Building energy-saving control system and method

Abstract

The invention relates to the technical field of building energy saving, and discloses a building energy-saving control system and method.The method comprises the steps that firstly, multi-source environment data inside and outside a building is collected, and a time-space fusion data set is generated through timestamp synchronization and data alignment; dynamic energy consumption characteristics are extracted, and the future energy consumption state of each region is predicted by using a hidden Markov model; decomposing a historical energy consumption mode according to a prediction result and a non-negative matrix factorization algorithm to generate a multi-objective optimization control strategy; and finally, adjusting an equipment operation sequence through a dynamic priority scheduling algorithm and outputting an instruction. The system comprises a multi-source data acquisition module, a feature extraction module, a state prediction module, a strategy generation module, an equipment scheduling module and a feedback correction module. According to the method, energy consumption can be accurately predicted, multi-target optimization control, dynamic equipment scheduling and online strategy correction are realized, the building energy utilization efficiency is effectively improved, energy consumption is reduced, the comfort level is guaranteed, and equipment loss is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of building energy conservation, and specifically to a building energy conservation control system and method. Background Art

[0002] With the continuous growth of global energy demand and the increasingly severe environmental problems, building energy conservation has become an important topic for sustainable development. Building energy consumption accounts for a relatively large proportion of the total social energy consumption. Therefore, how to effectively reduce building energy consumption and improve building energy utilization efficiency has become an urgent problem to be solved in the current building field.

[0003] Traditional building energy conservation methods mainly rely on building design and equipment selection, such as using energy-saving lamps, high-efficiency air-conditioning systems, etc. Although these methods can reduce building energy consumption to a certain extent, they often lack consideration of dynamic changing factors during the actual operation of the building. The environmental conditions inside the building (such as light intensity, temperature and humidity, personnel density, etc.) and equipment energy consumption are constantly changing over time, and it is difficult for traditional methods to perform precise energy conservation control based on these real-time changing data.

[0004] In addition, most of the existing building energy conservation control systems make control decisions based on single factors or simple rules. For example, some systems only control the operation of the air conditioner according to the indoor temperature, ignoring the influence of other factors (such as light, personnel activities, etc.) on energy consumption. This single-factor control method cannot fully exploit the potential of building energy conservation and is difficult to achieve the overall optimal energy conservation effect. Moreover, due to the lack of in-depth analysis of historical energy consumption data and accurate prediction of future energy consumption, the existing control systems often lack foresight when formulating energy conservation strategies and cannot adjust the equipment operation status in advance to adapt to different energy consumption demands.

[0005] In terms of data processing, there are a large number of sensors of different types and different collection frequencies in the building. The data collected by these sensors have clock deviations and frequency differences. If these data cannot be effectively synchronized and aligned, it will lead to a reduction in the accuracy and availability of the data, thereby affecting subsequent analysis and decision-making. At the same time, facing the massive building data, how to extract valuable information from it and accurately reflect the dynamic characteristics of building energy consumption is also an urgent problem to be solved.

[0006] In terms of formulating energy conservation strategies, existing methods often have difficulty in balancing multiple objectives such as equipment energy consumption, indoor comfort, and equipment switching frequency. Some energy conservation strategies may cause a significant decrease in indoor comfort while reducing equipment energy consumption, affecting the user experience; while overly focusing on comfort may not achieve the ideal energy conservation effect. In addition, frequent switching of equipment will increase the wear and maintenance costs of the equipment and shorten the service life of the equipment, but the existing control systems rarely fully consider this factor when formulating strategies. Summary of the Invention

[0007] The purpose of the present invention is to provide a building energy-saving control system and method to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A building energy-saving control method, the method includes:

[0009] Collect multi-source environmental data inside and outside the building, perform timestamp synchronization and data alignment on the multi-source environmental data, and generate a spatio-temporal fusion data set;

[0010] Extract dynamic energy consumption characteristics based on the spatio-temporal fusion data set, and predict the future energy consumption status of each area in the building through a hidden Markov model;

[0011] Decompose the historical energy consumption pattern according to the prediction result and the non-negative matrix factorization algorithm, and generate a multi-objective optimization control strategy;

[0012] Adjust the device operation sequence based on the dynamic priority scheduling algorithm, and output control instructions to the building device execution end.

