Building energy-saving intelligent integrated control system and method

Abstract

The application belongs to the technical field of intelligent control, and discloses a kind of building energy-saving intelligent integrated control system and method;Including the building comprehensive data of the building internal space area is collected;The building comprehensive data is pretreated, and standard building data is obtained;According to the standard building data, the building dynamic model is established, and the energy consumption change curve and the comfort change curve in the future fixed time period of building internal space area are obtained by using dynamic model prediction;The energy consumption change curve and the comfort change curve in the future fixed time period of building internal space area are used as constraints, and a series of optimal control strategy sequences are obtained;The optimal control strategy sequence obtained by solving is distributed to each actuator, and integrated intelligent control of heating, refrigeration, ventilation, lighting and other equipment is realized;Not only improve the energy-saving level of building, but also create a comfortable environment for indoor, and comprehensively enhance the comprehensive performance of building.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and more specifically, to an intelligent integrated control system and method for building energy conservation. Background Technology

[0002] Patent application publication number CN118605146A discloses a smart building energy-saving control system, method, and electrical equipment. By constructing a three-dimensional topology network based on IP addresses, it can more clearly display the device form, overall structure, and topological relationships. To a certain extent, this helps to better regulate each controlled device, reduce unnecessary energy consumption, and the energy-saving control server determines the degree of adjustment for each controlled object based on its energy consumption and actual operating variables. This allows each intelligent node control device to adjust the corresponding controlled device, enabling each controlled object to reach its optimal operating state, thereby helping to reduce energy waste and achieve the goal of energy saving.

[0003] However, existing systems still lack comprehensive perception of the building's internal environmental conditions and energy consumption. Due to poor data quality, they cannot accurately grasp the building's operational status, making it difficult to formulate scientific and reasonable control strategies. Secondly, existing systems have limited modeling and prediction capabilities; they typically use simple empirical or data-driven models, failing to accurately describe the complex dynamic processes within the building. Predicted energy consumption and comfort changes often deviate from reality, limiting the foresight and effectiveness of control strategies. Furthermore, existing systems often consider only a single objective when solving control strategies, neglecting inherent contradictions and balances. This single-objective optimization often fails to truly meet actual needs. In addition, existing systems often employ simple search or rule-based methods in the solution process, limiting optimization capabilities. Due to the high complexity and variability of building systems, it is difficult to obtain truly optimal control strategies, and control effects are difficult to achieve ideal levels. It is evident that existing building energy-saving control systems have significant shortcomings in multiple aspects, restricting overall system performance and making it difficult to truly meet actual needs in real-world application scenarios.

[0004] In view of this, the present invention proposes an intelligent integrated control system and method for building energy conservation to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent integrated control system for building energy conservation, comprising: a data acquisition module for acquiring comprehensive building data of the internal space area of ​​a building; and preprocessing the comprehensive building data to obtain standard building data;

[0006] The comprehensive prediction module is used to build a building dynamic model based on standard building data, and use the dynamic model to predict the energy consumption change curve and comfort change curve of the building's internal space area over a fixed period of time in the future.

[0007] The strategy solving module is used to solve for a series of optimal control strategy sequences by taking the energy consumption change curve and comfort change curve of the building's internal space area over a fixed future time period as constraints. The solved optimal control strategy sequences are then distributed to the actuators of the building. The modules are connected to each other via wired and / or wireless means to realize data transmission between the modules.

[0008] Furthermore, the comprehensive building data includes environmental parameters and energy consumption data within the building; the environmental parameters include temperature, humidity, carbon dioxide concentration, illuminance, and fresh air volume; the energy consumption data includes heating capacity, cooling capacity, and lighting power.

[0009] Furthermore, the method for preprocessing the comprehensive building data includes:

[0010] The building data is cleaned and dimensionless processing is performed on the data of different dimensions to obtain preliminary building data. The preliminary building data is arranged into a corresponding data time series according to the time order of data types. The data time series is segmented to obtain m segments of building data. Standard building data is extracted from each of the m segments of building data.

[0011] Furthermore, the methods for performing data segmentation include:

[0012] For each data time series, it is viewed as a state sequence of a discrete-time Markov chain. Based on the range of data values ​​in the time series, it is discretized into a finite number of states, thus determining the state space of the Markov chain. A counting matrix C is initialized, with all elements in C initially set to 0. The time series data is traversed. For each adjacent time point pair (Y_t, Y_t+1), where Y_t represents the state at time t and Y_t+1 represents the state at time t+1, the state indices i and j corresponding to Y_t and Y_t+1 are determined, respectively. The value of the element C(i, j) in the counting matrix C corresponding to state indices i and j is incremented by 1, i.e., C(i, j) records the number of transitions from state z_i to state z_j. After traversing the entire time series, for each state s_i, the frequency of its transitions to other states is calculated, i.e., C(i, j) is divided by the sum of all transitions from state z_i. The resulting frequencies are the elements of the transition probability matrix P.

[0013] Using the obtained transition probability matrix, for each time point t in the data time series, calculate the conditional probability P(t, z_i) that the data value at that time point belongs to each state z_i;

[0014] For each time point t, calculate the maximum value of the conditional probability P_max that the data value at that time point belongs to each state;

[0015] Based on the maximum value P_max and the preset probability threshold θ=a×P_max, when P(t, z_i) is greater than or equal to θ, it is determined that time point t belongs to state z_i; a is a trade-off parameter.

[0016] Scan the entire data time series and group adjacent time points belonging to the same state into a data segment to obtain several preliminary segmented data segments; perform cluster analysis on the several preliminary segmented data segments to obtain m segments of building data.

[0017] Furthermore, the methods for performing cluster analysis include:

[0018] Calculate the feature vector for each preliminary segment of data. The feature vector is obtained by concatenating the weighted mean vector, variance vector, mean vector, maximum value, minimum value, kurtosis, and skewness.

