Comprehensive energy management method and system

Abstract

The invention relates to the technical field of monitoring control, and discloses a comprehensive energy management method and system, and the method comprises the steps: obtaining historical energy data, equipment operation parameters and internal and external environment data of a target smart entity, the internal and external environment data comprising building characteristic data, season data, building internal temperature data and building internal humidity data; inputting the historical energy data, the equipment operation parameters and the internal and external environment data into a trained first adjustment model, and outputting a first energy demand result; inputting the building characteristic data and the first energy demand result into a trained second adjustment model, and outputting a second energy demand result; and formulating an energy adjustment strategy based on the second energy demand result. By implementing the method and the system, response and adjustment can be quickly made when environment and energy demand changes, challenges brought by changes of operating environment and building characteristics are effectively handled, and efficient energy utilization and sustainable development of the smart entity are realized.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and control technology, and in particular to a comprehensive energy management method and system. Background Art

[0002] Traditional energy management methods mainly rely on manual inspections and traditional monitoring equipment, which can collect key operating parameters such as temperature, humidity, power consumption, etc., and display or record them through instruments for operators to observe and analyze. Operators manually adjust the operating status of energy equipment based on these data and experience. With the development and application of intelligent technology, smart entities in different industries, such as hospitals, oil parks, communication plants, thermal power plants, etc., are actively exploring and applying advanced energy management technologies.

[0003] Smart entity energy management has become a new development trend. Smart entities use advanced energy management systems to monitor the energy consumption of various areas of the hospital in real time and achieve remote control and automatic adjustment. For example, the smart hospital system collects a large amount of real-time operating data, including energy consumption, temperature, humidity and other parameters of medical equipment, and achieves accurate prediction and optimization of energy demand through data analysis and modeling.

[0004] However, in actual applications, if the operating environment and building characteristics of the smart entity change, such as the adjustment of the hospital department layout, the update of medical equipment or the change of climate, the new data distribution may be significantly different from the training data. In this case, the accuracy of the traditional prediction model may drop significantly, thus affecting the effect of energy management. Summary of the invention

[0005] In view of this, the present invention provides a comprehensive energy management method and system to solve the problem that the prediction accuracy of the energy monitoring and adjustment model is greatly reduced due to changes in the operating environment and building characteristics.

[0006] In a first aspect, the present invention provides a comprehensive energy management method, which includes: obtaining historical energy data, equipment operating parameters and internal and external environmental data of a target smart entity, the internal and external environmental data including building characteristic data, seasonal data, building internal temperature data and building internal humidity data; inputting the historical energy data, equipment operating parameters and internal and external environmental data into a trained first adjustment model, and outputting a first energy demand result, the first adjustment model being a hybrid model based on a self-attention mechanism and a gated recurrent unit; inputting the building characteristic data and the first energy demand result into a trained second adjustment model, and outputting a second energy demand result, the second adjustment model being a quantum support vector machine model; formulating an energy adjustment strategy based on the second energy demand result, and adjusting the energy of the target smart entity based on the energy adjustment strategy.

[0007] The comprehensive energy management method provided in this embodiment realizes efficient management and precise regulation of smart energy through comprehensive and real-time data collection and multi-level energy demand forecasting. First, through equipment operation parameter monitoring and internal and external environmental data monitoring, the operation parameters of each device in the smart entity and the internal and external environmental data can be obtained in real time. These data cover key operation parameters such as equipment operation efficiency data, equipment power data, equipment utilization data and equipment heat dissipation efficiency data, as well as important environmental factors such as building characteristics, seasons, temperature, humidity, etc. This all-round data collection mechanism not only ensures the real-time and accuracy of the data, but also provides reliable data support for subsequent energy demand forecasting and strategy formulation. Secondly, by adopting a two-layer prediction, using historical energy data, real-time equipment operation parameters and internal and external environmental data, combined with a hybrid model based on self-attention mechanism and gated recurrent unit, preliminary energy demand forecasting is performed. This model combines the sensitivity of the self-attention mechanism to time series data and the processing ability of the gated recurrent unit to sequence data, so as to more accurately capture the complex relationship between data and improve the accuracy of the prediction. Afterwards, based on the first-layer prediction output, the impact of building characteristic data on energy demand is further considered. By introducing the quantum support vector machine model, the first-layer prediction results can be corrected and improved to generate more accurate second-layer energy demand prediction results. The powerful computing power and high-dimensional data processing capabilities of the quantum support vector machine model can still maintain the stability and accuracy of the prediction when facing a complex and changeable data environment. Finally, according to the second-layer energy demand prediction results, an adjustment strategy that meets the current energy demand is intelligently generated, which not only takes into account the real-time energy demand, but also takes into account the architectural characteristics and environmental changes of the smart entity to ensure the effectiveness and pertinence of the adjustment measures. Finally, in the energy adjustment execution stage, the method adjusts the energy of the smart entity in real time according to the strategy. By accurately controlling the operating parameters of the equipment, optimizing the energy allocation plan, etc., the efficient utilization of energy and the goal of energy conservation and emission reduction are achieved. At the same time, the method also has a real-time feedback mechanism, which can continuously adjust and optimize the adjustment strategy according to the actual effect to ensure the continuity and adaptability of energy management. In summary, the comprehensive energy management method provided in this embodiment effectively solves many drawbacks in traditional energy management methods through real-time data collection, accurate energy demand prediction and intelligent energy adjustment strategy. Especially when facing changes in the smart entity environment and energy demand, this method can respond and adjust quickly to ensure the flexibility and efficiency of energy management, providing strong guarantees for the sustainable development of smart entities.