[0013] Preferably, the multi-source environmental data includes light intensity, temperature and humidity, personnel density, and device energy consumption data; the method for timestamp synchronization and data alignment includes:

[0014] Correct the clock deviation of the data collected by each sensor through linear interpolation and align it to a unified time reference; based on the data acquisition frequency difference, use the sliding window weighted average method to perform high-frequency compensation on low-frequency data.

[0015] Preferably, the method for extracting the dynamic energy consumption characteristics includes:

[0016] Perform segmented standardization processing on the spatio-temporal fusion data set to eliminate the dimension difference; use discrete wavelet transform to decompose the time-frequency components of each dimension data, extract the energy entropy as the feature vector; perform dimensionality reduction on the high-dimensional feature vector through principal component analysis, and retain the components with a variance contribution rate greater than the threshold.

[0017] Preferably, the method for the hidden Markov model to predict the future energy consumption status includes:

[0018] Divide the historical energy consumption status into a discrete hidden state set, and the observation sequence is the dynamic energy consumption characteristics; iteratively optimize the state transition matrix and the observation probability matrix through the Baum-Welch algorithm; use the Viterbi algorithm to decode the hidden state sequence of the future N time steps, where N is a positive integer.

[0019] Preferably, the method for the non-negative matrix factorization algorithm to decompose the historical energy consumption pattern includes:

[0020] Construct a historical energy consumption data matrix, where the rows represent timestamps and the columns represent device types; solve the basis matrix and the coefficient matrix by the alternating least squares method, so that the column vectors of the basis matrix represent independent energy consumption patterns, and the coefficient matrix represents pattern weights; introduce a sparsity constraint term to enhance pattern interpretability.

[0021] Preferably, the method for generating the multi-objective optimization control strategy includes:

[0022] Define device energy consumption, comfort deviation, and device switching frequency as optimization objectives; use an improved particle swarm optimization algorithm to solve the Pareto front, where the inertia weight dynamically decays with the number of iterations, and the individual learning factor is updated using Gaussian perturbation.

[0023] Preferably, the implementation method of the dynamic priority scheduling algorithm includes:

[0024] Set the basic priority according to the device type, with the air conditioning system having a higher priority than the lighting system; calculate the ratio of the device energy consumption to the comfort contribution in real time, and dynamically increase the priority of the device with a higher contribution; use a directed acyclic graph topological sorting to avoid device instruction conflicts.

[0025] Preferably, it further includes an online correction method for the control strategy:

[0026] Collect the actual energy consumption data feedback from the device execution end, and calculate the residual with the predicted value; when the residual exceeds the threshold, trigger the incremental support vector regression model to update the observation probability matrix of the hidden Markov model.

[0027] Preferably, the update method of the incremental support vector regression model includes:

[0028] Map the residual data to a high-dimensional kernel space, and retain the support vector set and Lagrange multipliers; screen the boundary samples in the new data through the KKT conditions, and only perform model parameter iterative updates on the boundary samples.

[0029] Preferably, the present invention further includes a building energy-saving control system, and the system includes:

[0030] A multi-source data acquisition module for synchronously acquiring building internal and external environment data;

[0031] A feature extraction module connected to the data acquisition module for generating dynamic energy consumption features;

[0032] A state prediction module for outputting the future energy consumption state based on the hidden Markov model;

[0033] A strategy generation module for generating a control strategy using non-negative matrix factorization and a multi-objective optimization algorithm;

[0034] A device scheduling module for sending instructions to the execution end according to the dynamic priority scheduling algorithm;

[0035] The feedback correction module updates the prediction model parameters online according to the actual energy consumption data.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] The present invention collects multi-source environmental data inside and outside the building, including light intensity, temperature and humidity, personnel density, and equipment energy consumption data, etc., and synchronizes the timestamps and aligns the data of these data to generate a spatio-temporal fusion dataset. Based on this dataset, dynamic energy consumption characteristics are extracted, and the hidden Markov model is used to predict the future energy consumption status of each area in the building. Compared with the traditional prediction methods based on single factor or simple rules, the present invention can comprehensively consider various factors affecting energy consumption and improve the accuracy of energy consumption prediction. For example, through the analysis of historical data, it is found that in the case of strong light intensity and low personnel density, there are specific changing trends in the lighting energy consumption and air-conditioning energy consumption of a certain area. The model of the present invention can capture these complex relationships, so as to more accurately predict the future energy consumption and provide a reliable basis for formulating energy-saving strategies in advance.