[0019] The methods for calculating the weighted mean vector include:

[0020] For the The initial segmented data has a time range of [missing information]. [t_l _ start,t_l _ end] ,in, For the first The starting time point of the initial segmented data. For the first The end time point of each preliminary segment of data; based on the end time point Define the time weighting function; the formula for the time weighting function is:

[0021] ;in, It is an attenuation coefficient. This is the periodic adjustment coefficient. For time Time weighting; This is the initial time point of the function. For periodicity;

[0022] The first is calculated based on the time weighting function. The weighted mean vector of the initial segmented data ;

[0023] ;in, For data points Confidence weights, data points For the initial segmentation of data in time Data points;

[0024] ,in, The mean of the initial segmented data. for The standard deviation of the neighborhood and For confidence level parameters;

[0025] Based on the calculated feature vectors, a distance matrix D is constructed. The element D(I,J) in the distance matrix D represents the Mahalanobis distance between the I-th preliminary segment data and the J-th preliminary segment data. Based on the distance matrix D, a merging iteration is performed to obtain m segments of building data.

[0026] Furthermore, the method of merging and iterating based on the distance matrix D includes:

[0027] Each initial segment of data in the distance matrix D is treated as a separate cluster. The eigenvector of each cluster is calculated, and the overall distance between clusters is calculated based on these eigenvectors. The formula for calculating the overall distance is:

[0028] ;in, For clusters and The combined distance between them For clusters eigenvectors, For clusters eigenvectors, It is an adaptive covariance matrix. For clusters Time center, It is a cluster Time center, It is a time decay parameter;

[0029] The inter-cluster merging cost is calculated based on the comprehensive distance between clusters; the formula for calculating the merging cost is as follows:

[0030] ;in, It is a cluster shape penalty term. For clusters and The cost of merging between them;

[0031] Cluster shape penalty ;in, For clusters variance For clusters variance For clusters skewness, For clusters skewness, and These are learnable parameters;

[0032] Among all the current pairwise clusters, find the two clusters with the minimum merging cost. and Merge them into a new cluster new cluster eigenvectors based on and The feature vectors are obtained by weighting them with the number of samples; the new cluster is calculated. Calculate the distances to other clusters and update the distance matrix;

[0033] When the sum of the squared deviations within all clusters is less than the preset merging threshold, merging stops, and m clusters corresponding to m building data are finally obtained.

[0034] Furthermore, the methods for establishing the building dynamic model include:

[0035] The interior space of the building is discretized into several grid units, each corresponding to a magnetic spin. The magnetic spin can take two states: up or down, representing the two states of the grid unit.

[0036] Each grid cell contains several standard building data points; based on the standard building data, adjacent grid cells are defined. and The strength of the coupling between them ;in, Grid unit and grid unit The distance between centers Grid unit Any standard building data in the database, Grid unit Any standard building data in the database, Grid unit and grid unit The relative azimuth angle between them; For anisotropic parameters, For phase parameters; It is an exponential parameter; For intensity adjustment parameters, This is the intensity adjustment index parameter; For distance index parameters;

[0037] External field terms are defined using environmental parameters and energy consumption data; that is... ; Grid unit Outdoor events, Grid unit The corresponding environment parameter vector, Grid unit The corresponding energy consumption data vector; For mapping functions;

[0038] The standard building data of each grid cell is randomly divided into a training set and a test set according to a certain ratio, with the data in the training set serving as the training data. The loss function is to minimize the mean square error between the output and the training data. The standard building data of the p-th training data is defined as X_p, and the corresponding real environmental parameters and energy consumption data are obtained. Using the current coupling parameters, the output is obtained by simulating time evolution. The coupling parameters are initialized with a set of random values, and the training data is iterated several times. The iteration process is as follows: a small batch of training data is randomly sampled from the training set, the loss function is calculated for this small batch of training data, and the gradient of the loss function with respect to the coupling parameters is calculated. The coupling parameters are updated according to the gradient using stochastic gradient descent or its variant algorithm. The process continues until the preset maximum number of iterations is reached, and the optimal coupling parameters are obtained, resulting in the completed building dynamic model.

[0039] Furthermore, the method of using dynamic models to predict the energy consumption change curves and comfort change curves of the building's interior space over a fixed future time period includes:

[0040] Obtain the current value of the standard building data, determine the initial state of each cell based on the current value of the standard building data, and combine the initial states of all cells into a state vector;

[0041] In time According to the current state vector And the optimal coupling parameters are used to calculate each lattice cell. Spin flip probability ;

[0042] ;in, Grid unit In time The spin state, Grid unit In time The spin state, For random disturbance terms;

[0043] Based on the spin flipping probability Monte Carlo sampling is used, corresponding to the lattice unit. Generate a uniformly random number in the interval [0, 1]. ;like Then the grid unit In time spin state Set as That is, spin flipping, otherwise the spin state remains unchanged; all lattice elements in time The spin states form a new state vector; this process is repeated until the predicted maximum time step is reached. This yields a series of state vectors;

[0044] For each time step, the state vector corresponds to a set of environment parameter vectors. and energy consumption data vector Through the reverse mapping function Convert to and ;

[0045] Environmental parameter vector At the maximum time step All sampling points within the time frame form an environmental parameter variation curve; for each time step, the corresponding comfort index is calculated, forming a comfort variation curve; the comfort index represents the percentage of dissatisfaction. ; Utilizing energy consumption data vectors At the maximum time step Total computational energy consumption for all sampling points within the range This constitutes the energy consumption change curve.