[0008] In an optional implementation, when the target smart entity is a hospital, the historical energy data includes medical gas usage data, electricity consumption data, refrigeration and heating energy consumption data, and energy consumption distribution data of each department, the equipment operation parameters include medical equipment operation efficiency data, medical equipment power data, medical equipment utilization data, and medical equipment heat dissipation efficiency data, and the internal and external environmental data include building characteristic data, seasonal data, building internal temperature data, and building internal humidity data. Before inputting the historical energy data, equipment operation parameters, and internal and external environmental data into the trained first adjustment model, the method further includes: based on the historical energy data, equipment operation parameters, and The internal and external environmental data obtain a parameter embedding vector; the parameter embedding vector is linearly transformed to obtain a first embedding vector, a second embedding vector and a third embedding vector, respectively, the first embedding vector is a search vector, the second embedding vector is an identification vector, and the third embedding vector is an information vector; the scaled dot product attention mechanism is used to calculate the attention weight based on the first embedding vector and the second embedding vector; the third embedding vector is weightedly summed based on the attention weight to obtain the self-attention layer output value; based on the reset gate, update gate and self-attention layer output value of the first adjustment model, the target hidden state is determined, and the self-attention layer output value of the target hidden state is output as the first energy demand result.

[0009] In an optional embodiment, a parameter embedding vector is obtained based on historical energy data, equipment operating parameters and internal and external environmental data, including: normalizing medical gas usage data, electricity consumption data, refrigeration and heating energy consumption data, energy consumption distribution data of each department, operating efficiency data of medical equipment, power data of medical equipment, utilization data of medical equipment and heat dissipation efficiency data of medical equipment, extracting features and obtaining a first feature set; performing category feature encoding conversion on building characteristic data and seasonal data to obtain a second feature set; normalizing building internal temperature data and building internal humidity data to extract features and obtain a third feature set; obtaining a first feature weight, a second feature weight and a third feature weight, the first feature weight, the second feature weight and the third feature weight respectively corresponding to the first feature set, the second feature set and the third feature set, and the first feature weight, the second feature weight and the third feature weight are obtained by training a first adjustment model; using an attention mechanism to fuse the first feature set, the second feature set and the third feature set based on the first feature weight, the second feature weight and the third feature weight to obtain a parameter embedding vector.

[0010] In an optional embodiment, a parameter embedding vector is linearly transformed to obtain a first embedding vector, a second embedding vector and a third embedding vector, respectively, including: obtaining a first weight matrix, a first bias item, a second weight matrix, a second bias item, a third weight matrix and a third bias item, the first weight matrix and the first bias item correspond to the first embedding vector, the second weight matrix and the second bias item correspond to the second embedding vector, and the third weight matrix and the third bias item correspond to the third embedding vector; linearly transforming the parameter embedding vector using the first weight matrix and the first bias item to obtain the first embedding vector; linearly transforming the parameter embedding vector using the second weight matrix and the second bias item to obtain the second embedding vector; linearly transforming the parameter embedding vector using the third weight matrix and the third bias item to obtain the third embedding vector.

[0011] In an optional embodiment, an attention weight is calculated based on the first embedding vector and the second embedding vector using a scaled dot product attention mechanism, including determining an inner product of a target vector based on the first embedding vector and the second embedding vector; determining a target scaling factor based on the second embedding vector; and determining an attention weight based on the target vector inner product and the target scaling factor using a normalized exponential function.

[0012] In an optional embodiment, the building characteristic data includes building energy consumption data, building information model data and thermal conduction efficiency data, and the building characteristic data and the first energy demand result are input into a trained second adjustment model to output a second energy demand result, including: extracting features from the building energy consumption data and the thermal conduction efficiency data to obtain a first correction feature; encoding and converting the building information model data to obtain a second correction feature; extracting features from the first energy demand result to obtain a third correction feature; generating a fused correction feature based on the first correction feature, the second correction feature and the third correction feature; performing quantum feature mapping on the fused correction feature using quantum bits and quantum gate operations to obtain a quantum state fused correction feature; and inputting the quantum state fused correction feature into a trained quantum support vector machine model to obtain a second energy demand result.

[0013] In an optional implementation, the quantum state fusion correction feature is input into a trained quantum support vector machine model to obtain a second energy demand result, including: inputting the quantum state fusion correction feature into a quantum register of a quantum computing platform, using the quantum computing platform to perform a predefined quantum gate operation on the quantum state fusion correction feature to obtain a measurement result; and decoding the measurement result to obtain a second energy demand result.

[0014] In a second aspect, the present invention provides an integrated energy management system, which includes: an acquisition module for acquiring historical energy data, equipment operating parameters and internal and external environmental data of a target smart entity, the internal and external environmental data including building characteristic data, seasonal data, building internal temperature data and building internal humidity data; a first output module for inputting historical energy data, equipment operating parameters and internal and external environmental data into a trained first adjustment model, and outputting a first energy demand result, the first adjustment model being a hybrid model based on a self-attention mechanism and a gated recurrent unit; a second output module for inputting building characteristic data and the first energy demand result into a trained second adjustment model, and outputting a second energy demand result, the second adjustment model being a quantum support vector machine model; an adjustment module for formulating an energy adjustment strategy based on the second energy demand result, and adjusting the energy of the target smart entity based on the energy adjustment strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 is a flow chart of a comprehensive energy management method according to an embodiment of the present invention; Figure 2 is a structural block diagram of an integrated energy management system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0018] Traditional energy management methods mainly rely on manual inspections and traditional monitoring equipment, which can collect key operating parameters such as temperature, humidity, power consumption, etc., and display or record them through instruments for operators to observe and analyze. Operators manually adjust the operating status of energy equipment based on these data and experience. With the development and application of intelligent technology, smart entities in different industries are actively exploring and applying advanced energy management technologies, such as hospitals, petroleum, communications, thermal power plants and other smart entities.

