A green building energy-saving design method and system based on deep learning

Abstract

The application discloses a kind of green building energy-saving design method and system based on deep learning, belong to data processing technical field, by using deep learning model to construct green building energy consumption prediction model, and with the historical sample data as data support, using multi-strategy collaborative optimization algorithm to optimize green building energy consumption prediction model, obtain the green building energy consumption prediction model after optimization, then can dispatch the green building energy consumption prediction model after optimization to real-time building energy consumption feature analysis, realize energy consumption prediction, finally can according to prediction energy consumption data carry out energy-saving management, not only can effectively improve energy-saving management efficiency, real-time green building energy-saving management, it is more conducive to staff to carry out energy-saving management design.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a green building energy-saving design method and system based on deep learning. BACKGROUND

[0002] The building industry plays a decisive role in global energy consumption and greenhouse gas emissions, and is a key field for achieving energy-saving and emission-reducing targets. With the acceleration of urbanization, the electricity consumption and carbon emissions of residential buildings have increased sharply. Precise prediction and analysis of the electricity consumption of residential buildings can help develop effective energy-saving and emission-reducing strategies, which is of great significance for energy-saving and emission-reducing. With the rapid development of China's economy, the problem of building energy consumption is becoming increasingly serious. Green building energy-saving design has become an important means to reduce building energy consumption and improve building environmental quality. However, the traditional building energy-saving design method has the following shortcomings: long design cycle and low efficiency; strong dependence on designer experience, and uneven design quality; and difficulty in realizing real-time optimization of building energy consumption and environment. SUMMARY

[0003] The application provides a green building energy-saving design method and system based on deep learning, which solves the problems of low energy-saving design efficiency and difficulty in realizing real-time optimization in the prior art.

[0004] In a first aspect, the application provides a green building energy-saving design method based on deep learning, comprising:

[0005] Collecting historical building energy consumption characteristics corresponding to a target green building and historical energy consumption data corresponding to the historical building energy consumption characteristics, and obtaining historical sample data after preprocessing the historical building energy consumption characteristics and the historical energy consumption data corresponding to the historical building energy consumption characteristics;

[0006] Using a deep learning model to construct a green building energy consumption prediction model, using the historical sample data as data support, and using a multi-strategy collaborative optimization algorithm to optimize the green building energy consumption prediction model to obtain an optimized green building energy consumption prediction model;

[0007] Collecting real-time building energy consumption characteristics corresponding to the target green building, analyzing the real-time building energy consumption characteristics by using the optimized green building energy consumption prediction model, determining predicted energy consumption data, and performing green building energy-saving management according to the predicted energy consumption data.

[0008] Further, collecting historical building energy consumption characteristics corresponding to a target green building and historical energy consumption data corresponding to the historical building energy consumption characteristics, and obtaining historical sample data after preprocessing the historical building energy consumption characteristics and the historical energy consumption data corresponding to the historical building energy consumption characteristics, comprising:

[0009] Collect the daily average temperature, daily maximum temperature, daily minimum temperature, date type, season type and weather of the target green building on any day to obtain the historical building energy consumption characteristics corresponding to the green building;

[0010] For the historical building energy consumption characteristics corresponding to any green building, collect the historical energy consumption data corresponding to the historical building energy consumption characteristics;

[0011] After normalizing the historical building energy consumption characteristics, the normalized historical building energy consumption characteristics are obtained, and the normalized historical building energy consumption characteristics are constructed as input samples, and the historical energy consumption data corresponding to the historical building energy consumption characteristics are constructed as expected output labels to obtain historical sample data.

[0012] Further, a deep learning model is used to construct a green building energy consumption prediction model, including: using a CNN-LSTM model to construct a green building energy consumption prediction model.

[0013] Further, using the historical sample data as data support, a multi-strategy collaborative optimization algorithm is used to optimize the green building energy consumption prediction model to obtain an optimized green building energy consumption prediction model, including:

[0014] The hyperparameters of the green building energy consumption prediction model are initialized by using a chaotic mapping initialization method or a random initialization method, and the hyperparameters are encoded to obtain an encoding vector, and a plurality of different encoding vectors are obtained;

[0015] The input samples in the historical sample data are used as actual inputs, and the expected output labels in the historical sample data are used as expected outputs to obtain the fitness of each encoding vector;

[0016] According to the order of the fitness of the encoding vectors from large to small, all the encoding vectors are sequentially divided into a best region search population, a collaborative balance search population and a fast global search population;

[0017] For any first target encoding vector in the best region search population, a boundary probability transfer strategy is used to search the first target encoding vector to obtain a searched first target encoding vector;

[0018] For any second target encoding vector in the collaborative balance search population, a spiral balance search strategy is used to search the second target encoding vector to obtain a searched second target encoding vector;

[0019] For any third target encoding vector in the fast global search population, an optimal attraction and worst alert search strategy is used to search the third target encoding vector to obtain a searched third target encoding vector;

[0020] Determine whether the optimization termination condition is met. If so, determine the optimized green building energy consumption prediction model based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search. Otherwise, return to the step of determining the fitness corresponding to each encoding vector.