[0046] Furthermore, the solution method for the optimal control strategy sequence includes:

[0047] The optimization objectives are defined as minimizing energy consumption and maximizing comfort, that is, minimizing total energy consumption and minimizing comfort index over a future period; the optimization objectives are expressed as two objective functions. and ;

[0048] ; ;

[0049] Define the optimization variables as control parameters; define the constraints, including the energy consumption variation curve. The peak value does not exceed the preset energy consumption limit. The peak and trough ranges of the comfort variation curve are all within the preset comfort threshold range;

[0050] Initialize a policy population, determine the policy population size N2 and the dimension n2 of the optimization variables; for each individual in the policy population, randomly generate a set of optimization variables as the individual's position vector;

[0051] Two objective functions and The weighted sum is used as the update function for the policy population; the individual is calculated. The value of the update function That is, the update function in The function value at that point is denoted as the updated value;

[0052] Calculate the centroid of the current population As a potential optimal solution; among which... Let d be the position vector of individual d. For each individual Calculate individual With the center of mass European distance and unit vector ;

[0053] For each individual Generate a random step size that follows a normal distribution. Update the individual based on the random step size and unit vector. The position vector of the individual is obtained. New position vector And calculate the new position vector. The value of the update function ;

[0054] ;in, For constant parameters, Let be a random variable that follows a specific distribution;

[0055] Sort the policies by update value from smallest to largest to obtain the sorted policy population. Divide the sorted policy population into two parts: the first part consists of better individuals and the second part consists of worse individuals.

[0056] The number of better individuals is ;in, and To dynamically adjust parameters, This represents the current iteration number; It is a rounding function;

[0057] For each worse individual Randomly select a better individual Calculate the Euclidean distance between the two, denoted as the difference distance. Based on the gap distance, update the worse individuals. position vector To obtain worse individuals New position vector ; Calculate worse individuals New position vector The updated value;

[0058] ;in, For better individuals and worse individuals The generated random step size follows a normal distribution. For a better individual The position vector;

[0059] Repeatedly iterate and update the position vectors of individuals and worse individuals until the value of the update function converges. Finally, output the individual with the smallest update value in the current policy population, which is called the best individual. The corresponding set of optimization variables is the optimal control policy sequence.

[0060] A building energy-saving intelligent integrated control method, which is based on the building energy-saving intelligent integrated control system, includes: Step 1, collecting comprehensive building data of the internal space area of ​​the building; preprocessing the comprehensive building data to obtain standard building data;

[0061] Step 2: Establish a building dynamic model based on standard building data, and use the dynamic model to predict the energy consumption change curve and comfort change curve of the building's internal space area over a fixed period of time in the future.

[0062] Step 3: Using the energy consumption change curve and comfort change curve of the building's internal space area over a fixed future time period as constraints, solve for a series of optimal control strategy sequences; then distribute the solved optimal control strategy sequences to each actuator in the building.

[0063] The technical effects and advantages of the intelligent integrated control system and method for building energy conservation of this invention are as follows:

[0064] This invention comprehensively senses the environmental state and energy consumption within a building, laying a solid data foundation for subsequent refined control. By processing the raw data, it not only improves data quality but also reduces computational complexity. Secondly, it employs a dynamic modeling method, which accurately describes the complex dynamic behavior of the building's interior space and precisely predicts changes in energy consumption and comfort over a future period. This provides crucial support for developing forward-looking control strategies, facilitating proactive adjustment and optimization of the building's operational status. Furthermore, in the control strategy solution process, the two objectives of energy saving and comfort are organically combined, achieving multi-objective optimization. By balancing the relationship between these two objectives, a balanced optimal control strategy can be obtained, maximizing the satisfaction of energy saving and comfort needs and avoiding the one-sidedness that may result from single-objective optimization. In addition, an improved multi-objective optimization algorithm is used, possessing advantages such as strong global search capability, fast convergence speed, and good robustness. It can efficiently solve for the optimal control strategy sequence under constraints, thereby improving control accuracy and performance. Finally, it achieves integrated intelligent control of equipment, not only improving the building's energy efficiency but also creating a comfortable indoor environment, comprehensively enhancing the building's overall performance. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of an intelligent integrated control system for building energy conservation according to the present invention;

[0066] Figure 2 This is a schematic diagram of an intelligent integrated control method for building energy conservation according to the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example 1

[0069] Please see Figure 1 As shown in the figure, this embodiment of a building energy-saving intelligent integrated control system includes: a data acquisition module, used to collect comprehensive building data of the internal space area of ​​the building; and to preprocess the comprehensive building data to obtain standard building data;

[0070] The comprehensive prediction module is used to build a building dynamic model based on standard building data, and use the dynamic model to predict the energy consumption change curve and comfort change curve of the building's internal space area over a fixed period of time in the future.

[0071] The strategy solving module uses the energy consumption and comfort variation curves of the building's internal space over a fixed future time period as constraints to solve for a series of optimal control strategy sequences. The solved optimal control strategy sequences are then distributed to the building's actuators to achieve integrated intelligent control of heating, cooling, ventilation, lighting, and other equipment. The modules are connected via wired and / or wireless means to achieve data transmission between modules.

[0072] Furthermore, comprehensive building data includes environmental parameters and energy consumption data within the building. Environmental parameters include temperature, humidity, carbon dioxide concentration, illuminance, and fresh air volume. Energy consumption data includes heating, cooling, and lighting power. Illuminance refers to the luminous flux received per unit area, a physical quantity representing light intensity, reflecting the lighting conditions of the building's interior spaces. Appropriate illuminance levels are crucial for the comfort of the building's interior environment; excessively high or low illuminance can negatively impact visual comfort. Fresh air volume refers to the amount of fresh air supplied to the building's interior spaces through a fresh air system. Inside buildings, human activities and the operation of office equipment generate pollutants such as heat, carbon dioxide, and volatile organic compounds. If indoor air is not replaced over a long period, air quality will gradually deteriorate, affecting human health and comfort. Therefore, a fresh air system is needed to continuously supply fresh air to the interior to dilute and remove indoor pollutants, maintaining good indoor air quality.

[0073] Heating capacity refers to the heat provided by a heating system to the interior space of a building, usually measured in kilowatt-hours. The heating system consumes primary energy (such as natural gas, electricity, etc.) to provide heat to the building, and the amount of heating capacity directly determines the energy consumption level of the heating system. Cooling capacity refers to the heat removed from the interior space of a building by an air conditioning system, usually measured in kilowatt-hours. The air conditioning compressor needs to consume electricity to do work to complete the cooling. The greater the cooling capacity, the higher the electricity consumption of the air conditioning system.