[0019] Smart entity energy management has become a new development trend. Smart entities use advanced energy management systems to monitor the energy consumption of various areas of the hospital in real time and achieve remote control and automatic adjustment. For example, the smart hospital system collects a large amount of real-time operating data, including energy consumption, temperature, humidity and other parameters of medical equipment, and achieves accurate prediction and optimization of energy demand through data analysis and modeling.

[0020] However, in actual applications, if the operating environment and building characteristics of the smart entity change, such as the adjustment of the hospital department layout, the update of medical equipment or the change of climate, the new data distribution may be significantly different from the training data. In this case, the accuracy of the traditional prediction model may drop significantly, thus affecting the effect of energy management.

[0021] Smart entities such as hospitals, oil, communications, and thermal power plants contain multiple buildings with different functions. Taking hospitals as an example, there are multiple buildings related to medical services, diagnostic equipment, and auxiliary facilities. Among them, the main building of the hospital covers core functional areas such as operating rooms, treatment rooms, and wards. The energy consumption data of the main building continuously records the various energy uses of the hospital during operation, such as electricity consumption and air conditioning energy consumption. These data are important indicators reflecting the energy efficiency and operating costs of the hospital, and are crucial to the energy management and energy-saving optimization of the hospital. The building information model data of the main building provides a three-dimensional layout of the hospital space, including the location of medical equipment, the layout of ventilation and air-conditioning systems, etc., which provides strong support for the analysis of energy efficiency and environmental comfort. At the same time, the thermal conductivity efficiency data of the hospital's walls, roofs and other building structures are also key parameters for evaluating the thermal insulation performance of the hospital.

[0022] For the hospital's cooling and heating facilities, including central air-conditioning systems, boiler rooms, etc., their energy consumption data, such as water pump energy consumption and cooling energy consumption, are also important references for measuring the energy efficiency of these facilities. Building information model data reveals the internal structure and connection methods of cooling and heating facilities, providing convenience for analyzing the relationship between cooling and heating effects and energy consumption. In addition, heat transfer efficiency data also covers equipment such as heat exchangers and radiators, which plays an important role in the hospital's energy management and energy conservation and emission reduction.

[0023] By monitoring these energy consumption data in real time, hospitals can quickly identify abnormal energy consumption and possible equipment failures, and take timely measures to prevent energy waste and equipment damage. Combined with building information model data, hospitals can deeply analyze their energy usage, discover energy-saving potential and improvement points, and implement targeted energy-saving measures. Heat transfer efficiency data helps to evaluate the operating status of equipment and the need for preventive maintenance, ensuring the efficient and stable operation of medical equipment.

[0024] In summary, building energy consumption data, building information model data and heat transfer efficiency data play a key role in the energy management and optimization of smart entities, providing comprehensive and accurate information support for hospital managers, and are important tools for promoting green and sustainable development and energy conservation and emission reduction in hospitals. However, in previous technical practices, these data are often ignored, which affects the accuracy and efficiency of energy management models.

[0025] The comprehensive energy management method provided in this embodiment realizes efficient management and precise regulation of smart energy through comprehensive and real-time data collection and multi-level energy demand forecasting. First, through equipment operation parameter monitoring and internal and external environmental data monitoring, the operation parameters of each device in the smart entity and the internal and external environmental data can be obtained in real time. These data cover key operation parameters such as equipment operation efficiency data, equipment power data, equipment utilization data and equipment heat dissipation efficiency data, as well as important environmental factors such as building characteristics, seasons, temperature, humidity, etc. This all-round data collection mechanism not only ensures the real-time and accuracy of the data, but also provides reliable data support for subsequent energy demand forecasting and strategy formulation. Secondly, by adopting a two-layer prediction, using historical energy data, real-time equipment operation parameters and internal and external environmental data, combined with a hybrid model based on self-attention mechanism and gated recurrent unit, preliminary energy demand forecasting is performed. This model combines the sensitivity of the self-attention mechanism to time series data and the processing ability of the gated recurrent unit to sequence data, so as to more accurately capture the complex relationship between data and improve the accuracy of the prediction. Afterwards, based on the first-layer prediction output, the impact of building characteristic data on energy demand is further considered. By introducing the quantum support vector machine model, the first-layer prediction results can be corrected and improved to generate more accurate second-layer energy demand prediction results. The powerful computing power and high-dimensional data processing capabilities of the quantum support vector machine model can still maintain the stability and accuracy of the prediction when facing a complex and changeable data environment. Finally, according to the second-layer energy demand prediction results, an adjustment strategy that meets the current energy demand is intelligently generated, which not only takes into account the real-time energy demand, but also takes into account the architectural characteristics and environmental changes of the smart entity to ensure the effectiveness and pertinence of the adjustment measures. Finally, in the energy adjustment execution stage, the method adjusts the energy of the smart entity in real time according to the strategy. By accurately controlling the operating parameters of the equipment, optimizing the energy allocation plan, etc., the efficient utilization of energy and the goal of energy conservation and emission reduction are achieved. At the same time, the method also has a real-time feedback mechanism, which can continuously adjust and optimize the adjustment strategy according to the actual effect to ensure the continuity and adaptability of energy management. In summary, the comprehensive energy management method provided in this embodiment effectively solves many drawbacks in traditional energy management methods through real-time data collection, accurate energy demand prediction and intelligent energy adjustment strategy. Especially when facing changes in the smart entity environment and energy demand, this method can respond and adjust quickly to ensure the flexibility and efficiency of energy management, providing strong guarantees for the sustainable development of smart entities.