[0021] Furthermore, for any first target encoding vector in the optimal region search population, a boundary probability transition strategy is used to search the first target encoding vector to obtain the searched first target encoding vector, including:

[0022] For any first target encoding vector in the optimal region search population, based on the boundaries of each dimension's hyperparameters, the boundary transition value corresponding to the first target encoding vector is obtained as follows:

[0023]

[0024] in, Indicates the first t In the optimization process, the i-th first target encoding vector corresponds to the th... j dimensional hyperparameters, Indicates the first i The first target encoding vector corresponding to the first j Dimensional hyperparameters The corresponding boundary transition values ​​are i=1,2,…,L, where L represents the total number of the first target encoding vectors, and j=1,2,…,K, where K represents the total dimension of the hyperparameters. This represents the upper bound of the j-th dimension hyperparameter. This represents the lower bound of the j-th dimension hyperparameter. Represents the first random number between (0,1). This represents the second random number between (0,1). The first element of the optimal encoding vector is... j dimensional hyperparameters;

[0025] For any first target encoding vector in the optimal region search population, obtain the neighborhood search value of the first target encoding vector:

[0026]

[0027] in, This represents the neighborhood search value of the first target encoding vector. This represents the j-th dimension hyperparameter corresponding to a random other first target encoding vector. Represents pi (π). This indicates the preset maximum number of optimization iterations, and cos represents the cosine function. This represents a third random number between (0, 0.55);

[0028] The first decision factor is generated as follows: ;in, Indicates the first decision factor. This represents the control hyperparameter corresponding to the first decision factor, and is set to 0.499; This represents an exponential function with the natural constant e as its base.

[0029] Generate a fourth random number between (0,1) And determine the fourth random number Is it less than the first decision factor? If so, the neighborhood search value is used as the first target encoding vector after the search; otherwise, the fitness of the boundary transition value is judged.

[0030] Determine whether the fitness value of the boundary transition value is greater than the fitness value of the corresponding first target encoding vector. If so, use the boundary transition value as the first target encoding vector after the search; otherwise, use the original first target encoding vector directly as the first target encoding vector after the search.

[0031] Furthermore, for any second target encoding vector in the cooperative balance search population, a spiral balance search strategy is used to search for the second target encoding vector, resulting in the searched second target encoding vector, including:

[0032] For any second objective encoding vector in the cooperative equilibrium search population, obtain the fitness corresponding to the second objective encoding vector.

[0033] Based on the fitness corresponding to the second target encoding vector, the second target encoding vector is arranged in descending order to obtain the arranged second target encoding vector;

[0034] Based on the rearranged second target encoding vector, a search is performed on the m-th second target encoding vector to obtain the searched second target encoding vector as follows:

[0035]

[0036] in, Indicates the first t During the second optimization process, the first m The second target encoding vector after the permutation of the nth arrangement j dimensional hyperparameters, j =1,2,…,K, K Let m represent the total dimension of the hyperparameters, m = M / 2 + 1, M / 2 + 2, ..., M, where M represents the total number of second target encoding vectors, and M is an even number; Indicates the firstm The second target encoding vector after the first search. j dimensional hyperparameters, Represents a natural constant function. This represents the spiral search control factor, and ; This represents the coefficient of variation, and is set to 5. Represents pi (π). Represents the cosine function. This represents an exponential function with base e. This indicates the preset maximum number of optimization attempts. This represents a random spiral shape control factor uniformly distributed between [-1, 1], and abs represents the absolute value function. This represents the worst-performing encoded vector among all encoded vectors. j dimensional hyperparameters;

[0037] Based on the second target encoding vector after permutation, a search is performed on the g-th second target encoding vector to obtain the searched second target encoding vector as follows:

[0038]

[0039] in, The first digit represents the optimal coding vector among all coding vectors. j dimensional hyperparameters, g =1,2,…,M / 2 Indicates the first t During the second optimization process, the first g The second target encoding vector after the permutation of the nth arrangement j dimensional hyperparameters, Indicates the first g The second target encoding vector after the first search. j dimensional hyperparameters, This represents the directional search factor with a random dimension of -1 or 1.

[0040] Furthermore, for any third target encoding vector in the fast global search population, an optimal attraction and worst vigilance search strategy is used to search for the third target encoding vector, resulting in the searched third target encoding vector, including:

[0041] Generate a fifth random number between (0,1). ;

[0042] Determine the fifth random number Is it less than the preset second decision factor? If so, then the third target encoding vector is searched according to the optimal parameter encoding to obtain the searched third target encoding vector; otherwise, the third target encoding vector is searched according to the worst parameter encoding to obtain the searched third target encoding vector; where the second decision factor Set to t / , Indicates the maximum number of optimization attempts;

[0043] Based on the optimal parameter encoding, the third target encoding vector is searched, and the third target encoding vector after the search is obtained as follows:

[0044]

[0045] in, Indicates the first t During the second optimization process, the first k The third target encoding vector of the first j dimensional hyperparameters, k =1,2,…,P, where P represents the total number of third target encoding vectors. j =1,2,…,K, K This represents the total dimension of the hyperparameters. Indicates the first k The third target encoding vector after the first search j dimensional hyperparameters, The first digit represents the optimal coding vector among all coding vectors. j dimensional hyperparameters, This represents the sixth random number between (0,1). This represents the seventh random number between (0,1). Represents the sine function. Indicates the weighting factor for the first position. Indicates the weighting factor for the second position;

[0046] Based on the worst parameter encoding, the third target encoding vector is searched, and the resulting third target encoding vector is:

[0047]

[0048] in, This represents the eighth random number between (0,1). This represents the worst-performing encoded vector among all encoded vectors. j dimensional hyperparameters, Indicates the first t During the second optimization process, the first k The fitness corresponding to each third target encoding vector The fitness of the worst-case encoding vector is represented. This represents a constant term, and is set to 0.001.