[0074] Data cleaning is performed on the comprehensive building data, which involves removing outliers and noisy data, and dimensionless processing is performed on data of different dimensions so that they can be calculated and compared on the same order of magnitude. Common standardization methods include min-max standardization and Z-score standardization to obtain preliminary building data.

[0075] The preliminary building data is arranged into a corresponding time series according to the time order of data types; the time series of data is segmented to obtain m segments of building data; standard building data is extracted from each of the m segments of building data.

[0076] Specifically, for each segment of building data, various environmental parameters and energy consumption data are extracted, such as temperature, humidity, carbon dioxide concentration, illuminance, fresh air volume, heating capacity, cooling capacity, and lighting power. For each environmental parameter and energy consumption data, its comprehensive statistical characteristics within the data segment are calculated. The comprehensive statistical characteristics include mean, variance, maximum value, minimum value, kurtosis, and skewness.

[0077] The comprehensive statistical characteristics of each environmental parameter and energy consumption data are concatenated into a vector, which serves as the standard building data for that segment of building data; for m segments of building data, m vectors are obtained by repeating this process, which are the standard building data.

[0078] The original time-series building data was compressed into a vector of statistically significant standard building data, which can comprehensively describe the characteristics of each data segment, laying the foundation for subsequent modeling and optimization. At the same time, the standard building data vector has a relatively low dimension, which helps to reduce computational complexity.

[0079] Methods for segmenting data include:

[0080] For each data time series (such as a temperature time series), it is regarded as a state sequence of a discrete-time Markov chain. Based on the range of data values ​​in the data time series, it is discretized into a finite number of states, that is, the state space of the Markov chain is determined. For example, the temperature data is divided into three states: high temperature, medium temperature, and low temperature.

[0081] Initialize a counting matrix C, with all elements in C initialized to 0. Iterate through the time series data. For each adjacent time point pair (Y_t, Y_t+1), where Y_t represents the state at time t and Y_t+1 represents the state at time t+1, determine the state indices i and j corresponding to Y_t and Y_t+1, respectively. Increment the value of the element C(i, j) in the counting matrix C corresponding to state indices i and j by 1. That is, C(i, j) records the number of times the state z_i transitions to state z_j. After iterating through the entire time series, for each state s_i, calculate the frequency of its transitions to other states, i.e., C(i, j) divided by the sum of all transitions from state z_i. Repeat all the frequencies obtained, which are all the elements of the transition probability matrix P.

[0082] Using the transition probability matrix, for each time point t in the data time series, calculate the conditional probability P(t, z_i) that the data value at that time point belongs to each state z_i.

[0083] For each time point t, calculate the maximum value P_max of the conditional probability that the data value at that time point belongs to each state.

[0084] Based on the maximum value P_max, a preset probability threshold θ = a × P_max is used. When P(t, z_i) is greater than or equal to θ, it is determined that time point t belongs to state z_i. a is a trade-off parameter, which takes the value of a positive real number and is usually set to a value in the interval (0, 1). Different values ​​of a will result in different probabilities of the threshold θ, which in turn affects the ease with which time point is classified into a certain state.

[0085] Scan the entire data time series and group adjacent time points belonging to the same state into one data segment. That is, if time points t and t+1 belong to the same state z_i, they are divided into one data segment; thus, several preliminary segmented data segments are obtained. Cluster analysis is performed on several preliminary segmented data segments to obtain m segments of building data.

[0086] Specifically, the feature vector of each preliminary segment of data is calculated. The feature vector is obtained by concatenating the weighted mean vector, variance vector, mean vector, maximum value, minimum value, kurtosis and skewness. The variance vector is the variance of each variable (data point) in the preliminary segment of data.

[0087] The methods for calculating the weighted mean vector include:

[0088] For the The initial segmented data has a time range of [missing information]. [t_l _ start,t_l _ end] ,in, For the first The starting time point of the initial segmented data. For the first The end time point of each preliminary segment of data; based on the end time point Define the time weighting function; the formula for the time weighting function is:

[0089] ;in, It is an attenuation coefficient. This is a periodic adjustment coefficient that controls the strength of the periodicity. For time Time weighting; The initial time point of the function controls the phase of the function, determining the starting position of the entire function in time. It is a cycle, such as a day having a 24-hour cycle, or a year having a 365-day cycle.

[0090] The first is calculated based on the time weighting function. The weighted mean vector of the initial segmented data ;

[0091] ;in, For data points Confidence weights, data points For the initial segmentation of data in time Data points.

[0092] ,in, The mean of the initial segmented data. for The standard deviation of the neighborhood and The confidence level parameter is learnable. Data points that are far from the mean and have a large variance in their neighborhood have a low confidence level. By learning the parameter, the confidence level weight can be adaptively determined.

[0093] Based on the calculated feature vectors, a distance matrix D is constructed. The element D(I, J) in the distance matrix D represents the Mahalanobis distance between the I-th preliminary segment data and the J-th preliminary segment data. Based on the distance matrix D, a merging iteration is performed. Specifically, each preliminary segment data in the distance matrix D is regarded as a separate cluster, the feature vector of each cluster is calculated, and the comprehensive distance between clusters is calculated based on the feature vectors. The closer clusters are assigned a smaller distance.

[0094] The formula for calculating the total distance is:

[0095] ;

[0096] in, For clusters and The combined distance between them For clusters eigenvectors, For clusters eigenvectors, It is an adaptive covariance matrix that automatically adjusts the weights of different feature dimensions based on the data distribution. For clusters Time center, It is a cluster Time center, It is a time decay parameter.

[0097] The inter-cluster merging cost is calculated based on the comprehensive distance between clusters; the formula for calculating the merging cost is as follows:

[0098] ;in, It is a cluster shape penalty term. For clusters and The cost of merging between them.

[0099] Cluster shape penalty ;in, For clusters variance For clusters variance For clusters skewness, For clusters skewness, and It is a learnable parameter. The cluster shape penalty term tends to merge clusters with similar shapes (similar variance and skewness) to avoid producing malformed clusters.

[0100] Among all the current pairwise clusters, find the two clusters with the minimum merging cost. and Merge them into a new cluster new cluster eigenvectors based on and The feature vectors are obtained by weighting them with the number of samples; the new cluster is calculated. Calculate the distances to other clusters and update the distance matrix.