[0026] According to an embodiment of the present invention, an embodiment of an integrated energy management method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] In this embodiment, a comprehensive energy management method is provided. Figure 1 is a flow chart of a comprehensive energy management method according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps: Step S101, obtaining historical energy data, equipment operating parameters and internal and external environmental data of the target smart entity, the internal and external environmental data including building characteristic data, seasonal data, internal temperature data of the building and internal humidity data of the building.

[0028] Specifically, when the target smart entity is a hospital, the historical energy data includes medical gas usage data, power consumption data, cooling and heating energy consumption data, and energy consumption distribution data of each department. The equipment operation parameters include the operation efficiency data of medical equipment, the power data of medical equipment, the utilization data of medical equipment, and the heat dissipation efficiency data of medical equipment. The internal and external environmental data include building characteristics data, seasonal data, internal temperature data of the building, and internal humidity data of the building. Among them, the historical energy data reflects the energy usage of the hospital in the past period of time. The medical gas usage data represents the consumption of medical gases such as oxygen and nitrogen. These data are very critical for evaluating the operating efficiency of medical equipment and the energy needs of the hospital. The power consumption data represents the power consumption of the hospital as a whole and in each department and each equipment, which helps to analyze the efficiency and peak period of power use. The cooling and heating energy consumption data represents the energy consumed by the hospital to maintain the indoor environment, especially the energy consumption of air conditioning and heating systems. The energy consumption distribution data of each department represents that the energy consumption of different departments varies due to their different functions and equipment usage. These data help to understand which departments are energy consumers, so as to carry out targeted optimization. Equipment operation parameters provide detailed information about the operation of medical equipment. The operation efficiency data of medical equipment characterizes the performance of the equipment during operation, such as whether the expected effect is achieved and whether there is excessive energy consumption. The power data of medical equipment characterizes the power consumption of the equipment during operation, which helps to analyze the energy efficiency ratio and potential energy saving space of the equipment. The utilization data of medical equipment characterizes the frequency and duration of use of the equipment, reflecting the use of the equipment and possible idle time. The heat dissipation efficiency data of medical equipment characterizes the heat generated by the equipment during operation and the heat dissipation effect, which is crucial for evaluating the stability and life of the equipment. Building characteristic data is a comprehensive data set covering multiple aspects related to building energy performance, structure and thermal efficiency. Building energy consumption data characterizes the actual energy consumption of the building during operation, including fuel consumption, electricity consumption, etc., reflecting the overall energy efficiency level of the building. By analyzing these data, the energy use pattern and efficiency of the building can be understood, thereby identifying potential energy saving opportunities. Building Information Model (BIM) data characterizes the three-dimensional structure and detailed information of the building, including the precise location and attributes of building elements, systems, equipment, etc., which helps to fully understand and analyze the design, construction and operation of the building. BIM data can be used to accurately simulate and analyze the energy performance and thermal comfort of buildings, as well as optimize energy use strategies. Thermal conductivity efficiency data characterizes the thermal performance of building envelopes such as exterior walls, roofs, and windows, including the thermal conductivity of materials, thermal insulation performance, and possible thermal bridge effects. Thermal conductivity efficiency data is essential for evaluating the thermal insulation performance of buildings and formulating effective energy-saving measures. Climate change in different seasons has a great impact on the energy consumption of hospitals, such as the difference in energy consumption for cooling in summer and heating in winter.The temperature and humidity of the indoor environment are crucial to the comfort of patients and the stable operation of medical equipment, reflecting the effectiveness of the hospital's environmental control system.

[0029] Step S102, inputting historical energy data, equipment operating parameters and internal and external environmental data into the trained first regulation model, and outputting the first energy demand result, the first regulation model is a hybrid model based on the self-attention mechanism and the gated recurrent unit.

[0030] Specifically, the first adjustment model is a hybrid model based on the self-attention mechanism and the gated recurrent unit. The self-attention mechanism can capture the global dependencies in the data, so that the model can pay attention to different parts of the data and perform weighted processing according to their importance. The gated recurrent unit is a variant of the recurrent neural network that can effectively process sequence data and capture the temporal dependencies in the sequence. Combining the self-attention mechanism with the gated recurrent unit can make full use of the advantages of both, and can not only pay attention to the global information in the data, but also capture the temporal dependencies in the sequence data. Prior to this step, the first adjustment model has been trained with a large amount of historical data to learn how to predict energy demand based on the input data. During the training process, the first adjustment model will continuously adjust its internal parameters to minimize the prediction error, thereby improving the accuracy of the prediction. When the historical energy data, equipment operating parameters, and internal and external environmental data are input into the trained first adjustment model, the model will predict the future energy demand of the hospital based on these data, that is, the first energy demand result.

[0031] Specifically, step S102 includes: Step S1021, obtaining a parameter embedding vector based on historical energy data, equipment operating parameters, and internal and external environmental data.

[0032] Specifically, the above step S1021 includes: Step a1, normalize the medical gas usage data, electricity consumption data, refrigeration and heating energy consumption data, energy consumption distribution data of each department, operation efficiency data of medical equipment, power data of medical equipment, utilization data of medical equipment and heat dissipation efficiency data of medical equipment, extract features, and obtain the first feature set.