[0038] According to the prediction results and the non-negative matrix factorization algorithm to decompose the historical energy consumption pattern, a multi-objective optimization control strategy is generated. This strategy defines the equipment energy consumption, comfort deviation degree, and equipment switching frequency as optimization objectives, and uses an improved particle swarm optimization algorithm to solve the Pareto front. This enables, when formulating the control strategy, to reduce the equipment energy consumption while ensuring that the indoor comfort is within a reasonable range and reducing the equipment switching frequency. For example, during the peak period of air-conditioning use in summer, the system can optimize the operation combination of air-conditioning and ventilation equipment according to the indoor and outdoor temperature, personnel distribution, etc., not only meeting the comfort requirements of indoor personnel but also avoiding the frequent start and stop of equipment, reducing equipment loss and energy consumption, and achieving the balance between energy saving and comfort.

[0039] Based on the dynamic priority scheduling algorithm, the equipment operation sequence is adjusted. The basic priority is set according to the equipment type, such as the air-conditioning system having a higher priority than the lighting system, and at the same time, the ratio of the equipment energy consumption to the comfort contribution degree is calculated in real time, and the priority of the equipment with a high contribution degree is dynamically increased. The directed acyclic graph topological sorting is used to avoid equipment instruction conflicts. This scheduling method can reasonably arrange the operation sequence of equipment according to the real-time state of the building and improve the equipment operation efficiency. For example, in the scenario of a crowded meeting room, when it is detected that people enter, the system will give priority to increasing the priority of air-conditioning and lighting equipment, quickly adjusting the indoor environment to meet the usage needs of people, and at the same time ensuring the coordinated work between equipment and avoiding conflicts.

[0040] The present invention also includes an online correction method for the control strategy. The actual energy consumption data fed back by the execution end of the acquisition device is collected, and the residual between the actual value and the predicted value is calculated. When the residual exceeds the threshold, an incremental support vector regression model is triggered to update the observation probability matrix of the hidden Markov model. In this way, the system can adjust the prediction model and control strategy in a timely manner according to the actual operation conditions, improving the adaptability and accuracy of the system. For example, when temporary decoration is carried out in a certain area of the building, resulting in changes in personnel activities and equipment usage, the system can respond quickly, update the prediction model, and make the subsequent control strategy more in line with the actual situation, continuously ensuring the energy-saving effect.

[0041] In the data acquisition and processing link, the clock deviation of the data collected by each sensor is corrected by linear interpolation and aligned to a unified time reference. The sliding window weighted average method is used to perform high-frequency compensation on low-frequency data, ensuring the accuracy and consistency of the data. In terms of feature extraction, the spatio-temporal fusion data set is subjected to segmented normalization processing, and effective features are extracted using methods such as discrete wavelet transform and principal component analysis, which can more accurately reflect the dynamic changes of building energy consumption. These data processing methods provide high-quality data support for subsequent prediction and control, improving the performance of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is the working principle diagram of the building energy-saving control method described in the present invention;

[0043] Figure 2 is the working flow chart of multi-source environmental data acquisition and processing;

[0044] Figure 3 is the working flow chart of the non-negative matrix factorization algorithm;

[0045] Figure 4 is the working flow chart of the dynamic priority scheduling algorithm;

[0046] Figure 5 is the working flow chart of the online correction of the control strategy. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] Please refer to Figures 1-5 , the present invention provides a technical solution: a building energy-saving control system and method, the method includes:

[0049] Various sensors such as light sensors, temperature and humidity sensors, personnel density sensors, and equipment energy consumption sensors are installed inside and outside the building to collect multi-source environmental data. Due to the differences in the clocks and acquisition frequencies of different sensors, the clock deviations of the data collected by each sensor need to be corrected by linear interpolation and aligned to a unified time reference; at the same time, based on the differences in data acquisition frequencies, the sliding window weighted average method is used to perform high-frequency compensation on low-frequency data to generate a spatio-temporal fusion data set.