[0049] Furthermore, based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search, the optimized green building energy consumption prediction model is determined, including:

[0050] Based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search, the optimal encoding vector is obtained again.

[0051] The hyperparameters contained in the re-acquired optimal encoding vector are used as the final hyperparameters of the green building energy consumption prediction model, thus obtaining the optimized green building energy consumption prediction model.

[0052] Furthermore, real-time building energy consumption characteristics corresponding to the target green building are collected, and the optimized green building energy consumption prediction model is used to analyze the real-time building energy consumption characteristics to determine the predicted energy consumption data. Energy-saving management of the green building is then performed based on the predicted energy consumption data, including:

[0053] Collect real-time building energy consumption characteristics corresponding to the target green building, and schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics and determine the predicted energy consumption data;

[0054] Based on multiple preset energy consumption data intervals, the energy consumption data interval to which the predicted energy consumption data belongs is determined, thus obtaining the target energy consumption data interval;

[0055] The system schedules the target energy consumption data range corresponding to the target energy-saving management scheme and executes the target energy-saving management scheme to realize energy-saving management of green buildings; wherein, each energy consumption data range is preset with an energy-saving management scheme.

[0056] Secondly, the present invention provides a green building energy-saving design system based on deep learning, comprising: a historical data acquisition module, a deep learning module, and a building energy-saving management module;

[0057] The historical data acquisition module is used to collect the historical building energy consumption characteristics and the historical energy consumption data corresponding to the target green building, and to preprocess the historical building energy consumption characteristics and the historical energy consumption data to obtain historical sample data.

[0058] The deep learning module is used to construct a green building energy consumption prediction model using a deep learning model, and uses the historical sample data as data support to optimize the green building energy consumption prediction model using a multi-strategy collaborative optimization algorithm to obtain the optimized green building energy consumption prediction model.

[0059] The building energy conservation management module is used to collect the real-time building energy consumption characteristics corresponding to the target green building, and to schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics, determine the predicted energy consumption data, and carry out energy conservation management of the green building based on the predicted energy consumption data.

[0060] This invention provides a deep learning-based green building energy-saving design method and system. It constructs a green building energy consumption prediction model using a deep learning model, and uses historical sample data as support. A multi-strategy collaborative optimization algorithm is employed to optimize the green building energy consumption prediction model, resulting in an optimized model. This optimized model can then be used to analyze real-time building energy consumption characteristics, enabling energy consumption prediction. Finally, energy-saving management can be implemented based on the predicted energy consumption data. This not only effectively improves energy-saving management efficiency and enables real-time energy-saving management of green buildings, but also facilitates the design of energy-saving management systems for staff. Attached Figure Description

[0061] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0062] Figure 1 A flowchart illustrating a deep learning-based green building energy-saving design method provided in an embodiment of the present invention.

[0063] Figure 2 This is a schematic diagram of a deep learning-based green building energy-saving design system provided in an embodiment of the present invention.

[0064] Among them, 21-historical data acquisition module, 22-deep learning module, and 23-building energy conservation management module.

[0065] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0066] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0067] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0068] like Figure 1 As shown, this embodiment of the invention provides a deep learning-based green building energy-saving design method, including:

[0069] S11. Collect the historical building energy consumption characteristics and historical energy consumption data corresponding to the target green building, and after preprocessing the historical building energy consumption characteristics and historical energy consumption data corresponding to the historical building energy consumption characteristics, obtain historical sample data.

[0070] The energy consumption characteristics of historical buildings corresponding to the target green building refer to some energy-related features. We can collect the energy consumption characteristics of historical buildings and the historical energy consumption data corresponding to the energy consumption characteristics of historical buildings at the same time. After learning from the collected data, we can make predictions.

[0071] S12. A green building energy consumption prediction model is constructed using a deep learning model, and the historical sample data is used as data support. A multi-strategy collaborative optimization algorithm is used to optimize the green building energy consumption prediction model to obtain the optimized green building energy consumption prediction model.

[0072] A green building energy consumption prediction model can be constructed using a deep learning model with data prediction capabilities. After constructing the green building energy consumption prediction model, hyperparameter optimization is required to achieve data recognition. In existing technologies, gradient descent is conventionally used for hyperparameter optimization, resulting in poor optimization capabilities and consequently poor energy consumption prediction performance. Therefore, this invention employs a multi-strategy collaborative optimization algorithm to optimize the green building energy consumption prediction model, thereby improving the algorithm's optimization speed and global optimization capabilities.