[0033] Specifically, the first feature extraction subunit plays a vital role in the energy management and equipment monitoring system of the smart entity. This subunit is mainly responsible for normalizing the medical gas usage data, power consumption data, refrigeration and heating energy consumption data, energy consumption distribution data of each department, and the operation efficiency data, power data, utilization data and heat dissipation efficiency data of medical equipment. Normalization is a key step in data preprocessing. It can convert data of different units or ranges into a unified standard, thereby eliminating the impact of dimensional differences on the analysis results. After completing the data normalization, the first feature extraction subunit further extracts key features from these processed data. These features may involve the operating status of the equipment, energy utilization efficiency, energy consumption mode, etc. The feature extraction method may include statistical methods, machine learning or deep learning technology. The specific method to be selected depends on the characteristics of the data and the accuracy of the required features. For example, from the operation efficiency data, features such as the average operation efficiency, the maximum operation efficiency and the change trend of the operation efficiency can be extracted; from the utilization data, features such as the average utilization rate of the equipment and the stability of the utilization rate can be extracted. These extracted features can comprehensively reflect the working status and energy utilization of medical equipment, providing an important basis for subsequent data analysis, energy forecasting, and the formulation of equipment maintenance strategies. Finally, through normalization processing and fine feature extraction, the first feature extraction subunit will generate a comprehensive and critical first feature set, which is the basis for efficient decision-making of smart entity energy management and equipment monitoring systems.

[0034] Step a2, performing category feature coding conversion on the building characteristic data and the season data to obtain a second feature set.

[0035] Specifically, building characteristic data and seasonal data contain a large amount of category information and cannot be directly used for numerical calculations or training of machine learning models. Therefore, it is necessary to convert them into numerical features that can be understood by the model through category feature encoding conversion. Seasonal data usually contains four categories: spring, summer, autumn, and winter. Similar encoding methods can be used for the encoding of seasonal data. For example, label encoding can be used to map each season to an integer, or unique encoding can be used to convert each season into a four-bit binary vector. The encoded seasonal data can reflect the impact of different seasons on energy consumption and equipment operation, which helps the model to predict and make decisions more accurately. After the above encoding conversion, the second feature extraction subunit will generate a second feature set containing numerical features. This set not only contains the category information of building characteristic data and seasonal data, but also ensures that this information can be effectively used by subsequent models or algorithms.

[0036] Step a3, normalizing the building internal temperature data and the building internal humidity data, extracting features, and obtaining a third feature set.

[0037] Specifically, for temperature data and humidity data, due to their different value ranges and units, directly performing feature extraction or model training may lead to inaccurate results. Therefore, the third feature extraction subunit first normalizes these data and converts them into a unified scale. Through normalization and feature extraction, the third feature extraction subunit eventually generates a third feature set containing multiple key features.

[0038] Step a4, obtain the first feature weight, the second feature weight and the third feature weight, the first feature weight, the second feature weight and the third feature weight correspond one-to-one to the first feature set, the second feature set and the third feature set respectively, and the first feature weight, the second feature weight and the third feature weight are obtained through training of the first adjustment model.

[0039] Specifically, the input of the first adjustment model is the features in the first feature set, the second feature set and the third feature set, and the output is the target variables such as predicted energy consumption and equipment operation status. During the model training process, the parameters of the model are continuously adjusted to minimize the gap between the predicted value and the actual value, so as to obtain the optimal model weight.

[0040] Step a5, using the attention mechanism to fuse the first feature set, the second feature set and the third feature set based on the first feature weight, the second feature weight and the third feature weight to obtain a parameter embedding vector.

[0041] Specifically, the feature fusion subunit receives feature sets from the first feature extraction subunit, the second feature extraction subunit, and the third feature extraction subunit, as well as the corresponding feature weights. Using the attention mechanism, each feature set is weighted according to the feature weight. This process can be seen as "focusing" and "integrating" feature information from different sources, so that important features receive more attention, while minor features are appropriately suppressed. After weighted processing, the feature fusion subunit concatenates or adds feature vectors from different sources to obtain a parameter embedding vector that integrates the equipment operating status, environmental conditions, and building characteristics.

[0042] Step S1022, linearly transform the parameter embedding vector to obtain a first embedding vector, a second embedding vector and a third embedding vector, respectively. The first embedding vector is a search vector, the second embedding vector is an identification vector, and the third embedding vector is an information vector.

[0043] Specifically, the above step S1022 includes: Step b1, obtain a first weight matrix, a first bias item, a second weight matrix, a second bias item, a third weight matrix and a third bias item, the first weight matrix and the first bias item correspond to the first embedding vector, the second weight matrix and the second bias item correspond to the second embedding vector, and the third weight matrix and the third bias item correspond to the third embedding vector.

[0044] Specifically, a first weight matrix (W_1), a first bias item (b_1), a second weight matrix (W_2), a second bias item (b_2), a third weight matrix (W_3), and a third bias item (b_3) are obtained. The first weight matrix, the first bias item, the second weight matrix, the second bias item, the third weight matrix, and the third bias item are obtained by model training.

[0045] Step b2: linearly transform the parameter embedding vector using the first weight matrix and the first bias term to obtain a first embedding vector.