[0050] The spatio-temporal fusion data set is subjected to piecewise normalization to eliminate the dimensional differences. The time-frequency components of the data in each dimension are decomposed by discrete wavelet transform, and the energy entropy is extracted as the feature vector. The high-dimensional feature vector is reduced in dimension by principal component analysis, and the components with a variance contribution rate greater than the threshold are retained to obtain the dynamic energy consumption characteristics. The historical energy consumption states are divided into a discrete hidden state set. Taking the dynamic energy consumption characteristics as the observation sequence, the state transition matrix and the observation probability matrix of the hidden Markov model are iteratively optimized by the Baum-Welch algorithm, and the Viterbi algorithm is used to decode the hidden state sequence in the next N time steps to predict the future energy consumption states of each area in the building.

[0051] A historical energy consumption data matrix is constructed, where the rows represent timestamps and the columns represent equipment types. The basis matrix and the coefficient matrix are solved by the alternating least squares method, a sparsity constraint term is introduced, and the historical energy consumption patterns are decomposed by the non-negative matrix factorization algorithm. The equipment energy consumption, comfort deviation, and equipment switching frequency are defined as the optimization objectives, and the improved particle swarm optimization algorithm is used to solve the Pareto front to generate a multi-objective optimization control strategy.

[0052] The basic priorities are set according to the equipment types, for example, the air conditioning system has a higher priority than the lighting system. The ratio of the equipment energy consumption to the comfort contribution is calculated in real time, and the priority of the equipment with a high contribution is dynamically increased. The directed acyclic graph topological sorting is used to avoid equipment instruction conflicts, and the equipment operation sequence is adjusted based on the dynamic priority scheduling algorithm, and the control instructions are output to the building equipment execution end.

[0053] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0054] Embodiment 1:

[0055] In this embodiment, the multi-source environmental data includes light intensity, temperature and humidity, personnel density, and equipment energy consumption data. These data are collected by different types of sensors. For example, light intensity sensors are installed near windows, on rooftops, etc. of buildings to monitor the external light conditions in real time; temperature and humidity sensors are distributed in various rooms, corridors, etc. inside the building to obtain the indoor temperature and humidity information; personnel density sensors can be installed at doorways, passages, etc. to count the personnel flow through technologies such as infrared sensing and image recognition, and then estimate the personnel density; equipment energy consumption sensors are connected to various building equipment to collect the energy consumption data of the equipment in real time.

[0056] During the collection process, due to the possible deviation of the clocks of each sensor, the collected data may have inconsistent times. Therefore, a method of correcting the clock deviation of the data collected by each sensor through linear interpolation is adopted to align all the data to a unified time reference. Assuming that there is a time deviation between the data collected by sensor A and sensor B, based on an accurate time source, the accurate time points of sensor A and sensor B under the unified time reference are calculated through linear interpolation, so as to achieve timestamp synchronization.

[0057] In addition, there are also differences in the data collection frequencies of different sensors. For low-frequency data, the sliding window weighted average method is used for high-frequency compensation. For example, the collection frequency of a certain equipment energy consumption sensor is low, while the collection frequency of the temperature and humidity sensor is high. Taking the collection frequency of the temperature and humidity sensor as a reference, a sliding window is set, and the low-frequency equipment energy consumption data is weighted and averaged within the window. Assuming that the size of the sliding window is n, and the data within the window is x1, x2, …, x n , and the corresponding weights are w1, w2, …, w n , and the high-frequency compensated data after weighted average The weight w i is determined according to factors such as the time correlation of the data. Usually, the data closer to the current time has a greater weight, so as to improve the time resolution of the low-frequency data and make it match the high-frequency data on the time scale, and finally generate a spatio-temporal fusion data set.

[0058] Embodiment 2:

[0059] In this embodiment, dynamic energy consumption characteristics are extracted from the spatio-temporal fusion data set. The spatio-temporal fusion data set is subjected to piecewise normalization processing to eliminate the dimensional differences between different data dimensions. Different types of environmental data, such as light intensity, temperature and humidity, and equipment energy consumption data, have different dimensions and value ranges. Through piecewise normalization processing, the data is mapped to the same scale range, making different data comparable. For example, for the light intensity data x within a certain time period, the formula Standardize the data, where min(x) and max(x) are the minimum and maximum values of the light intensity data during this period, and the processed data x ′ ranges from 0 to 1.

