputs the performance index J p to provide a quantization basis for the switching control law design unit;

[0079] The switching control law design unit designs the control law u p under different working modes according to the performance index J output by the performance index construction unit and the state tracking error e(t). The design process includes constructing a Lyapunov candidate function and designing a control law that satisfies the stability condition; s (t), and the design process includes constructing a Lyapunov candidate function and designing a control law that satisfies the stability condition;

[0080] First, construct the Lyapunov candidate function:

[0081] V(t) = e(t) T Pe(t),

[0082] where P is a positive definite matrix and V(t) is the Lyapunov candidate function;

[0083] Utilize the stability condition:

[0084] where, is the time derivative of the Lyapunov candidate function, α is the system decay rate, and β is the performance index influence coefficient;

[0085] Design the switching control law under the condition of satisfying the stability condition:

[0086] u s (t) = -Ke(t) - Lsgn(e(t)),

[0087] where K is the state feedback gain matrix, L is the switching gain matrix, and sgn(e(t)) is the state error sign function;

[0088] The switching control law u s (t) output by the switching control law design unit is used to switch the working mode of the control input;

[0089] The stability verification unit verifies the stability of the control system using the Lyapunov function. The verification process uses the discrete Lyapunov difference formula, and the verification formula is:

[0090] ΔV(t) = V(t + 1) - V(t) ≤ -γV(t),

[0091] where ΔV(t) is the discrete difference of the Lyapunov function, and γ is the stability margin;

[0092] The stability verification unit receives the switching control law u s (t) to verify whether the system state satisfies the convergence condition under different operating modes.

[0093] Preferably, in the distributed cooperative optimization design module, based on the performance index, the switching control diagram output by the switching control module, the local state information of the system, and the global energy-saving optimization requirements, it further includes:

[0094] The local control law design unit uses the performance index and the switching control law u s (t) output by the switching control module and the local state X i (t) output by the physical modeling unit to construct the local control law output The construction formula is:

[0095]

[0096] where is the local control law output of the i-th region, u s (t) is the switching control law output by the performance index and the switching control module, X i (t) is the actual state of the i-th region, X ref,i (t) is the reference state of the i-th region, K i is the state feedback gain matrix of the i-th region;

[0097] The output of the local control law design unit is used to optimize and adjust the states of each region by the local control law for each region;

[0098] The global coordination optimization unit uses the local control law output output by the local control law design unit and the state information of each region to construct a global coordination optimization problem. The construction formula is:

[0099]

[0100] where U glob (t) is the global coordination control input, T is the total number of regions, α i is the state tracking weight of the i-th region, β i is the control input consistency weight of the i-th region, X i (t) is the actual state of the i-th region, X ref,i (t) is the reference state of the i-th region, is the output of the local control law for the i-th region;

[0101] The global coordination optimization unit solves to obtain the global coordination control input U glob (t) provides a global consistency basis for subsequent collaborative feedback updates;

[0102] The collaborative feedback update unit uses the global coordination control input U glob (t) output by the global coordination optimization unit and the local control law output by the local control law design unit Adjust the final output u of each region's control law i (t), and the construction formula is:

[0103]

[0104] where u i (t) is the final control law output of the i-th region, γ i is the collaborative feedback update weight of the i-th region, is the output of the local control law for the i-th region, U glob (t) is the global coordination control input;

[0105] The u i (t) output by the collaborative feedback update unit is used as the control instruction for each region's execution unit to achieve multi-region collaborative optimization control.

[0106] A terminal device includes a display unit, a data processing unit, a communication unit, and an execution unit. The display unit is used to display the air conditioner operation status and energy consumption information. The data processing unit is used to process the collected data. The communication unit is used to perform real-time data interaction with the central control system. The execution unit is used to execute the optimization control instruction.

[0107] A storage medium contains program codes of a computer program. The program codes are used to implement system modeling, online deep prediction, robust adaptive model predictive control, switching control, distributed collaborative optimization, and integrated control processes. After the program codes are executed on a computer, a central air conditioner energy management system is formed.