[0046] Specifically, the parameter embedding vector is linearly transformed using the first weight matrix and the first bias term, and the formula for the first embedding vector is as follows: [ \text{first embedding vector} = W_1 \times \text{parameter embedding vector} + b_1 ] Among them, (\times) represents matrix multiplication, and (W_1 \times \text{parameter embedding vector}) represents the matrix multiplication of the first weight matrix and the parameter embedding vector. The purpose of this step is to perform a dot product operation on each element of the parameter embedding vector with the corresponding row of the weight matrix to obtain a new vector. This process can capture the linear relationship between different features of the input data. After the matrix multiplication, the bias term (b_1) is added to increase the flexibility of the model. The bias term can be regarded as an adjustable threshold that allows the model to fine-tune the output without changing the weight matrix. This is crucial for the model to learn and adapt to different data distributions. The first embedding vector obtained is a representation of the parameter embedding vector after linear transformation. It may contain higher-level features or information in the original data, which will be passed to the next layer of the neural network for further processing.

[0047] Step b3: linearly transform the parameter embedding vector using the second weight matrix and the second bias term to obtain a second embedding vector.

[0048] Specifically, the parameter embedding vector is linearly transformed using the second weight matrix and the second bias term to obtain the second embedding vector, and the formula is as follows: [ \text{Second embedding vector} = W_2 \times \text{Parameter embedding vector} + b_2 ] Among them, (\times) represents matrix multiplication. Similarly, the parameter embedding vector is transformed into the second embedding vector through matrix multiplication and the addition of the bias term. This process is mathematically similar to the operation in step b1, but different parameters ((W_2) and (b_2)) are used, so different aspects or features of the input data are captured. Unlike the first embedding vector, the second embedding vector is used as an identification vector here. It may play a role of identification or classification in the network, helping the network to distinguish different input patterns or categories. By interacting with other embedding vectors (such as calculating similarity or distance), the identification vector can assist the network in making more accurate decisions or predictions.

[0049] Step b4: linearly transform the parameter embedding vector using the third weight matrix and the third bias term to obtain a third embedding vector.

[0050] Specifically, the parameter embedding vector is linearly transformed using the third weight matrix and the third bias term to obtain the third embedding vector, and the formula is as follows: [ \text{third embedding vector} = W_3 \times \text{parameter embedding vector} + b_3 ] Among them, (\times) represents matrix multiplication. Similar to the linear transformations in steps b2 and b3, the transformation in step b4 is also achieved by matrix multiplication and addition. However, it is important to understand that each transformation has its own specific weight matrix and bias term, which means that each embedding vector is extracted from the same original parameter embedding vector, but focuses on different features or information aspects. This specificity enables the model to understand and process the input data at multiple levels. The third embedding vector is designated as the information vector here. Its role may be to carry detailed information about the input data that is crucial to the final decision or prediction of the model. The information vector may contain complex feature representations of the input data, which will be further decoded and utilized in subsequent network layers. In a broader model context, the third embedding vector can be combined with the search vector and the identification vector to perform complex tasks such as information retrieval, semantic matching, or classification. By comparing and combining these embedding vectors, the model is able to make more detailed and accurate inferences.

[0051] Step S1023, using the scaled dot product attention mechanism to calculate the attention weight based on the first embedding vector and the second embedding vector.

[0052] Specifically, the above step S1023 includes: Step c1, determining the inner product of a target vector based on the first embedding vector and the second embedding vector.

[0053] Specifically, the formula for calculating the dot product between the first embedding vector and the second embedding vector is as follows: [ \text{Inner product of target vector} = \text{first embedding vector} \cdot \text{second embedding vector} ] Among them, cdot is the dot product operation, which means multiplying the corresponding elements of two vectors and then summing them. \text{} is a command commonly used in LaTeX mathematical typesetting, which is used to display the text in brackets in text mode instead of mathematical symbols. The purpose of this is to enhance the readability of the formula so that certain parts of the formula can be clearly labeled or explained in natural language. \text{target vector inner product}: This part is the label or description of the formula, which is used to explain that the following equation represents the inner product of two embedded vectors, and this result is called the "target vector inner product". \text{first embedded vector} and \text{second embedded vector}: These two parts are also text descriptions, referring to the two embedded vectors involved in the inner product operation. The "first embedded vector" and "second embedded vector" here are obtained by linear transformation in the previous step.

[0054] Step c2, determining a target scaling factor based on the second embedding vector.

[0055] Specifically, the scaling factor is usually used to prevent the dot product result from being too large, causing the softmax function to enter the saturation region. This scaling factor is usually the inverse of the square root of the dimension of the second embedding vector. Assuming that the dimension of the second embedding vector (second embedding vector) is (d_k), the scaling factor is: [ \text{Target scaling factor} = \frac{1}{\sqrt{d_k}} ] Here, frac and sqrt represent fraction and square root respectively.

[0056] In step c3, the attention weight is determined based on the target vector inner product and the target scaling factor using a normalized exponential function.

[0057] Specifically, the normalized exponential function (softmax function) is used to determine the attention weight based on the inner product of the target vector and the target scaling factor. The formula is as follows: [ \text{Attention weight} = \text{softmax}\left( \frac{\text{Inner product of target vector}}{\sqrt{d_k}} \right) ] Here, the softmax function maps each input element to the interval (0, 1), and the sum of all output elements is 1, thereby obtaining the attention weight, and \text{} is used to insert text in mathematical formulas. In LaTeX, math mode and text mode are two different typesetting modes. In math mode, all characters are treated as mathematical symbols, and the \text{} command allows you to insert normal text in math mode, so that the text is displayed in normal fonts and formats instead of mathematical symbols. For example, in this formula, "attention weight" and "target vector inner product" are both surrounded by \text{}, so they are displayed in the format of normal text. \left and \right are used to create automatically resizable brackets in mathematical formulas. \left( starts a left bracket, \right) ends a right bracket, and the size of the brackets is automatically adjusted according to the content they surround. In this formula, \left( and \right) surround the fraction \frac{\text{target vector inner product}}{\sqrt{d_k}} to ensure that the brackets can appropriately accommodate the size of this fraction.