[0073] S13. Collect the real-time building energy consumption characteristics corresponding to the target green building, and schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics, determine the predicted energy consumption data, and carry out energy-saving management of green buildings based on the predicted energy consumption data.

[0074] Energy management of green buildings based on the predicted energy consumption data can be achieved by adopting an energy management plan preset by staff.

[0075] In this embodiment of the invention, historical building energy consumption characteristics and historical energy consumption data corresponding to the target green building are collected. After preprocessing the historical building energy consumption characteristics and historical energy consumption data, historical sample data is obtained, including:

[0076] Collect the target green building's average daily temperature, maximum daily temperature, minimum daily temperature, date type, season type, and weather on any given day to obtain the corresponding historical building energy consumption characteristics;

[0077] For any green building, collect historical energy consumption data corresponding to the historical building's energy consumption characteristics.

[0078] After normalizing the energy consumption characteristics of the historical buildings, the normalized energy consumption characteristics of the historical buildings are obtained. The normalized energy consumption characteristics of the historical buildings are used as input samples, and the historical energy consumption data corresponding to the energy consumption characteristics of the historical buildings are used as expected output labels to obtain historical sample data.

[0079] It is worth noting that the energy consumption characteristics of historical buildings corresponding to green buildings mentioned above are only a preferred embodiment provided by the present invention, and other characteristics can also be used as the energy consumption characteristics of historical buildings corresponding to green buildings.

[0080] In this embodiment of the invention, a green building energy consumption prediction model is constructed using a deep learning model, including: constructing a green building energy consumption prediction model using a CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory Network) model.

[0081] This invention proposes a green building energy consumption prediction model based on a hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM). It combines the advantages of CNN in feature extraction and the ability of LSTM in time series prediction, which can effectively improve the energy consumption prediction capability.

[0082] In this embodiment of the invention, the historical sample data is used as data support, and a multi-strategy collaborative optimization algorithm is employed to optimize the green building energy consumption prediction model, resulting in an optimized green building energy consumption prediction model, including:

[0083] The hyperparameters of the green building energy consumption prediction model are initialized using a chaotic mapping initialization method or a random initialization method, and the hyperparameters are encoded to obtain encoding vectors. Multiple different encoding vectors are then obtained.

[0084] Using the input samples in the historical sample data as the actual input and the expected output label in the historical sample data as the expected output, the fitness corresponding to each encoding vector is obtained.

[0085] For example, the encoded vector can be applied to the green building energy consumption prediction model. Then, the input sample in the historical sample data is used as the actual input, and the expected output label in the historical sample data is used as the expected output. The error function value corresponding to the encoded vector is obtained. Then, the error function value is added to a preset constant (such as 0.001), and the reciprocal is taken to obtain the fitness corresponding to the encoded vector.

[0086] Based on the fitness of the encoded vectors in descending order, all encoded vectors are divided into three groups: the optimal region search population, the cooperative balance search population, and the fast global search population.

[0087] For any first target encoding vector in the optimal region search population, a boundary probability transition strategy is used to search the first target encoding vector to obtain the searched first target encoding vector;

[0088] For any second target encoding vector in the cooperative balance search population, a spiral balance search strategy is used to search for the second target encoding vector to obtain the searched second target encoding vector.

[0089] For any third target encoding vector in the fast global search population, the optimal attraction and worst vigilance search strategy is used to search for the third target encoding vector to obtain the searched third target encoding vector.

[0090] Determine whether the optimization termination condition is met (e.g., the current number of optimizations reaches the preset maximum number of optimizations). If so, determine the optimized green building energy consumption prediction model based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search. Otherwise, return to the step of determining the fitness corresponding to each encoding vector.

[0091] Optionally, after searching the first target encoding vector, the second target encoding vector, and the third target encoding vector, out-of-bounds processing can be performed on the first target encoding vector, the second target encoding vector, and the third target encoding vector to ensure that the algorithm is always effective.

[0092] In this embodiment of the invention, for any first target encoding vector in the optimal region search population, a boundary probability transition strategy is used to search the first target encoding vector to obtain the searched first target encoding vector, including:

[0093] For any first target encoding vector in the optimal region search population, based on the boundaries of each dimension's hyperparameters, the boundary transition value corresponding to the first target encoding vector is obtained as follows:

[0094]

[0095] in, Indicates the first t In the optimization process, the i-th first target encoding vector corresponds to the th... j dimensional hyperparameters, Indicates the first i The first target encoding vector corresponding to the first j Dimensional hyperparameters The corresponding boundary transition values ​​are i=1,2,…,L, where L represents the total number of the first target encoding vectors, and j=1,2,…,K, where K represents the total dimension of the hyperparameters. This represents the upper bound of the j-th dimension hyperparameter. This represents the lower bound of the j-th dimension hyperparameter. Represents the first random number between (0,1). This represents the second random number between (0,1). The first element of the optimal encoding vector is... j dimensional hyperparameters;

[0096] For any first target encoding vector in the optimal region search population, obtain the neighborhood search value of the first target encoding vector:

[0097]

[0098] in, This represents the neighborhood search value of the first target encoding vector. This represents the j-th dimension hyperparameter corresponding to a random other first target encoding vector. Represents pi (π). This indicates the preset maximum number of optimization iterations, and cos represents the cosine function. This represents a third random number between (0, 0.55);

[0099] The first decision factor is generated as follows: ;in, Indicates the first decision factor. This represents the control hyperparameter corresponding to the first decision factor, and is set to 0.499; This represents an exponential function with the natural constant e as its base.