[0108] The present invention provides a central air conditioner energy management system based on data analysis. It has the following beneficial effects:

[0109] 1. The present invention uses data analysis and online deep prediction technologies to achieve real-time prediction of the air conditioner state and achieve precise control effects. Compared with the existing single model prediction scheme, it solves the problems of large prediction errors and response lags.

[0110] 2. The present invention adopts the robust adaptive model predictive control technology to realize the online iterative update of the control input and achieve the energy consumption optimization effect. Compared with the traditional static control method, it overcomes the defect of insufficient response to external disturbances.

[0111] 3. The present invention adopts the distributed collaborative optimization design technology to realize the multi-region linkage collaborative control and achieve the overall energy-saving effect. Compared with the traditional local independent regulation method, it makes up for the problem of insufficient collaboration between regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] Figure 1 It is the system diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0113] To enable those skilled in the art to understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, rather than all, of the embodiments of the present invention. Based on the embodiments of the present invention, other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0114] The present invention will be described in detail below in conjunction with the accompanying drawings:

[0115] Embodiment:

[0116] Please refer to the attached Figure 1 , the embodiment of the present invention provides a central air-conditioning energy-saving management system based on data analysis. The central air-conditioning energy-saving management system includes:

[0117] The system modeling and data acquisition module is used to establish the air-conditioning physical model and collect the room temperature, humidity, and wind speed data;

[0118] In the system modeling and data acquisition module, based on the dynamic characteristics of the central air-conditioning system operation state and the energy consumption optimization requirements, it further includes:

[0119] The physical modeling unit constructs a mathematical model describing the change of room temperature, humidity, and wind speed over time according to the thermodynamic characteristics and fluid dynamics characteristics of the central air-conditioning system, and defines the system state variables, control inputs, disturbance terms, and uncertainties. The algorithm formula is:

[0120]

[0121] Wherein, E1(t) is the system state vector, V1(t) is the control input vector, Φ1(·) is the system dynamic function, Γ1(·) is the control coupling matrix, Δ1(·) is the uncertainty term, and Ω1(t) is the external disturbance term;

[0122] The system state model output by the physical modeling unit provides variable definitions for the data acquisition unit and state evolution constraint conditions for the data preprocessing unit;

[0123] The data acquisition unit obtains the real-time operation data of the central air-conditioning system through sensor devices and converts the acquired data into a format that can be used by the mathematical model. Let the original data measured by the sensor be Y2(t). Therefore, the converted system state variable format is:

[0124] E2(t) = Ψ2(Y2(t)) + ε2(t),

[0125] where E2(t) is the converted system state variable, Y2(t) is the original sensor data, Ψ2(·) is the data conversion function, and ε2(t) is the measurement error;

[0126] The output data of the data acquisition unit enters the data preprocessing unit for filtering and anomaly detection;

[0127] The data preprocessing unit uses filtering, anomaly detection, and missing data completion algorithms to improve data quality and provides data input to the subsequent online deep prediction module;

[0128] Data filtering uses weighted moving average filtering to process data noise:

[0129]

[0130] where E3(t) is the filtered state data, w 3,k is the weighting coefficient, and N is the time window length;

[0131] Anomaly detection sets the anomaly detection threshold Θ3. If the data satisfies:

[0132] ||E3(t) - E3(t - 1)|| > Θ3, mark the data as abnormal data and eliminate it,

[0133] where Θ3 is the anomaly detection threshold;

[0134] Missing data completion uses the nearest neighbor interpolation method to complete missing data:

[0135]

[0136] where K is the number of nearest neighbor samples, and t i is the time point of the nearest non-missing data;

[0137] The output E3(t) of the data preprocessing unit is used as the input data of the online deep prediction module;

[0138] The state mapping unit combines the physical model and the preprocessed data to construct standardized state variables to match the input format of the online depth prediction module. The state mapping function is:

[0139] E4(t) = M4(E3(t)),

[0140] where E4(t) is the standardized state data, and M4(·) is the standardization mapping function;

[0141] The standardized data E4(t) output by the state mapping unit is directly used in the online depth prediction module to achieve future state prediction;

[0142] The online depth prediction module is used to calculate the future state and correct the prediction error using a deep neural network;

[0143] In the online depth prediction module, based on the standardized state data and the prediction accuracy optimization requirements provided by the system modeling and data acquisition module, it further includes:

[0144] The data input unit processes the standardized state data E4(t) provided by the system modeling and data acquisition module and converts it into the input format of the deep neural network model. The input format conversion is:

[0145] X5(t) = F5(E4(t)),

[0146] where X5(t) is the neural network input data vector, E4(t) is the standardized state data, and F5(·) is the data preprocessing function;

[0147] The output X5(t) of the data input unit is used as the input of the deep learning prediction unit;

[0148] The deep learning prediction unit uses a deep neural network to predict the future state of the central air-conditioning system and calculates the state at the future τ moment based on the current state data X5(t) The prediction model is:

[0149]

[0150] where, is the predicted future state, N5(·) is the deep neural network prediction model, and θ5 is the parameter of the neural network;

[0151] The output of the deep learning prediction unit is used as the input of the prediction error correction unit;

[0152] The prediction error correction unit calculates the prediction error and adjusts the neural network model parameters through an adaptive update mechanism. The error calculation formula is:

[0153]

[0154] Among them, E6(t + τ) is the prediction error, and X5(t + τ) is the true state, is the predicted future state;

[0155] For parameter update, online gradient adjustment is performed using the adaptive learning rate η6 to update the network parameter θ5:

[0156]

[0157] Among them, L(E6) is the loss function, is the gradient of the loss function with respect to the neural network parameters, and η6 is the adaptive learning rate;

[0158] The updated parameter θ5 of the output of the prediction error correction unit is fed back to the deep learning prediction unit to form an adaptive prediction optimization closed loop;

[0159] The robust adaptive model predictive control module is used to construct an optimal control problem with energy consumption index and state tracking index and solve the control input;

[0160] In the robust adaptive model predictive control module, based on the future state prediction result output by the online depth prediction module and the system energy consumption and state tracking requirements, it further includes:

[0161] An optimization problem construction unit, according to the future state prediction result provided by the online depth prediction module and the expected state X ref (t + τ) constructs an optimal control problem with energy consumption index and state tracking index, establishes system dynamic constraints and control input constraints, and constructs the optimization objective:

[0162]

[0163] Among them, J is the control optimization objective, and N p is the prediction time domain length, X(t + k + 1) is the system state at time t + k + 1, and X ref (t + k + 1) is the desired state, u(t + k) is the control input, λ1 is the state tracking index weight, and λ2 is the energy consumption index weight;

[0164] The optimization problem construction unit outputs the optimization objective J and related constraint conditions for the control solution unit to solve the control input;

[0165] The control solution unit, based on the optimization objective and constraint conditions output by the optimization problem construction unit, uses the online solution algorithm to solve and obtain the optimal control sequence u * (t) in real time, and feeds back the first control input to the execution unit, and the dynamic constraint description:

[0166] X(t + k + 1) = f(X(t + k), u(t + k)) + δ(t + k),

[0167] where f(·) is the dynamic function constructed by the physical modeling unit, and δ(t + k) is the modeling error and external disturbance;

[0168] The control solution unit updates the control input using the iterative gradient descent algorithm, and the calculation formula is:

[0169]

[0170] where η is the online update learning rate, is the gradient of the control input corresponding to the optimization objective J;

[0171] The control solution unit outputs the optimal control sequence u * The first element of (t) is used as the current control input and fed back to the system execution unit to achieve actual control;

[0172] The robustness adjustment unit adjusts the dynamic constraint parameters and the control input update rate through error analysis according to the optimal control input obtained by the control solution unit and the actual operation feedback of the system, to achieve control robustness and adaptive performance. The robustness compensation formula:

[0173]

[0174] where Δu(t) is the compensation control input, K is the compensation coefficient, X(t) is the actual system state, is the predicted state output by the online depth prediction module;

[0175] The robustness adjustment unit combines the compensation control input Δu(t) with the optimal control input u * (t) to form the final control input u(t): u(t) = u * (t) + Δu(t), ensuring that the system can operate stably under the action of disturbances and model errors;

[0176] The performance index and switching control module is used to design control laws for different working modes and prove the system stability with the Lyapunov function;

[0177] In the performance index and switching control module, based on the control input and state tracking error output by the robust adaptive model predictive control module and the system stability requirements, it further includes:

[0178] The performance index construction unit constructs the performance index function J according to the state tracking error e(t) = X(t) - X ref (t) and the control input u(t) output by the robust adaptive model predictive control module p, The construction formula is:

[0179] J p = e(t) T Qe(t) + u(t) T Ru(t),

[0180] where, e(t) is the state tracking error, Q is the state tracking weight matrix, u(t) is the output control input of the robust adaptive model predictive control module, R is the control energy consumption weight matrix, X(t) is the actual system state provided by the physical modeling unit, and X ref (t) is the reference state output by the online depth prediction module after data standardization, and J p is the performance index function;

[0181] The performance index construction unit outputs the performance index J p to provide a quantization basis for the switching control law design unit;

[0182] The switching control law design unit designs the control law u p (t) in different working modes according to the performance index J output by the performance index construction unit and the state tracking error e(t). The design process includes constructing a Lyapunov candidate function and designing a control law that satisfies the stability condition; s (t), and the design process includes constructing a Lyapunov candidate function and designing a control law that satisfies the stability condition;

[0183] First, construct the Lyapunov candidate function:

[0184] V(t) = e(t) T Pe(t),

[0185] where, P is a positive definite matrix and V(t) is the Lyapunov candidate function;

[0186] Utilize the stability condition:

[0187] where, is the time derivative of the Lyapunov candidate function, α is the system decay rate, and β is the performance index influence coefficient;

[0188] Design the switching control law under the condition of satisfying the stability condition:

[0189] u s (t) = -Ke(t) - Lsgn(e(t)),

[0190] where, K is the state feedback gain matrix, L is the switching gain matrix, and sgn(e(t)) is the state error sign function;

[0191] The switching control law u output by the switching control law design unit s(t) is used to switch the working mode of the control input;

[0192] The stability verification unit verifies the stability of the control system by using the Lyapunov function. The verification process adopts the discrete Lyapunov difference formula, and the verification formula is:

[0193] ΔV(t) = V(t + 1) - V(t) ≤ -γV(t),

[0194] where ΔV(t) is the discrete difference of the Lyapunov function, and γ is the stability margin;

[0195] The stability verification unit receives the switching control law u s (t) to verify whether the system state satisfies the convergence condition under different working modes;

[0196] The distributed cooperative optimization design module is used to realize the optimization of the multi - area local control law and the global coordination control;

[0197] In the distributed cooperative optimization design module, based on the performance index, the switching control diagram output by the switching control module, the local state information of the system, and the global energy - saving optimization requirements, it further includes:

[0198] The local control law design unit uses the performance index and the switching control law u s (t) output by the switching control module and the local state X i (t) output by the physical modeling unit to construct the local control law output The construction formula is:

[0199]

[0200] where, is the local control law output of the i - th area, u s (t) is the switching control law output by the performance index and the switching control module, X i (t) is the actual state of the i - th area, X ref,i (t) is the reference state of the i - th area, K i is the state - feedback gain matrix of the i - th area;

[0201] The output by the local control law design unit is used to optimize and adjust the states of each area by the local control law of each area;

[0202] The global coordination optimization unit uses the local control law output output by the local control law design unit and the state information of each area to construct the global coordination optimization problem. The construction formula is:

[0203]

[0204] Among them, U glob (t) is the global coordinated control input, T is the total number of regions, and α i is the state tracking weight of the i-th region, and β i is the control input consistency weight of the i-th region. X i (t) is the actual state of the i-th region, and X ref,i (t) is the reference state of the i-th region. is the output of the local control law of the i-th region;

[0205] The global coordinated optimization unit solves to obtain the global coordinated control input U glob (t) provides a global consistency basis for subsequent collaborative feedback updates;

[0206] The collaborative feedback update unit uses the global coordinated control input U glob (t) output by the global coordinated optimization unit and the local control law output by the local control law design unit to adjust the final output u i (t) of each region's control law. The construction formula is:

[0207]

[0208] Among them, u i (t) is the final control law output of the i-th region, and γ i is the collaborative feedback update weight of the i-th region. is the local control law output of the i-th region, and U glob (t) is the global coordinated control input;

[0209] The u i (t) output by the collaborative feedback update unit is used as the control instruction for each region's execution unit to achieve multi-region collaborative optimization control;

[0210] The integrated controller implementation process module is used to integrate data transfer, control instruction generation, and feedback update of each module to form a closed-loop control.