[0101] When the sum of the squared deviations within all clusters is less than the preset merging threshold, merging stops, and m clusters corresponding to m building data are finally obtained.

[0102] The clustering process further merges the original over-segmented data segments to obtain larger and more compact data segment representations, automatically discovering the inherent similarity patterns in the data and avoiding subjective manual segmentation. At the same time, by setting reasonable termination conditions, the degree of merging can be controlled to obtain the optimal data segmentation results.

[0103] The original building composite data time series is divided into multiple data segments, with each data type corresponding to m data segments. The segmentation method can adaptively discover potential pattern transformations from the data, which is beneficial for subsequent modeling and optimization.

[0104] The following are methods for creating a dynamic building model:

[0105] The interior space of the building is discretized into several grid units, each corresponding to a magnetic spin. The magnetic spin value can be either up or down (-1 or 1), representing the two states of the grid unit (such as high temperature / low temperature).

[0106] Each grid cell contains several standard building data points; based on the standard building data, adjacent grid cells are defined. and The strength of the coupling between them ;in, Grid unit and grid unit The distance between centers (either Euclidean distance or other distance metric), Grid unit Any standard building data in the database, Grid unit Any standard building data in the database, Grid unit and grid unit The relative azimuth angle between them; in the built environment, adjacent grid units may have different coupling strengths at different azimuth angles (such as different heat conduction in the east-west and north-south directions); The anisotropy parameter controls the amplitude, with a value range of [0, 1]. This is a phase parameter that controls the phase offset, and its value range is [0, 2π). It is an exponential parameter that controls the steepness; This is a strength adjustment parameter that controls the numerical range of the coupling strength. The intensity adjustment index parameter controls the smoothness. This is the distance exponent parameter, which controls the degree of coupling in the distance.

[0107] The external field term is defined using environmental parameters and energy consumption data. This external field term reflects the influence of the external environment on the state of the lattice cells; that is... ; Grid unit Outdoor events, Grid unit The corresponding environmental parameter vector (such as temperature, humidity, illuminance, etc.), Grid unit The corresponding energy consumption data vector (such as heating capacity, cooling capacity, lighting power, etc.); This is a mapping function, such as a linear function, a multilayer perceptron (MLP) function, a kernel function, or an attention function.

[0108] The standard building data of each grid cell is randomly divided into a training set and a test set according to a certain ratio (e.g., 7:3). The data in the training set is used as the training data. The loss function is to minimize the mean square error between the output and the training data (targeted training grid cells).

[0109] Define the standard building data of the p-th training data as X_p, and obtain the corresponding real environmental parameters and energy consumption data; using the current coupling parameters (including coupling strength and external field terms), obtain the output Ŷ_p (the model's predicted output for the input X_p, which includes the predicted environmental parameters and energy consumption data) by simulating time evolution.

[0110] The coupling parameters are initialized with a set of random values. The training data is iterated several times. The iteration process is as follows: a small batch of training data is randomly sampled from the training set. The loss function is calculated for this small batch of training data, and the gradient of the loss function with respect to the coupling parameters is calculated. The coupling parameters are updated according to the gradient using stochastic gradient descent or its variants (such as Adam, RMSProp, etc.). The process continues until the preset maximum number of iterations is reached, and the optimal coupling parameters (coupling strength and external field term) are obtained, which can fit the training data well and thus obtain better prediction and optimization capabilities. The completed building dynamic model is then obtained.

[0111] Obtain the current value of the standard building data, determine the initial state (spin orientation) of each grid cell based on the current value of the standard building data, and combine the initial states of all grid cells into a state vector;

[0112] In time According to the current state vector And with the optimal coupling parameters (coupling strength and external field terms), calculate each lattice element. Spin flip probability ;

[0113] ;in, Grid unit In time The spin state (referring to the value of the magnetic spin corresponding to each lattice unit at a certain moment) takes the value of 1 or -1. Grid unit In time The spin state has a value of 1 or -1. The random disturbance term follows a certain probability distribution, such as a Gaussian distribution or other distributions, in order to simulate the influence of various random factors on the state of lattice cells in actual situations, making it more realistic.

[0114] Based on the spin flipping probability Monte Carlo sampling is used to determine the lattice unit. In time Whether to flip the spin; i.e., the corresponding lattice unit. Generate a uniformly random number in the interval [0, 1]. ;like Then the grid unit In time spin state Set as That is, spin flipping, otherwise the spin state remains unchanged; all lattice elements in time The spin states form a new state vector; this process is repeated until the predicted maximum time step is reached. This yields a series of state vectors.

[0115] For each time step, the state vector corresponds to a set of environment parameter vectors. and energy consumption data vector Through the reverse mapping function Convert to and The inverse mapping function can be a linear function, a nonlinear function (such as MLP), or other functions.

[0116] Environmental parameter vector At the maximum time step All sampling points within the time frame form an environmental parameter variation curve; for each time step, the corresponding comfort index is calculated, forming a comfort variation curve; the comfort index represents the percentage of dissatisfaction. Specifically, the percentage of dissatisfaction. It is related to the average thermal sensation index and is used to predict what percentage of people will be dissatisfied with their current thermal environment given a given average thermal sensation index value.

[0117] Average thermal sensation index The thermal comfort index is a widely used indicator used to assess the subjective thermal sensation of the human body to the thermal environment. The value range is [-3, +3]. An average thermal comfort index of 0 indicates thermal neutrality, meaning that the human body feels neither cold nor hot, and is in the most comfortable state. An average thermal comfort index less than 0 indicates that the human body feels cold. The smaller the average thermal comfort index value, the colder the feeling. An average thermal comfort index greater than 0 indicates that the human body feels hot. The larger the average thermal comfort index value, the hotter the feeling.

[0118] The average thermal sensation index needs to take into account multiple environmental and human parameters, including air temperature, mean radiant temperature, relative humidity, air flow velocity, human activity level, and clothing conditions, and is calculated from the environmental parameter variation curve.