[0058] The entire attention weight calculation process can be expressed as the following formula: [ \text{Attention weight} = \frac{\exp\left( \frac{\text{first embedding vector} \cdot \text{second embedding vector}}{\sqrt{d_k}} \right)}{\sum_i \exp\left( \frac{\text{first embedding vector} \cdot \text{the (i)th second embedding vector}}{\sqrt{d_k}} \right)} ] Among them, \text{} is used in the formula to display normal text content instead of mathematical symbols. In this formula, "attention weight", "first embedding vector" and "second embedding vector" are all surrounded by \text{} to ensure that they are displayed in the format of ordinary text, so that readers can understand the meaning of these words. \left and \right are used to automatically adjust the size of the brackets to adapt to the height and width of the content in the brackets. In this formula, \left( and \right) represent the left bracket and the right bracket, respectively, which surround the parameters of the exponential function \exp, so that the size of the brackets can automatically match the height of the internal formula, thereby improving the readability of the formula. \sum is a summation symbol, which means adding up a series of values. In this formula, \sum_i means summing all possible indexes i. The term being summed is \exp\left( \frac{\text{first embedding vector} \cdot \text{(i)th second embedding vector}}{\sqrt{d_k}} \right), which means that the sum of the results of the dot products of all "second embedding vectors" and "first embedding vectors" after the exponential operation is used as part of the denominator to normalize the attention weights. exp is the exponential function used to calculate the power of e. It is used to map the inner product result to the positive real number space, so that larger inner product values ​​will get larger weights. d_k is the dimension of the second embedding vector. When calculating the scaled dot product attention, in order to stabilize the calculation of the softmax function, the dot product result is usually scaled by the inverse of the square root of the second embedding vector dimension. This helps prevent the softmax function from entering the saturation region when the dot product result is large. ∑ is used to calculate the sum of the exp function values ​​corresponding to all second embedding vectors. This sum ensures that the final attention weight is a probability distribution, that is, the sum of the attention weights of all second embedding vectors is 1. i is an index variable used to traverse all second embedding vectors in the summation operation. For each second embedding vector, calculate its dot product with the first embedding vector and scale it by dividing it by the square root of the dimension of the second embedding vector (\sqrt{d_k})). Apply an exponential function (exp) to the scaled dot product result to obtain a non-negative weight value. Calculate the sum of the exp values ​​of all second embedding vectors. Divide the exp value of each second embedding vector by this sum to obtain the final attention weight. In this way, the model can dynamically assign attention weights to different parts of the input, thereby more accurately focusing on the most important information for the task. The summation in the denominator is performed on all possible second embedding vectors to ensure that the sum of the attention weights is 1. In practical applications, if there are multiple second embedding vectors, attention weights need to be calculated separately for each second embedding vector.

[0059] Step S1024, performing weighted summation on the third embedding vector based on the attention weight to obtain a self-attention layer output value.

[0060] Specifically, each third embedding vector is multiplied by its corresponding attention weight, and then all weighted third embedding vectors are added together. This process can be expressed mathematically as: [ \text{Self-attention layer output value} = \sum_n (\text{attention weight}_n \times \text{the (n)th third embedding vector}) ] Where n is the index variable used to traverse all third embedding vectors.

[0061] Step S1025, based on the reset gate, update gate and self-attention layer output value of the first adjustment model, determine the target hidden state, and output the self-attention layer output value of the target hidden state as the first energy demand result.

[0062] Specifically, the hidden state of the first conditioning model is combined with the output value of the self-attention layer, such as by simple concatenation, weighted summation, or more complex operations. The target hidden state is calculated by combining the reset gate, update gate, and the output value of the self-attention layer. This state contains both the long-term dependencies of the sequence and the attention of the self-attention layer to important information.

[0063] Step S103, inputting the building characteristic data and the first energy demand result into the trained second adjustment model, and outputting the second energy demand result, wherein the second adjustment model is a quantum support vector machine model.

[0064] Specifically, the building characteristic data provides detailed information about the hospital building itself, which helps to more accurately understand the impact of the building on energy demand. The first energy demand result is a preliminary energy demand forecast derived from the first adjustment model, which is based on historical energy data, equipment operating parameters, and internal and external environmental data. The second adjustment model uses the Quantum Support Vector Machine (QSVM) model. Quantum Support Vector Machine is a variant of the traditional Support Vector Machine (SVM). It combines the theory and algorithms of quantum computing and is designed to handle more complex and higher-dimensional data, and may provide more accurate prediction and classification results. The use of the QSVM model can handle the possible nonlinear relationship between the building characteristic data and the first energy demand result, as well as extract features that have a significant impact on energy demand from these data. The second adjustment model also needs to be trained before use to learn how to accurately predict energy demand from the input data. During the training process, the second adjustment model uses known data sets (including building characteristic data and corresponding energy demand data) to adjust its parameters to improve the accuracy of the prediction. When the building characteristic data and the first energy demand result are input into the trained second regulation model, the model will integrate this information and output a more accurate and detailed energy demand forecast, namely the second energy demand result. The second energy demand result takes into account the specific characteristics of the building and the preliminary energy demand forecast, so it is more accurate and targeted than the first energy demand result. The second energy demand result provides a more reliable basis for energy management for hospitals, helping them to formulate more accurate energy procurement plans, optimize energy use strategies, and implement more effective energy-saving measures.