[0100] Generate a fourth random number between (0,1) And determine the fourth random number Is it less than the first decision factor? If so, the neighborhood search value is used as the first target encoding vector after the search; otherwise, the fitness of the boundary transition value is judged.

[0101] Determine whether the fitness value of the boundary transition value is greater than the fitness value of the corresponding first target encoding vector. If so, use the boundary transition value as the first target encoding vector after the search; otherwise, use the original first target encoding vector directly as the first target encoding vector after the search.

[0102] This invention employs a boundary probability transition strategy to search for the first target encoding vector. In the early stages of the algorithm, this strategy guides the first target encoding vector to search towards the boundary. During the position update process, the change magnitude is adjusted based on the optimal encoding vector, thereby preventing the problem of missing the global optimal solution due to excessively fast convergence. In the later stages of the algorithm, it tends to perform neighborhood search, which can effectively merge the regions between two first target encoding vectors. Furthermore, the search accuracy increases as the algorithm progresses, effectively improving the algorithm's search accuracy and its ability to find the global optimal solution.

[0103] In this embodiment of the invention, for any second target encoding vector in the cooperative balance search population, a spiral balance search strategy is used to search for the second target encoding vector to obtain the searched second target encoding vector, including:

[0104] For any second objective encoding vector in the cooperative equilibrium search population, obtain the fitness corresponding to the second objective encoding vector.

[0105] Based on the fitness corresponding to the second target encoding vector, the second target encoding vector is arranged in descending order to obtain the arranged second target encoding vector;

[0106] Based on the rearranged second target encoding vector, a search is performed on the m-th second target encoding vector to obtain the searched second target encoding vector as follows:

[0107]

[0108] in, Indicates the first t During the second optimization process, the first m The second target encoding vector after the permutation of the nth arrangement j dimensional hyperparameters, j =1,2,…,K, K Let m represent the total dimension of the hyperparameters, m = M / 2 + 1, M / 2 + 2, ..., M, where M represents the total number of second target encoding vectors, and M is an even number; Indicates the first m The second target encoding vector after the first search. j dimensional hyperparameters, Represents a natural constant function. This represents the spiral search control factor, and ; This represents the coefficient of variation, and is set to 5. Represents pi (π). Represents the cosine function. This represents an exponential function with base e. This indicates the preset maximum number of optimization attempts. This represents a random spiral shape control factor uniformly distributed between [-1, 1], and abs represents the absolute value function. This represents the worst-performing encoded vector among all encoded vectors. j dimensional hyperparameters;

[0109] Based on the second target encoding vector after permutation, a search is performed on the g-th second target encoding vector to obtain the searched second target encoding vector as follows:

[0110]

[0111] in, The first digit represents the optimal coding vector among all coding vectors. j dimensional hyperparameters, g =1,2,…,M / 2 Indicates the first t During the second optimization process, the first g The second target encoding vector after the permutation of the nth arrangement j dimensional hyperparameters, Indicates the first g The second target encoding vector after the first search. j dimensional hyperparameters, This represents the directional search factor with a random dimension of -1 or 1.

[0112] The embodiments of the present invention employ a spiral balanced search strategy to search the second target encoding vector. This strategy allows for a wider search of the second target encoding with poor fitness, while providing a better search for the second target encoding with better fitness. Furthermore, it integrates cosine and exponential functions to balance global and local search capabilities during the search process.

[0113] In this embodiment of the invention, for any third target encoding vector in the fast global search population, an optimal attraction and worst vigilance search strategy is used to search for the third target encoding vector to obtain the searched third target encoding vector, including:

[0114] Generate a fifth random number between (0,1). ;

[0115] Determine the fifth random number Is it less than the preset second decision factor? If so, then the third target encoding vector is searched according to the optimal parameter encoding to obtain the searched third target encoding vector; otherwise, the third target encoding vector is searched according to the worst parameter encoding to obtain the searched third target encoding vector; where the second decision factor Set to t / , Indicates the maximum number of optimization attempts;

[0116] Based on the optimal parameter encoding, the third target encoding vector is searched, and the third target encoding vector after the search is obtained as follows:

[0117]

[0118] in, Indicates the first t During the second optimization process, the first k The third target encoding vector of the first j dimensional hyperparameters, k =1,2,…,P, where P represents the total number of third target encoding vectors. j =1,2,…,K, K This represents the total dimension of the hyperparameters. Indicates the first k The third target encoding vector after the first search j dimensional hyperparameters, The first digit represents the optimal coding vector among all coding vectors. j dimensional hyperparameters, This represents the sixth random number between (0,1). This represents the seventh random number between (0,1). Represents the sine function. Indicates the weighting factor for the first position. Indicates the weighting factor for the second position;

[0119] Based on the worst parameter encoding, the third target encoding vector is searched, and the resulting third target encoding vector is:

[0120]

[0121] in, This represents the eighth random number between (0,1). This represents the worst-performing encoded vector among all encoded vectors. j dimensional hyperparameters, Indicates the first t During the second optimization process, the first k The fitness corresponding to each third target encoding vector The fitness of the worst-case encoding vector is represented. This represents a constant term, and is set to 0.001.