[0060] Use discrete wavelet transform to decompose the time-frequency components of each dimension data and extract energy entropy as the feature vector. Discrete wavelet transform can decompose the signal at different time and frequency scales, thus revealing the time-frequency characteristics of the signal. For each dimension data after standardization, such as temperature and humidity data, it is decomposed into sub-band signals of different frequencies through discrete wavelet transform. Then, calculate the energy entropy of each sub-band signal. Energy entropy can reflect the uncertainty and complexity of the signal, and as a feature vector, it can effectively characterize the dynamic change characteristics of the data. Assume the energy of a certain sub-band signal is E i , then the energy entropy H of this sub-band signal is H = -∑ i p i log2p i , where

[0061] Reduce the dimension of the high-dimensional feature vector through principal component analysis. Since the extracted energy entropy feature vector has a high dimension and contains a large amount of redundant information, directly using it will increase the computational complexity and may affect the model performance. Principal component analysis can project the high-dimensional data onto a few mutually orthogonal principal components, and these principal components can retain most of the variance information of the original data. Set a variance contribution rate threshold and retain the principal components with a variance contribution rate greater than this threshold. For example, after calculation, the sum of the variance contribution rates of the first three principal components is greater than 90%, then retain these three principal components, reduce the high-dimensional feature vector to three dimensions, which not only reduces the data dimension but also retains the key information, facilitating the subsequent prediction of the energy consumption state.

[0062] Example 3:

[0063] This example focuses on the method of predicting the future energy consumption state using the hidden Markov model. Through this method, the future energy consumption trends of each area in the building can be accurately predicted, providing an important basis for formulating energy-saving control strategies.

[0064] In this example, use the hidden Markov model to predict the future energy consumption state of each area in the building. Divide the historical energy consumption state into a discrete hidden state set. According to the actual situation and experience of building energy consumption, divide the energy consumption state into several different levels, such as low energy consumption, medium energy consumption, and high energy consumption, and these levels constitute the discrete hidden state set. The observation sequence is the feature vector obtained through the above dynamic energy consumption feature extraction method.

[0065] Iteratively optimize the state transition matrix and the observation probability matrix through the Baum-Welch algorithm. The state transition matrix describes the probabilities of transitioning between different hidden states, and the observation probability matrix represents the probabilities of observing specific observation sequences under different hidden states. Assume there are N states in the hidden state set, and the state transition matrix A = (a ij ), where a ij represents the probability of transitioning from state i to state j; the observation probability matrix B = (b jk ), where b jk represents the probability of observing the observation value k under state j. The Baum-Welch algorithm iteratively calculates new model parameters based on the observation sequence and the current model parameters (state transition matrix and observation probability matrix) to maximize the likelihood probability of the model for the observation sequence.

[0066] Adopt the Viterbi algorithm to decode the hidden state sequence for the next N time steps. The Viterbi algorithm is a dynamic programming algorithm that can quickly find the most likely hidden state sequence given an observation sequence when the state transition matrix and the observation probability matrix are known. For example, to predict the energy consumption states for the next 5 time steps, the Viterbi algorithm will calculate the most likely hidden state at each time step based on the optimized state transition matrix and observation probability matrix, thus obtaining the hidden state sequence for the next 5 time steps. This sequence reflects the future energy consumption trends of each area in the building and provides predictive data support for formulating subsequent energy-saving control strategies.

[0067] Example 4:

[0068] In this example, first use the non-negative matrix factorization algorithm to decompose the historical energy consumption patterns. Construct a historical energy consumption data matrix, where the rows of the matrix represent timestamps and the columns represent equipment types. For example, with a one-day time span and one-hour timestamps, there will be 24 rows in the matrix; assuming there are equipment such as air conditioning systems, lighting systems, and elevator systems in the building, the number of columns in the matrix is the number of equipment types.

[0069] Solve the basis matrix and the coefficient matrix through the alternating least squares method. During the solution process, make the column vectors of the basis matrix represent independent energy consumption patterns and the coefficient matrix represent pattern weights. The alternating least squares method is an iterative algorithm. In each iteration, fix one of the matrices and solve the other matrix, and keep alternating until the convergence condition is met. For example, first fix the coefficient matrix and solve the basis matrix; then fix the basis matrix and solve the coefficient matrix, and obtain relatively accurate basis matrix and coefficient matrix through multiple iterations.

[0070] To enhance the interpretability of the model, a sparsity constraint term is introduced. The sparsity constraint term can make most of the elements in the coefficient matrix approach zero, making it easier to discover the main energy consumption patterns and corresponding key devices when analyzing the energy consumption patterns. Assuming the coefficient matrix is W, the sparsity constraint term λ∑ i,j |W ij | is introduced, where λ is a parameter for adjusting the sparsity degree. During the solution process, by adjusting the value of λ, the coefficient matrix is made to have an appropriate sparsity.