[0211] The system modeling and data acquisition module establishes a mathematical model of the central air conditioner, collects real-time environmental data, and ensures that the system can accurately reflect the operating state of the air conditioner. By combining physical modeling with data acquisition, the data reliability is improved, providing input for subsequent prediction and control. Compared with traditional simple sensing monitoring, an accurate state mapping is constructed to avoid regulation deviation caused by data errors and improve the overall perception ability of the system.

[0212] The online depth prediction module predicts the future state of the air conditioner through a deep neural network, enabling dynamic adjustment and improving the accuracy of regulation. Compared with traditional static prediction methods, it can adaptively learn and continuously optimize the prediction accuracy, enabling the system to maintain stable and efficient operation in complex environments and reducing the impact of prediction errors on control.

[0213] The robust adaptive model predictive control module constructs a control strategy for energy consumption optimization and state tracking, and uses an online solution algorithm to update the control input in real time to achieve optimal energy-saving regulation. Compared with traditional fixed control methods, it has an adaptive ability and can dynamically optimize the control strategy according to environmental changes, improving the energy efficiency ratio and reducing the operating cost.

[0214] The performance index and switching control module ensures system stability and meets the requirements of different working modes by constructing a Lyapunov function and designing a switching control law. Compared with traditional single control strategies, it can automatically switch the control mode according to different operating states, improving the adaptability and energy-saving effect of the air conditioner and avoiding energy waste caused by mode mismatch.

[0215] The distributed collaborative optimization design module realizes the collaborative optimization of the multi-region air conditioning system, ensuring that the regulation of each region can meet its own needs and can be coordinated with the overall system. Compared with traditional independent region control methods, it can dynamically balance the energy consumption requirements of the local and global areas, improve the energy-saving efficiency of the overall system, and avoid resource waste caused by conflicts in the regulation of each region.

[0216] The integrated controller implementation process module integrates the data flow and control process of each functional unit to form a closed-loop control system, ensuring the efficient and stable operation of the system. Compared with traditional independent control methods, it uniformly schedules each subsystem, realizes information sharing and collaborative optimization, and improves the intelligent level of the overall management of the central air conditioner.

[0217] In summary, the present invention constructs an efficient central air conditioner energy-saving management system through multi-level data analysis, intelligent prediction, and optimal control. Each module cooperates with each other, can accurately predict the system state, can adaptively optimize the control strategy, and simultaneously realizes multi-region coordinated regulation. Compared with traditional methods, it improves the prediction accuracy, response speed, and energy-saving effect, making the air conditioner management intelligent and efficient, and providing strong support for smart buildings and green buildings.

[0218] A terminal device, which includes a display unit, a data processing unit, a communication unit, and an execution unit. The display unit is used to display the operating state and energy consumption information of the air conditioner. The data processing unit is used to process the collected data. The communication unit is used to perform real-time data interaction with the central control system. The execution unit is used to execute the optimal control instruction.

[0219] A storage medium contains program codes carried by a computer program. The program codes are used to implement system modeling, online depth prediction, robust adaptive model predictive control, switching control, distributed collaborative optimization, and an integrated control process. After the program codes are executed on a computer, a central air-conditioning energy management system is formed.

[0220] The terminal device integrates display, data processing, communication, and execution functions, making the central air-conditioning energy management intuitive and efficient. Through the display unit, users can monitor the operating status and energy consumption data of the air conditioner in real time, improving management transparency. The data processing unit ensures the efficient calculation and optimization of the collected data, improving the accuracy of regulation. The communication unit supports the real-time interaction between the device and the central control system, ensuring the efficient transmission of data and collaborative control. The execution unit can accurately execute the optimized control strategy, ensuring the implementation of the regulation instructions and further enhancing the energy-saving effect. Compared with traditional air-conditioning management devices, the terminal device has intelligence, visualization, and linkage, enabling users to conveniently manage the central air-conditioning system.