[0122] The embodiments of the present invention employ an optimal attraction and worst-case warning search strategy to search for the third target encoding vector. In the early stages of the algorithm, the search can be conducted in a direction away from the worst position, which can effectively avoid the algorithm getting stuck in local optima. At the same time, in the later stages of the algorithm, the search can be more inclined to be conducted around the optimal region, which improves the search accuracy. Furthermore, the fluctuating search is more conducive to finding the global optimal position.

[0123] The various search methods provided in this invention can work together to search more solution space regions, avoid search collisions, effectively avoid getting trapped in local optima, improve search speed and accuracy, and ultimately improve the accuracy of building energy consumption prediction.

[0124] In this embodiment of the invention, an optimized green building energy consumption prediction model is determined based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search, including:

[0125] Based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search, the optimal encoding vector is obtained again.

[0126] The hyperparameters contained in the re-acquired optimal encoding vector are used as the final hyperparameters of the green building energy consumption prediction model, thus obtaining the optimized green building energy consumption prediction model.

[0127] In this embodiment of the invention, real-time building energy consumption characteristics corresponding to the target green building are collected, and the optimized green building energy consumption prediction model is scheduled to analyze the real-time building energy consumption characteristics to determine the predicted energy consumption data. Energy-saving management of the green building is then performed based on the predicted energy consumption data, including:

[0128] Collect real-time building energy consumption characteristics corresponding to the target green building, and schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics and determine the predicted energy consumption data;

[0129] Based on multiple preset energy consumption data intervals, the energy consumption data interval to which the predicted energy consumption data belongs is determined, thus obtaining the target energy consumption data interval;

[0130] The system schedules the target energy consumption data range corresponding to the target energy-saving management scheme and executes the target energy-saving management scheme to realize energy-saving management of green buildings; wherein, each energy consumption data range is preset with an energy-saving management scheme.

[0131] For example, a threshold can be set. The interval above the threshold, considered high energy consumption, is designated as the first energy consumption data interval, and the interval below the threshold, considered high energy consumption, is designated as the second energy consumption data interval. When the predicted energy consumption data of a green building falls within the second energy consumption data interval, no management is required. When the predicted energy consumption data of a green building falls within the first energy consumption data interval, a preset energy-saving management scheme for that interval can be executed. The target energy-saving management scheme can be set as follows: turning off lower-priority electrical equipment, limiting air conditioning use, limiting the number of lights used, etc.

[0132] It is worth noting that the above is merely an example of an embodiment of the present invention. More energy consumption data ranges and corresponding energy-saving management schemes can be set to achieve more efficient green building energy-saving management.

[0133] This invention provides a deep learning-based energy-saving design method for green buildings. It constructs a green building energy consumption prediction model using a deep learning model, supports it with historical sample data, and optimizes the model using a multi-strategy collaborative optimization algorithm. This optimized model can then be used to analyze real-time building energy consumption characteristics, enabling energy consumption prediction. Ultimately, energy-saving management can be implemented based on the predicted energy consumption data. This not only effectively improves energy-saving management efficiency and enables real-time energy-saving management of green buildings, but also facilitates the design of energy-saving management systems for staff.

[0134] like Figure 2 As shown, the present invention provides a green building energy-saving design system based on deep learning, including: a historical data acquisition module 21, a deep learning module 22, and a building energy-saving management module 23;

[0135] The historical data acquisition module 21 is used to collect the historical building energy consumption characteristics and the historical energy consumption data corresponding to the target green building, and to preprocess the historical building energy consumption characteristics and the historical energy consumption data to obtain historical sample data.

[0136] The deep learning module 22 is used to construct a green building energy consumption prediction model using a deep learning model, and to optimize the green building energy consumption prediction model using a multi-strategy collaborative optimization algorithm with the historical sample data as data support, so as to obtain an optimized green building energy consumption prediction model.

[0137] The building energy conservation management module 23 is used to collect the real-time building energy consumption characteristics corresponding to the target green building, and to schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics, determine the predicted energy consumption data, and carry out energy conservation management of the green building based on the predicted energy consumption data.

[0138] This invention provides a deep learning-based green building energy-saving design system that can execute the above-described method and technical solution. Its principle and beneficial effects are similar and will not be repeated here.