[0119] It can be seen that when the average thermal sensation index is 0, the dissatisfaction percentage is 5%, which means that even in an ideal thermally neutral environment, 5% of people will feel dissatisfied, due to individual differences in human beings.

[0120] When the average thermal sensation index deviates from 0, the percentage of dissatisfaction increases rapidly; for example, when the average thermal sensation index is equal to 1, the percentage of dissatisfaction is 26%; when the average thermal sensation index is equal to 2, the percentage of dissatisfaction is 76%.

[0121] The goal is to minimize the percentage of dissatisfaction in order to provide an ideal thermal comfort environment. Therefore, the curve showing the change in the percentage of dissatisfaction is used as one of the optimization objectives to maximize the satisfaction of human thermal comfort needs while saving energy.

[0122] Using energy consumption data vectors At the maximum time step Total computational energy consumption for all sampling points within the range This constitutes the energy consumption change curve.

[0123] Based on the initial state of the grid and its evolution over time, the state changes of each grid cell over a future period are predicted, and the changes in environmental parameters, energy consumption data, comfort indicators, and energy consumption are calculated accordingly.

[0124] The methods for solving the optimal control policy sequence include:

[0125] The optimization objectives are defined as minimizing energy consumption and maximizing comfort, that is, minimizing total energy consumption and minimizing comfort indicators over a future period of time.

[0126] Optimization variables are defined as control parameters used to influence the operation of the building environment and energy system, including heating, cooling, fresh air volume and lighting power.

[0127] Define constraints including energy consumption variation curves The peak value does not exceed the preset energy consumption limit. The peak and trough ranges of the comfort variation curve are all within the preset comfort threshold range;

[0128] Minimizing energy consumption and maximizing comfort are expressed as two objective functions. and ;

[0129] ; ;

[0130] Initialize a policy population, determine the policy population size N2 and the dimension n2 of the optimization variables; for each individual in the policy population, randomly generate a set of optimization variables as the individual's position vector.

[0131] Based on the constraints of the optimization variables, the position vector is subjected to boundary treatment to ensure that it remains within the feasible region.

[0132] Two objective functions and The weighted sum is used as the update function for the policy population; the individual is calculated. The value of the update function That is, the update function in The function value at that point is denoted as the updated value.

[0133] Calculate the centroid of the current population , as a potential optimal solution; where, Let d be the position vector of individual d. For each individual Calculate individual With the center of mass European distance and unit vector .

[0134] For each individual Generate a random step size that follows a normal distribution. Update the individual based on the random step size and unit vector. The position vector of the individual is obtained. New position vector And calculate the new position vector. The value of the update function (Updated value).

[0135] ;in, For constant parameters, As a random variable that follows a specific distribution (such as Gaussian distribution or other distributions), a certain random perturbation is introduced into the displacement direction, which helps individuals escape local optima and improves the diversity and robustness of the algorithm.

[0136] Sort the policies by update value from smallest to largest to obtain the sorted policy population. Divide the sorted policy population into two parts: the first part consists of better individuals and the second part consists of worse individuals.

[0137] The number of better individuals is ;in, and To dynamically adjust parameters, This represents the current iteration number; It is a rounding function; generally speaking... The value should not be too small (to avoid underutilization of information) nor too large (to avoid excessive computational overhead).

[0138] For each worse individual Randomly select a better individual Calculate the Euclidean distance between the two, denoted as the difference distance. Based on the gap distance, update the worse individuals. position vector To obtain worse individuals New position vector ; Calculate worse individuals New position vector The update value; dynamically adjust the division of "better" and "worse" individuals according to the actual situation of the population, so as to make more flexible and efficient use of population information.

[0139] ;in, For better individuals and worse individuals The generated random step size follows a normal distribution. For a better individual The position vector;

[0140] Repeatedly iterate and update the position vectors of individuals and worse individuals until the value of the update function converges. Finally, output the individual with the smallest update value in the current policy population, which is called the best individual. The corresponding set of optimization variables is the optimal control policy sequence.

[0141] These control strategies are distributed to various actuators to achieve integrated intelligent control of heating, cooling, ventilation, lighting and other equipment.

[0142] This embodiment comprehensively perceives the environmental conditions and energy consumption within the building, laying a solid data foundation for subsequent refined control. Processing the raw data not only improves data quality but also reduces computational complexity. Secondly, the use of dynamic modeling methods accurately describes the complex dynamic behavior of the building's interior space and precisely predicts changes in energy consumption and comfort over a future period. This provides crucial support for developing forward-looking control strategies, facilitating proactive adjustment and optimization of the building's operational status. Furthermore, the integration of energy conservation and comfort objectives during the control strategy solution process achieves multi-objective optimization. By balancing these two objectives, a balanced optimal control strategy is obtained, maximizing the satisfaction of both energy conservation and comfort needs and avoiding the one-sidedness that may result from single-objective optimization. In addition, an improved multi-objective optimization algorithm is employed, possessing advantages such as strong global search capability, fast convergence speed, and good robustness. This algorithm efficiently solves for the optimal control strategy sequence under constraints, thereby improving control accuracy and performance. Finally, integrated intelligent control of equipment is achieved, not only improving the building's energy efficiency but also creating a comfortable indoor environment, comprehensively enhancing the building's overall performance.

[0143] Example 2

[0144] Please see Figure 2As shown, parts not described in detail in this embodiment are described in Embodiment 1. A building energy-saving intelligent integrated control method is provided, including:

[0145] Step 1: Collect comprehensive building data of the interior space areas of the building; preprocess the comprehensive building data to obtain standard building data;

[0146] Step 2: Establish a building dynamic model based on standard building data, and use the dynamic model to predict the energy consumption change curve and comfort change curve of the building's internal space area over a fixed period of time in the future.

[0147] Step 3: Using the energy consumption change curve and comfort change curve of the building's internal space area over a fixed future time period as constraints, solve for a series of optimal control strategy sequences; distribute the solved optimal control strategy sequences to the building's actuators to achieve integrated intelligent control of heating, cooling, ventilation, lighting and other equipment.