[0071] Next, a multi-objective optimization control strategy is generated. The device energy consumption, comfort deviation degree, and device switching frequency are defined as the optimization objectives. The device energy consumption directly reflects the energy consumption situation of the building; the comfort deviation degree is calculated based on the differences between parameters such as indoor temperature and humidity, light intensity, etc. and the human comfort standard range, and is used to measure the comfort level of users; the device switching frequency takes into account the impact of frequent device switching on the device lifespan and energy consumption.

[0072] An improved particle swarm optimization algorithm is used to solve the Pareto front. In the improved particle swarm optimization algorithm, the inertia weight dynamically decays with the number of iterations, and the individual learning factor is updated using Gaussian perturbation. The dynamic decay of the inertia weight enables the algorithm to have strong global search ability in the early stage and strong local search ability in the later stage. Assuming the initial value of the inertia weight ω is ω0, as the number of iterations t increases, where ω min is the minimum value of the inertia weight, and T is the maximum number of iterations. The individual learning factor is updated using Gaussian perturbation. In each iteration, Gaussian noise is added to the individual learning factor to increase the diversity of the algorithm and avoid falling into local optimal solutions. The Pareto front obtained by solving with this algorithm contains multiple optimal solutions that trade off between different objectives. Select a suitable control strategy from the Pareto front according to the actual requirements to achieve the balance between building energy conservation and user comfort.

[0073] Example 5:

[0074] In this example, first, the dynamic priority scheduling algorithm is implemented. The basic priority is set according to the device type. For example, considering the greater impact of the air conditioning system on indoor comfort and energy consumption, the basic priority of the air conditioning system is set higher than that of the lighting system. During the actual operation process, the ratio of the device energy consumption to the comfort contribution degree is calculated in real time. For the air conditioning system, by calculating the ratio of its energy consumption to the comfort improvement brought by indoor temperature and humidity regulation, its energy efficiency ratio is measured; for the lighting system, the ratio of its energy consumption to the provided lighting comfort is calculated. Dynamically increase the priority of devices with high contribution degrees. For example, if at a certain moment the ratio of the energy consumption of the air conditioning system to the comfort contribution degree is low, it means that it can provide a high comfort improvement with less energy consumption. At this time, increase the priority of the air conditioning system.

[0075] To avoid device instruction conflicts, directed acyclic graph topological sorting is adopted. Various devices in the building are regarded as nodes in the directed acyclic graph, and the control relationships between devices are regarded as directed edges. For example, when adjusting the operating mode of the air conditioning system, it may affect devices such as the ventilation system associated with it, and the relationships between these devices form directed edges. Through the directed acyclic graph topological sorting algorithm, the scheduling order of devices is determined to ensure that the execution of device instructions does not generate conflicts and to achieve the orderly operation of devices.

[0076] In addition, it also includes an online correction method for control strategies. The actual energy consumption data fed back by the execution end of the device is collected, and the residual between the actual value and the predicted value is calculated. When the residual exceeds the threshold, an incremental support vector regression model is triggered to update the observation probability matrix of the hidden Markov model. Suppose the predicted energy consumption value of a certain device is y, and the actually collected energy consumption value is y, and the residual e = y - y. If |e| > the threshold, the correction mechanism is started.

[0077] The update method of the incremental support vector regression model is as follows: The residual data is mapped to a high-dimensional kernel space, and the support vector set and Lagrange multipliers are retained. Support vectors are data points that play a key role in the model decision boundary, and Lagrange multipliers are used for constraint conditions in the solution process. The boundary samples in the newly added data are screened through the KKT conditions, and only the boundary samples are used to iteratively update the model parameters. This can reduce the computational amount and improve the efficiency of model update. For example, at a certain moment, new residual data is collected. According to the KKT conditions, it is judged which data belong to the boundary samples, and then the model parameters corresponding to these boundary samples are updated, so that the observation probability matrix of the hidden Markov model can more accurately reflect the actual situation, improve the accuracy of future energy consumption state prediction, and further optimize the control strategy to achieve more precise building energy-saving control.