[0139] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that the invention is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A deep learning-based green building energy-saving design method, characterized in that, include: Collect the historical building energy consumption characteristics and historical energy consumption data corresponding to the target green building, and after preprocessing the historical building energy consumption characteristics and historical energy consumption data, obtain historical sample data. A green building energy consumption prediction model is constructed using a deep learning model, and the historical sample data is used as data support. A multi-strategy collaborative optimization algorithm is used to optimize the green building energy consumption prediction model to obtain the optimized green building energy consumption prediction model. Collect real-time building energy consumption characteristics corresponding to the target green building, and schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics, determine the predicted energy consumption data, and carry out energy-saving management of green buildings based on the predicted energy consumption data. Using the historical sample data as support, a multi-strategy collaborative optimization algorithm is employed to optimize the green building energy consumption prediction model, resulting in an optimized green building energy consumption prediction model, including: The hyperparameters of the green building energy consumption prediction model are initialized using a chaotic mapping initialization method or a random initialization method, and the hyperparameters are encoded to obtain encoding vectors. Multiple different encoding vectors are then obtained. Using the input samples in the historical sample data as the actual input and the expected output label in the historical sample data as the expected output, the fitness corresponding to each encoding vector is obtained. Based on the fitness of the encoded vectors in descending order, all encoded vectors are divided into three groups: the optimal region search population, the cooperative balance search population, and the fast global search population. For any first target encoding vector in the optimal region search population, a boundary probability transition strategy is used to search the first target encoding vector to obtain the searched first target encoding vector; For any second target encoding vector in the cooperative balance search population, a spiral balance search strategy is used to search for the second target encoding vector to obtain the searched second target encoding vector. For any third target encoding vector in the fast global search population, the optimal attraction and worst vigilance search strategy is used to search for the third target encoding vector to obtain the searched third target encoding vector. Determine whether the optimization termination condition is met. If so, determine the optimized green building energy consumption prediction model based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search. Otherwise, return to the step of determining the fitness corresponding to each encoding vector.

2. The deep learning-based green building energy-saving design method according to claim 1, characterized in that, Collect historical building energy consumption characteristics and corresponding historical energy consumption data for the target green building. After preprocessing the historical building energy consumption characteristics and corresponding historical energy consumption data, historical sample data is obtained, including: Collect the target green building's average daily temperature, maximum daily temperature, minimum daily temperature, date type, season type, and weather on any given day to obtain the corresponding historical building energy consumption characteristics; For any green building, collect historical energy consumption data corresponding to the historical building's energy consumption characteristics. After normalizing the energy consumption characteristics of the historical buildings, the normalized energy consumption characteristics of the historical buildings are obtained. The normalized energy consumption characteristics of the historical buildings are used as input samples, and the historical energy consumption data corresponding to the energy consumption characteristics of the historical buildings are used as expected output labels to obtain historical sample data.

3. The deep learning-based green building energy-saving design method according to claim 1, characterized in that, A green building energy consumption prediction model is constructed using a deep learning model, including: constructing a green building energy consumption prediction model using a CNN-LSTM model.

4. The deep learning-based green building energy-saving design method according to claim 1, characterized in that, For any first target encoding vector in the optimal region search population, a boundary probability transition strategy is used to search the first target encoding vector, resulting in the searched first target encoding vector, including: For any first target encoding vector in the optimal region search population, based on the boundaries of each dimension's hyperparameters, the boundary transition value corresponding to the first target encoding vector is obtained as follows: in, Indicates the first t In the optimization process, the i-th first target encoding vector corresponds to the th... j dimensional hyperparameters, Indicates the first i The first target encoding vector corresponding to the first j Dimensional hyperparameters The corresponding boundary transition values ​​are i=1,2,…,L, where L represents the total number of the first target encoding vectors, and j=1,2,…,K, where K represents the total dimension of the hyperparameters. This represents the upper bound of the j-th dimension hyperparameter. This represents the lower bound of the j-th dimension hyperparameter. Represents the first random number between (0,1). This represents the second random number between (0,1). The first element of the optimal encoding vector is... j dimensional hyperparameters; For any first target encoding vector in the optimal region search population, obtain the neighborhood search value of the first target encoding vector: in, This represents the neighborhood search value of the first target encoding vector. This represents the j-th dimension hyperparameter corresponding to a random other first target encoding vector. Represents pi (π). This indicates the preset maximum number of optimization iterations, and cos represents the cosine function. This represents a third random number between (0, 0.55); The first decision factor is generated as follows: ;in, Indicates the first decision factor. This represents the control hyperparameter corresponding to the first decision factor, and is set to 0.499; This represents an exponential function with the natural constant e as its base. Generate a fourth random number between (0,1) And determine the fourth random number Is it less than the first decision factor? If so, the neighborhood search value is used as the first target encoding vector after the search; otherwise, the fitness of the boundary transition value is judged. Determine whether the fitness value of the boundary transition value is greater than the fitness value of the corresponding first target encoding vector. If so, use the boundary transition value as the first target encoding vector after the search; otherwise, use the original first target encoding vector directly as the first target encoding vector after the search.