[0148] Example 3

[0149] This embodiment discloses an electronic device, 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 operation mode of the building energy-saving intelligent integrated control method provided above.

[0150] Since the electronic device described in this embodiment is the electronic device used to implement the building energy-saving intelligent integrated control method described in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the building energy-saving intelligent integrated control method described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any electronic device used by those skilled in the art to implement the building energy-saving intelligent integrated control method described in this application embodiment falls within the scope of protection of this application.

[0151] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0152] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A building energy-saving intelligent integrated control system, characterized in that, include: The data acquisition module is used to collect comprehensive building data of the interior space areas of a building; The building data is preprocessed to obtain standard building data; The comprehensive prediction module is used to build a building dynamic model based on standard building data, and use the dynamic model to predict the energy consumption change curve and comfort change curve of the building's internal space area over a fixed period of time in the future. The methods for establishing dynamic building models include: The interior space of the building is discretized into several grid units, each corresponding to a magnetic spin. The magnetic spin can take two states: up or down, representing the two states of the grid unit. There are several standard building data in each grid unit; based on the standard building data, the coupling strength between adjacent grid units is defined and ; wherein, is the distance between the grid units and the center of the grid unit , is any standard building data in the grid unit , is any standard building data in the grid unit , is the relative azimuth angle between the grid units and the grid unit ; is an anisotropy parameter, is a phase parameter; is an index parameter; is a strength adjustment parameter, is a strength adjustment index parameter; is a distance index parameter;​ External field terms are defined using environmental parameters and energy consumption data; that is... ; Grid unit Outdoor events, Grid unit The corresponding environment parameter vector, Grid unit The corresponding energy consumption data vector; For mapping functions; The standard building data of each grid cell is randomly divided into a training set and a test set according to a certain ratio, with the data in the training set serving as the training data. The loss function is to minimize the mean square error between the output and the training data. The standard building data of the p-th training data is defined as X_p, and the corresponding real environmental parameters and energy consumption data are obtained. Using the current coupling parameters, the output is obtained by simulating time evolution. The coupling parameters are initialized to a set of random values, and the training data is iterated several times. The iteration process is as follows: a small batch of training data is randomly sampled from the training set, the loss function is calculated for this small batch of training data, and the gradient of the loss function with respect to the coupling parameters is calculated. The coupling parameters are updated according to the gradient using stochastic gradient descent or its variant algorithm. The process continues until the preset maximum number of iterations is reached, the optimal coupling parameters are obtained, and the completed building dynamic model is obtained. The strategy solving module is used to solve for a series of optimal control strategy sequences by taking the energy consumption change curve and comfort change curve of the building's internal space area over a fixed future time period as constraints. The solved optimal control strategy sequences are then distributed to the actuators of the building. The modules are connected to each other via wired and / or wireless means to realize data transmission between the modules.

2. The building energy-saving intelligent integrated control system according to claim 1, characterized in that, The building's comprehensive data includes environmental parameters and energy consumption data within the building; the environmental parameters include temperature, humidity, carbon dioxide concentration, illuminance, and fresh air volume; the energy consumption data includes heating capacity, cooling capacity, and lighting power.

3. The building energy-saving intelligent integrated control system according to claim 2, characterized in that, The methods for preprocessing comprehensive building data include: The building data is cleaned and dimensionless processing is performed on the data of different dimensions to obtain preliminary building data. The preliminary building data is arranged into a corresponding data time series according to the time order of data types. The data time series is segmented to obtain m segments of building data. Standard building data is extracted from each of the m segments of building data.

4. The building energy-saving intelligent integrated control system according to claim 3, characterized in that, The methods for performing data segmentation include: For each data time series, it is regarded as a state sequence of a discrete time Markov chain, and is discretized into a finite number of states according to the range of data values of the data time series, that is, the state space of the Markov chain is determined; a count matrix C is initialized, all elements in the count matrix C have initial values of 0, the time series data is traversed, for each adjacent time point pair (Y t, Y t+1), wherein Y t represents the state at time t, Y t+1 represents the state at time t+1, the state indexes i and j corresponding to Y t and Y t+1 are determined; the value of the element C(i, j) in the count matrix C corresponding to the state indexes i and j is added by 1, that is, C(i, j) records the number of times of transition from the state z_i to the state z_j; after traversing the entire time series, for each state s_i, the frequency of its transition to other states is calculated, that is, C(i, j) is divided by the sum of the number of all transitions from the state z_i; all frequencies obtained, that is, all elements of the transition probability matrix P are obtained; Using the obtained transition probability matrix, for each time point t in the data time series, the conditional probability P(t, z_i) that the data value at the time point belongs to each state z_i is calculated; For each time point t, the maximum value P_max of the conditional probability that the data value at the time point belongs to each state is calculated; Based on the maximum value P_max, a preset probability threshold θ=a×P_max is determined, when P(t, z_i) is greater than or equal to θ, it is determined that the time point t belongs to the state z_i; a is a trade-off parameter; Scanning the entire data time series, adjacent time points belonging to the same state are grouped into one data segment, that is, a plurality of preliminary segmented data are obtained; the plurality of preliminary segmented data are subjected to cluster analysis to obtain m building data.

5. The building energy-saving intelligent integrated control system according to claim 4, characterized in that, The cluster analysis includes: A feature vector of each preliminary segmented data is calculated, and the feature vector is obtained by splicing a weighted mean vector, a variance vector, a mean vector, a maximum value, a minimum value, kurtosis and skewness; The calculation method of the weighted mean vector includes: For the first preliminary segmented data, its time range is , wherein, is the starting time point of the first preliminary segmented data, is the ending time point of the first preliminary segmented data; a time weight function is defined based on the ending time point ; and a formula of the time weight function is: ; wherein, is a decay coefficient, is a periodicity modulation coefficient, is a time weight at time t; is an initial time point of the function, is a period; The weighted mean vector of the first preliminary segmented data is calculated based on a time weight function ;​ ; wherein is a confidence weight for data point , data point is a data point of the preliminary segmented data at time ; wherein, is the mean of the preliminary segmented data, is the standard deviation of the neighborhood of, and is a confidence parameter; Based on the calculated feature vector, a distance matrix D is constructed, and an element D(I, J) in the distance matrix D represents the Mahalanobis distance between the Ith preliminary segmented data and the Jth preliminary segmented data; based on the distance matrix D, a merging iteration is performed to obtain m building data.