[0221] The storage medium contains the core algorithms of a complete central air-conditioning energy management system, and implements system modeling, online depth prediction, robust adaptive control, switching control, distributed optimization, and overall process management through a computer program. The program codes ensure the stability and intelligence of the system operation, enabling the central air conditioner to perform accurate prediction and optimized control based on real-time data. Compared with traditional air-conditioning management solutions with preset rules or simple logic control, the intelligent algorithms of this storage medium can dynamically adapt to environmental changes, improve the energy-saving efficiency, and reduce manual intervention at the same time, realizing intelligent and efficient air-conditioning management.

[0222] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A central air-conditioning energy-saving management system based on data analysis, characterized in that The central air conditioning energy-saving management system comprises: The system modeling and data acquisition module is used to establish the air conditioning physical model and collect room temperature, humidity, and wind speed data; The online deep prediction module is used to calculate future states and correct prediction errors using deep neural networks; The robust adaptive model predictive control module is used to construct the optimization control problem with energy consumption index and state tracking index and solve the control input; The performance index and switching control module are used to design control laws for different working modes and the Lyapunov function is used to prove the stability of the system; The distributed collaborative optimization design module is used to realize multi-region local control law and global coordinated control optimization; The integrated controller realizes the process module to integrate the data transmission, control instruction generation and feedback update of each module to form a closed-loop control.

2. The central air-conditioning energy-saving management system based on data analysis according to claim 1, characterized in that The system modeling and data acquisition module further includes: The physical modeling unit constructs a mathematical model describing the changes of room temperature, humidity, and wind speed over time based on the thermodynamic and fluid dynamic characteristics of the central air-conditioning system, and defines system state variables, control inputs, disturbance terms, and uncertainties. The algorithm formula is: Where E1(t) is the system state vector, V1(t) is the control input vector, Φ1(·) is the system dynamic function, Γ1(·) is the control coupling matrix, Δ1(·) is the uncertainty term, and Ω1(t) is the external disturbance term; The system state model output by the physical modeling unit provides variable definitions for the data acquisition unit and provides state evolution constraints for the data preprocessing unit; The data acquisition unit obtains the real-time operation data of the central air-conditioning system through the sensor equipment, and converts the collected data into a format that can be used by the mathematical model. Assuming that the original data measured by the sensor is Y2(t), the converted system state variable format is: E2(t)=Ψ2(Y2(t))+ε2(t), Among them, E2(t) is the converted system state variable, Y2(t) is the original sensor data, Ψ2(·) is the data conversion function, and ε2(t) is the measurement error; The output data of the data acquisition unit enters the data preprocessing unit for filtering and abnormality detection.

3. A central air-conditioning energy-saving management system based on data analysis according to claim 1, characterized in that The system modeling and data acquisition module further includes: The data preprocessing unit uses filtering, anomaly detection and missing data completion algorithms to improve data quality and provide data input to the subsequent online depth prediction module; The data filtering uses weighted moving average filtering to process data noise: Among them, E3(t) is the filtered state data, w 3,k is the weighting coefficient, and N is the time window length; The anomaly detection sets an anomaly detection threshold θ3, if the data satisfies: ||E3(t)-E3(t-1)||>Θ3, the data is marked as abnormal data and removed, Among them, Θ3 is the anomaly detection threshold; The missing data is completed by using the nearest neighbor interpolation method: where K is the number of nearest neighbor samples, and t i is the time point of the nearest non-missing data; The output E3(t) of the data preprocessing unit is used as input data of the online depth prediction module; The state mapping unit combines the physical model and the preprocessed data to construct a standardized state variable to match the input format of the online depth prediction module. The state mapping function is as follows: E4(t) = M4(E3(t)), where E4(t) is the standardized state data, and M4(·) is the standardization mapping function; The standardized data E4(t) output by the state mapping unit is directly used in the online depth prediction module to achieve future state prediction.