5. The deep learning-based green building energy-saving design method according to claim 4, characterized in that, For any second target encoding vector in the cooperative equilibrium search population, a spiral equilibrium search strategy is used to search for the second target encoding vector, resulting in the searched second target encoding vector, including: For any second objective encoding vector in the cooperative equilibrium search population, obtain the fitness corresponding to the second objective encoding vector. Based on the fitness corresponding to the second target encoding vector, the second target encoding vector is arranged in descending order to obtain the arranged second target encoding vector; Based on the rearranged second target encoding vector, a search is performed on the m-th second target encoding vector to obtain the searched second target encoding vector as follows: in, Indicates the first t During the second optimization process, the first m The second target encoding vector after the permutation of the nth arrangement j dimensional hyperparameters, j =1,2,…,K, K Let m represent the total dimension of the hyperparameters, m = M / 2 + 1, M / 2 + 2, ..., M, where M represents the total number of second target encoding vectors, and M is an even number; Indicates the first m The second target encoding vector after the first search. j dimensional hyperparameters, Represents a natural constant function. This represents the spiral search control factor, and ; This represents the coefficient of variation, and is set to 5. Represents pi (π). Represents the cosine function. This represents an exponential function with base e. This indicates the preset maximum number of optimization attempts. This represents a random spiral shape control factor uniformly distributed between [-1, 1], and abs represents the absolute value function. This represents the worst-performing encoded vector among all encoded vectors. j dimensional hyperparameters; Based on the second target encoding vector after permutation, a search is performed on the g-th second target encoding vector to obtain the searched second target encoding vector as follows: in, The first digit represents the optimal coding vector among all coding vectors. j dimensional hyperparameters, g =1,2,…,M / 2 Indicates the first t During the second optimization process, the first g The second target encoding vector after the permutation of the nth arrangement j dimensional hyperparameters, Indicates the first g The second target encoding vector after the first search. j dimensional hyperparameters, This represents the directional search factor with a random dimension of -1 or 1.

6. The deep learning-based green building energy-saving design method according to claim 5, characterized in that, For any third target encoding vector in the fast global search population, an optimal attraction and worst vigilance search strategy is used to search for the third target encoding vector, resulting in the searched third target encoding vector, including: Generate a fifth random number between (0,1). ; Determine the fifth random number Is it less than the preset second decision factor? If so, then the third target encoding vector is searched according to the optimal parameter encoding to obtain the searched third target encoding vector; otherwise, the third target encoding vector is searched according to the worst parameter encoding to obtain the searched third target encoding vector; where the second decision factor Set to t / , Indicates the maximum number of optimization attempts; Based on the optimal parameter encoding, the third target encoding vector is searched, and the third target encoding vector after the search is obtained as follows: in, Indicates the first t During the second optimization process, the first k The third target encoding vector of the first j dimensional hyperparameters, k =1,2,…,P, where P represents the total number of third target encoding vectors. j =1,2,…,K, K This represents the total dimension of the hyperparameters. Indicates the first k The third target encoding vector after the first search j dimensional hyperparameters, The first digit represents the optimal coding vector among all coding vectors. j dimensional hyperparameters, This represents the sixth random number between (0,1). This represents the seventh random number between (0,1). Represents the sine function. Indicates the weighting factor for the first position. Indicates the weighting factor for the second position; Based on the worst parameter encoding, the third target encoding vector is searched, and the resulting third target encoding vector is: in, This represents the eighth random number between (0,1). This represents the worst-performing encoded vector among all encoded vectors. j dimensional hyperparameters, Indicates the first t During the second optimization process, the first k The fitness corresponding to each third target encoding vector The fitness of the worst-case encoding vector is represented. This represents a constant term, and is set to 0.

001.

7. The deep learning-based green building energy-saving design method according to claim 6, characterized in that, Based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search, the optimized green building energy consumption prediction model is determined, including: Based on the first target encoding vector, the second target encoding vector, and the third target encoding vector after the search, the optimal encoding vector is obtained again. The hyperparameters contained in the re-acquired optimal encoding vector are used as the final hyperparameters of the green building energy consumption prediction model, thus obtaining the optimized green building energy consumption prediction model.

8. The deep learning-based green building energy-saving design method according to claim 1, characterized in that, The process involves collecting real-time building energy consumption characteristics corresponding to the target green building, scheduling the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics, determining predicted energy consumption data, and performing energy-saving management of the green building based on the predicted energy consumption data, including: Collect real-time building energy consumption characteristics corresponding to the target green building, and schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics and determine the predicted energy consumption data; Based on multiple preset energy consumption data intervals, the energy consumption data interval to which the predicted energy consumption data belongs is determined, thus obtaining the target energy consumption data interval; The system schedules the target energy consumption data range corresponding to the target energy-saving management scheme and executes the target energy-saving management scheme to realize energy-saving management of green buildings; wherein, each energy consumption data range is preset with an energy-saving management scheme.

9. A deep learning-based green building energy-saving design system, wherein the deep learning-based green building energy-saving design system is capable of executing the deep learning-based green building energy-saving design method according to any one of claims 1-8, characterized in that, include: Historical data acquisition module, deep learning module, and building energy conservation management module; The historical data acquisition module is used to collect the historical building energy consumption characteristics and the historical energy consumption data corresponding to the target green building, and to preprocess the historical building energy consumption characteristics and the historical energy consumption data to obtain historical sample data. The deep learning module is used to construct a green building energy consumption prediction model using a deep learning model, and uses the historical sample data as data support to optimize the green building energy consumption prediction model using a multi-strategy collaborative optimization algorithm to obtain the optimized green building energy consumption prediction model. The building energy conservation management module is used to collect the real-time building energy consumption characteristics corresponding to the target green building, and to schedule the optimized green building energy consumption prediction model to analyze the real-time building energy consumption characteristics, determine the predicted energy consumption data, and carry out energy conservation management of the green building based on the predicted energy consumption data.

Images