6. The building energy-saving intelligent integrated control system according to claim 5, characterized in that, The merging iteration based on the distance matrix D includes: Each preliminary segmented data in the distance matrix D is regarded as a separate cluster, a feature vector of each cluster is calculated, and a comprehensive distance between clusters is calculated based on the feature vector; the calculation formula of the comprehensive distance is: ; wherein, is the integrated distance between clusters and , is the eigenvector of cluster , is the eigenvector of cluster , is an adaptive covariance matrix, is the temporal center of cluster , is the temporal center of cluster , is a temporal decay parameter; The merging cost between clusters is calculated based on the comprehensive distance between clusters; the calculation formula of the merging cost is: ; wherein, is a cluster shape penalty term, is a cluster and a merging cost between clusters cluster shape penalty term ; wherein, is the variance of the cluster , is the variance of the cluster , is the skewness of the cluster , is the skewness of the cluster , and are learnable parameters; find the two clusters with the minimum merging cost among all current pairs of clusters and merge them into a new cluster ; the feature vector of the new cluster is computed based on the feature vectors of and , weighted by their sample sizes; compute the distances of the new cluster to other clusters and update the distance matrix; When the sum of the intra-cluster dispersion squares of all clusters is less than a preset merging threshold, the merging is stopped, and finally m clusters corresponding to m building data are obtained.

7. The building energy-saving intelligent integrated control system according to claim 6, characterized in that, The method for predicting the energy consumption change curve and the comfort change curve of the building interior space region in a future fixed time period by using the dynamic model includes: Obtaining a current value of standard building data, determining an initial state of each grid cell according to the current value of the standard building data; and composing a state vector with the initial states of all the grid cells; At time , the flipping spin probability of each lattice cell is calculated according to the current state vector and the optimal coupling parameters ; where, is a lattice unit at time spin state, is a lattice unit at time spin state, is a random perturbation term; Based on the spin flipping probability Monte Carlo sampling is used, corresponding to the lattice cell. Generate a uniformly random number in the interval [0, 1]. ;like Then the grid unit In time spin state Set as That is, spin flipping, otherwise the spin state remains unchanged; all lattice elements in time The spin states form a new state vector; this process is repeated until the predicted maximum time step is reached. This yields a series of state vectors; For each time step, the state vector corresponds to a set of environment parameter vectors and energy consumption data vectors are converted to and and by a back-mapping function; environment parameter vector all sampling points within the maximum time step constitute the environment parameter change curve; for each time step, the corresponding comfort index is calculated to constitute the comfort change curve; the comfort index is the percentage of dissatisfaction ; using the energy consumption data vector all sampling points within the maximum time step corresponding to the calculation of total energy consumption , that is, constitute the energy consumption change curve.

8. The building energy-saving intelligent integrated control system according to claim 7, characterized in that, The solving method of the optimal control strategy sequence comprises: The optimization objectives are defined to include energy consumption minimization and comfort maximization, i.e., to minimize the total energy consumption in a future period of time and to minimize a comfort index; the optimization objectives are expressed as two objective functions and ; ； ； Define optimization variables as control parameters; define constraint conditions including energy consumption variation curve The peak value of the energy consumption variation curve does not exceed a preset upper limit of energy consumption The peak-to-valley range of the comfort variation curve is within a preset comfort threshold interval Initializing a strategy population, determining a strategy population size N2 and a dimension n2 of optimization variables, and randomly generating a set of optimization variables as a position vector of each individual in the strategy population; Two objective functions and The weighted sum is used as the update function for the policy population; the individual is calculated. The value of the update function That is, the update function in The function value at that point is denoted as the updated value; Compute the centroid of the current population as a potential optimal solution; wherein, is the position vector of individual d, for each individual , compute the Euclidean distance of individual d to the centroid and the unit vector in the direction of the centroid ; For each individual , a random step is generated subject to a normal distribution , the position vector of the individual is updated according to the random step and the unit vector, resulting in a new position vector of the individual ; and the value of the update function for the new position vector is calculated ;​ ; wherein, is a constant parameter, is a random variable subject to a specific distribution; Arranging the updated values in ascending order to obtain an arranged strategy population, and dividing the arranged strategy population into two parts, a better individual part and a worse individual part; The number of better individuals is ; wherein and are dynamic adjustment parameters, is the current iteration number; is the rounding function; For each worse individual Randomly select a better individual Calculate the Euclidean distance between the two, denoted as the difference distance. Based on the gap distance, update the worse individuals. position vector To obtain worse individuals New position vector ; Calculate worse individuals New position vector The updated value; ; wherein, is the position vector of the better individual and the worse individual a random step generated in compliance with a normal distribution, is the position vector of the better individual . Repeating the iteration of updating the position vectors of the individuals and the worse individuals until the value of the updating function converges, and finally outputting an individual with the minimum updated value in the current strategy population, denoted as a best individual, and a set of optimization variables corresponding to the best individual as the optimal control strategy sequence.

9. A building energy-saving intelligent integrated control method, which is implemented based on the building energy-saving intelligent integrated control system according to any one of claims 1 to 8, characterized in that, Comprise: Step 1, collecting building comprehensive data of an internal space region of a building; Pretreating the building comprehensive data to obtain standard building data; Step 2, establishing a building dynamic model according to the standard building data, and predicting an energy consumption change curve and a comfort change curve of the internal space region of the building in a future fixed time period by using the dynamic model; Step 3, taking the energy consumption change curve and the comfort change curve of the internal space region of the building in the future fixed time period as constraints, and solving a series of optimal control strategy sequences; and distributing the optimal control strategy sequences obtained by solving to each actuator of the building.

Images