4. An energy-saving management system for central air-conditioning based on data analysis according to claim 1, characterized in that In the online depth prediction module, based on the standardized state data and the prediction accuracy optimization requirements provided by the system modeling and data acquisition module, it further includes: The data input unit processes the standardized state data E4(t) provided by the system modeling and data acquisition module and converts it into the input format of the deep neural network model. The input format conversion is as follows: X5(t) = F5(E4(t)), where X5(t) is the neural network input data vector, E4(t) is the standardized state data, and F5(·) is the data preprocessing function; The output X5(t) of the data input unit is used as the input of the deep learning prediction unit; The deep learning prediction unit uses a deep neural network to predict the future state of the central air-conditioning system, and calculates the state at the future τ moment based on the current state data X5(t). The prediction model is as follows: where, is the predicted future state, N5(·) is the deep neural network prediction model, and θ5 is the parameter of the neural network; Output of the deep learning prediction unit As the input of the prediction error correction unit; The prediction error correction unit calculates the prediction error and adjusts the neural network model parameters through an adaptive update mechanism. The error calculation formula is: Among them, E6(t + τ) is the prediction error, and X5(t + τ) is the true state, which is the predicted future state; Parameter update: Online gradient adjustment is performed using the adaptive learning rate η6 to update the network parameters θ5: where \(L(E6)\) is the loss function, is the gradient of the loss function with respect to the neural network parameters, and \(\eta6\) is the adaptive learning rate; The output of the prediction error correction unit, the updated parameter θ5, is fed back to the deep learning prediction unit to form an adaptive prediction optimization closed loop.

5. A central air-conditioning energy-saving management system based on data analysis according to claim 1, characterized in that In the robust adaptive model predictive control module, based on the future state prediction results output by the online depth prediction module and the system energy consumption and state tracking requirements, it further includes: Optimization problem construction unit, according to the future state prediction results provided by the online depth prediction module and the expected state X ref (t + τ) constructs an optimal control problem including energy consumption index and state tracking index, establishes system dynamic constraints and control input constraints, and constructs the optimization objective: Among them, J is the control optimization objective, N p is the prediction horizon length, X(t + k + 1) is the system state at time t + k + 1, X ref (t + k + 1) is the desired state, u(t + k) is the control input, λ1 is the weight of the state tracking index, and λ2 is the weight of the energy consumption index; The optimization problem construction unit outputs the optimization objective J and related constraint conditions for the control solution unit to solve the control input.

6. A central air-conditioning energy-saving management system based on data analysis according to claim 1, characterized in that, In the robust adaptive model predictive control module, based on the future state prediction results output by the online depth prediction module and the system energy consumption and state tracking requirements, it further includes: The control solution unit constructs the optimization objective and constraint conditions of the unit output based on the optimization problem, and uses the online solution algorithm to solve in real time to obtain the optimal control sequence u * (t), and feeds back the first control input to the execution unit. The dynamic constraint description is as follows: X(t + k + 1) = f(X(t + k), u(t + k)) + δ(t + k), where f(·) is the dynamic function constructed by the physical modeling unit, and δ(t + k) is the modeling error and external disturbance; The control solution unit updates the control input using the iterative gradient descent algorithm. The calculation formula is: where η is the online update learning rate, is the gradient of the control input corresponding to the optimization objective J; The control and solution unit outputs the first element of the optimal control sequence u * (t) as the current control input, which is fed back to the system execution unit to achieve actual control; The robustness adjustment unit adjusts the dynamic constraint parameters and the control input update rate through error analysis according to the optimal control input obtained by the control solution unit and the actual operation feedback of the system to achieve control robustness and adaptive performance. The robustness compensation formula: where, Δu(t) is the compensation control input, K is the compensation coefficient, and X(t) is the actual system state, which is the predicted state output by the online depth prediction module; The robustness adjustment unit combines the compensation control input Δu(t) and the optimal control input u * (t) to form the final control input u(t): u(t) = u * (t) + Δu(t), ensuring that the system can operate stably under disturbances and model errors.