[0078] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. [[ID=1३]]

[0079] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood 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 building energy-saving control method, characterized in that: The following steps are involved: Collect multi-source environmental data inside and outside the building, synchronize timestamps and align data on the multi-source environmental data, and generate a spatiotemporal fusion dataset; Extracting dynamic energy consumption characteristics based on the spatiotemporal fusion data set and predicting the future energy consumption status of each area in the building through a hidden Markov model; Decompose historical energy consumption patterns based on prediction results and non-negative matrix factorization algorithm to generate multi-objective optimization control strategies; Adjust the equipment operation sequence based on the dynamic priority scheduling algorithm and output control instructions to the execution end of the building equipment.

2. The building energy-saving control method according to claim 1, characterized in that: The multi-source environmental data includes light intensity, temperature and humidity, personnel density and equipment energy consumption data; The method for timestamp synchronization and data alignment includes: The clock deviation of the data collected by each sensor is corrected by linear interpolation and aligned to a unified time base; Based on the difference in data acquisition frequency, the sliding window weighted average method is used to perform high-frequency compensation on low-frequency data.

3. The building energy-saving control method according to claim 1, characterized in that: The method for extracting the dynamic energy consumption characteristics includes: The spatiotemporal fusion dataset is segmented and standardized to eliminate dimensional differences; the time-frequency components of each dimensional data are decomposed using discrete wavelet transform, and the energy entropy is extracted as the feature vector; the high-dimensional feature vector is reduced in dimensionality through principal component analysis, and the components with variance contribution greater than the threshold are retained.

4. The building energy-saving control method according to claim 1, characterized in that: The method for predicting future energy consumption state using a hidden Markov model includes: The historical energy consumption state is divided into a discrete hidden state set, and the observation sequence is the dynamic energy consumption feature. The state transition matrix and the observation probability matrix are iteratively optimized using the Baum-Welch algorithm. The Viterbi algorithm is used to decode the hidden state sequence of the next N time steps, where N is a positive integer.

5. The building energy-saving control method according to claim 1, characterized in that: The method for decomposing the historical energy consumption pattern using the non-negative matrix decomposition algorithm includes: A historical energy consumption data matrix is constructed, in which rows represent timestamps and columns represent device types. The basis matrix and coefficient matrix are solved by alternating least squares method, so that the column vectors of the basis matrix represent independent energy consumption patterns and the coefficient matrix represents pattern weights. Sparsity constraints are introduced to enhance pattern interpretability.

6. The building energy-saving control method according to claim 1, characterized in that: The method for generating the multi-objective optimization control strategy includes: Equipment energy consumption, comfort deviation and equipment switching frequency are defined as optimization objectives. An improved particle swarm optimization algorithm is used to solve the Pareto frontier, in which the inertia weight dynamically decays with the number of iterations, and the individual learning factor is updated using Gaussian perturbation.

7. The building energy-saving control method according to claim 1, characterized in that: The implementation method of the dynamic priority scheduling algorithm includes: Set basic priority based on device type, with air conditioning systems taking precedence over lighting systems. Calculate the ratio of device energy consumption to comfort contribution in real time, and dynamically increase the priority of devices with high contribution. Use directed acyclic graph topology sorting to avoid device instruction conflicts.

8. The building energy-saving control method according to claim 1, characterized in that: It also includes online correction methods for control strategies: The actual energy consumption data fed back by the execution end of the collection device is calculated, and the residual with the predicted value is calculated; when the residual exceeds the threshold, the incremental support vector regression model is triggered to update the observation probability matrix of the hidden Markov model.

9. The building energy-saving control method according to claim 8, characterized in that: The updating method of the incremental support vector regression model includes: The residual data is mapped to a high-dimensional kernel space, retaining the support vector set and Lagrange multipliers; the boundary samples in the newly added data are filtered through the KKT condition, and the model parameters are iteratively updated only for the boundary samples.

10. A building energy-saving control system, characterized in that: include: Multi-source data acquisition module, used to synchronously collect environmental data inside and outside the building; a feature extraction module, connected to the data acquisition module, for generating dynamic energy consumption features; State prediction module, which outputs future energy consumption state based on hidden Markov model; Strategy generation module, which uses non-negative matrix decomposition and multi-objective optimization algorithm to generate control strategies; The device scheduling module sends instructions to the execution end according to the dynamic priority scheduling algorithm; Feedback correction module updates the prediction model parameters online according to actual energy consumption data.

Images