7. An energy-saving management system for a central air conditioner based on data analysis according to claim 1, characterized in that, In the performance index and switching control module, based on the control input and state tracking error output by the robust adaptive model predictive control module and the system stability requirements, it further includes: Performance index construction unit, constructs a performance index function J based on the output state tracking error e(t) = X(t) - X ref (t) and the control input u(t) p , and the construction formula is: J p = e(t) T Qe(t) + u(t) T Ru(t), Among them, e(t) is the state tracking error, Q is the state tracking weight matrix, u(t) is the output control input of the robust adaptive model predictive control module, R is the control energy consumption weight matrix, X(t) is the actual system state provided by the physical modeling unit, and X ref (t) is the reference state output by the online depth prediction module after data standardization, and J p is the performance index function; The performance index construction unit outputs a performance index J p to provide a quantization basis for the switching control law design unit; The switching control law design unit constructs the output performance index J of the unit according to the performance index p and the state tracking error e(t) to design the control law u s (t) in different working modes. The design process includes constructing a Lyapunov candidate function and designing a control law that satisfies the stability condition; First, construct the Lyapunov candidate function: V(t) = e(t) T Pe(t), where P is a positive definite matrix and V(t) is the Lyapunov candidate function; Using the stability condition: where, is the time derivative of the Lyapunov candidate function, α is the system decay rate, and β is the performance index influence coefficient; Design the switching control law under the stability condition: u s (t) = -Ke(t) - Lsgn(e(t)), where K is the state feedback gain matrix, L is the switching gain matrix, and sgn(e(t)) is the state error sign function; The switching control law u s (t) output by the switching control law design unit is used to switch the working mode of the control input; The stability verification unit verifies the stability of the control system by using the Lyapunov function. The discrete Lyapunov difference formula is adopted in the verification process, and the verification formula is as follows: ΔV(t) = V(t + 1) - V(t) ≤ -γV(t), where ΔV(t) is the discrete difference of the Lyapunov function, and γ is the stability margin; The stability verification unit receives the switching control law u s (t) to verify whether the system state satisfies the convergence condition under different working modes.

8. An energy-saving management system for a central air conditioner based on data analysis according to claim 1, characterized in that, In the distributed collaborative optimization design module, based on the performance index, the switching control law output by the switching control module, the local state information of the system, and the global energy-saving optimization requirements, it further includes: The local control law design unit constructs the local control law output by using the performance index and the switching control law u s (t) output by the switching control module and the local state X i (t) output by the physical modeling unit The construction formula is as follows: Among them, is the output of the local control law for the i-th region, u s (t) is the switching control law output by the performance index and the switching control module, X i (t) is the actual state of the i-th region, X ref,i (t) is the reference state of the i-th region, K i is the state feedback gain matrix for the i-th region; The output of the local control law design unit is used to optimize and adjust the states of each area by means of the sub-area local control law; The global coordination optimization unit designs the local control law output of the unit output by using the local control law and the state information of each region to construct a global coordination optimization problem. The construction formula is as follows: Among them, U glob (t) is the global coordinated control input, T is the total number of regions, α i is the state tracking weight of the i-th region, β i is the control input consistency weight of the i-th region, X i (t) is the actual state of the i-th region, X ref,i (t) is the reference state of the i-th region, is the output of the local control law of the i-th region; The global coordination optimization unit solves to obtain the global coordination control input U glob (t) provides a global consistency basis for subsequent collaborative feedback updates; The collaborative feedback update unit utilizes the global coordination optimization unit to output the global coordination control input U glob (t) and the local control law design unit to output the local control law output Adjust the final output u i (t) of each regional control law, and the construction formula is: where, u i (t) is the output of the final control law for the i-th region, γ i is the collaborative feedback update weight for the i-th region, is the output of the local control law for the i-th region, U glob (t) is the global coordinated control input; The u i (t) output by the collaborative feedback update unit is used as the control instruction for each regional execution unit to achieve multi-regional collaborative optimization control.

9. A terminal device, characterized in that, The terminal device includes a display unit, a data processing unit, a communication unit, and an execution unit. The display unit is used to display the operating state and energy consumption information of the air conditioner. The data processing unit is used to process the collected data. The communication unit is used to perform real-time data interaction with the central control system. The execution unit is used to execute the optimization control instruction.

10. A storage medium, characterized in that, The storage medium contains program codes of a computer program. The program codes are used to implement system modeling, online deep prediction, robust adaptive model predictive control, switching control, distributed collaborative optimization, and integrated control processes. After being executed on a computer, the program codes form a central air-conditioning energy-saving management system.

Images