[0065] Specifically, the building characteristic data includes building energy consumption data, building information model data and heat conduction efficiency data. The above step S103 includes: Step S1031, extracting features from building energy consumption data and heat conduction efficiency data to obtain first correction features.

[0066] Specifically, the collected building energy consumption data and thermal conductivity efficiency data are cleaned to remove outliers and missing values. According to the data characteristics and prediction task requirements, features closely related to energy demand are selected, such as energy consumption in different time periods, thermal conductivity in different areas, etc. The selected features are transformed, such as standardized, normalized or logarithmically transformed, to obtain the first corrected features.

[0067] Step S1032: convert the building information model data into a code to obtain a second correction feature.

[0068] Specifically, building information model data is a digital representation that integrates the physical and functional characteristics of a building project. It contains the geometric information, spatial relationships, material properties, and construction and management information of the building. In the energy management and equipment monitoring of smart entities, BIM data provides a rich source of information for analyzing and optimizing energy use. The second modified feature extraction unit first preprocesses the BIM data, including steps such as data cleaning, format conversion, and standardization to ensure the quality and consistency of the data. Then, using a special encoding algorithm or model, the BIM data is converted into a format that can be understood and utilized by the machine learning algorithm. This encoding conversion process may include multiple steps, such as feature selection, feature transformation, and feature dimensionality reduction. Feature selection aims to select the most relevant features for energy management and equipment monitoring from BIM data; feature transformation is to transform the original features to extract more meaningful feature representations; feature dimensionality reduction is to reduce the dimensionality of the data and reduce computational complexity and the risk of overfitting.

[0069] Step S1033: extracting features from the first energy demand result to obtain a third corrected feature.

[0070] Specifically, after receiving the first energy demand result, the third revised feature extraction unit will perform in-depth feature extraction on it. First, the unit will clean and preprocess the prediction results to ensure the quality and consistency of the data. Then, according to the characteristics of energy demand and the requirements of the prediction task, the unit will select features closely related to energy demand for extraction. These features may include the time distribution characteristics of energy demand, differences in energy demand in different regions, and the correlation between energy demand and weather conditions. By extracting these features, the third revised feature extraction unit can more comprehensively reflect the energy demand situation and provide a strong basis for subsequent analysis and optimization.

[0071] Step S1034, generating a fused correction feature based on the first correction feature, the second correction feature and the third correction feature.

[0072] Specifically, in the energy management and equipment monitoring of smart entities, features from different sources contain different aspects of information. The first correction feature mainly reflects the energy consumption and thermal conduction efficiency characteristics of the building; the second correction feature comes from the building information model data, which contains information such as the geometry, materials and spatial relationships of the building; and the third correction feature is a further refinement of the energy demand forecast results. Each of these features has unique value, but when used alone, they may not fully reflect the energy use and equipment operation status. Therefore, fusing these features can generate more representative fused correction features and improve the accuracy and reliability of subsequent tasks.

[0073] Step S1035, using quantum bits and quantum gate operations to perform quantum feature mapping on the fusion correction features to obtain quantum state fusion correction features.

[0074] Specifically, according to the requirements of quantum computing, the classical features are converted into a format that can be processed by quantum bits. According to the dimension and complexity of the fusion correction feature, the mapping unit determines the number of quantum bits required. These quantum bits are initialized to a specific initial state, usually a zero state or a one state, or a superposition state. The fusion correction feature is encoded on the quantum bit using quantum gate operations. This can be achieved by amplitude coding, angle coding or ground state coding. For example, amplitude coding can map the feature value to the amplitude of the quantum state, while angle coding converts the feature value to the angle of the quantum gate operation through a quantum rotation gate. According to the correlation between the features and the problem to be solved, a suitable quantum gate sequence is designed. These quantum gates can include single-bit gates (such as X gates, Y gates, Z gates, H gates, etc.) and multi-bit gates (such as CNOT gates, Toffoli gates, etc.) for rotating, entanglement and measuring quantum bits. By applying the designed quantum gate sequence, the initialized quantum bit is operated to convert it from the initial state to the quantum state corresponding to the fusion correction feature. After the quantum state conversion is completed, the mapping unit measures the quantum bit to extract the quantum state fusion correction feature. The measurement can be a projection measurement, a partial measurement or a weak measurement. The mapping unit processes the measurement results and converts them into a format that can be processed by a classical computer.

[0075] Step S1036, inputting the quantum state fusion correction feature into the trained quantum support vector machine model to obtain a second energy demand result.

[0076] Specifically, the above step S1036 includes: Step d1, input the quantum state fusion correction feature into the quantum register of the quantum computing platform, and use the quantum computing platform to perform a predefined quantum gate operation on the quantum state fusion correction feature to obtain a measurement result.

[0077] Specifically, the quantum gate operator receives the quantum state fusion correction characteristics, encodes these quantum states into a format acceptable to the quantum register, and inputs them into the quantum register of the quantum computing platform. Obtain a predefined quantum gate operation sequence, including single-bit gates (such as X, Y, Z, Hadamard gates, etc.) and multi-bit gates (such as CNOT, Toffoli gates, etc.). The operation sequence may include steps such as initializing the quantum state, rotating the quantum state, performing entanglement operations, and intermediate measurements. Using the computing power of the quantum computing platform, perform a predefined quantum gate operation sequence on the quantum state in the quantum register. During the execution process, the quantum state will evolve according to the operation sequence and gradually approach the solution space of the problem. After the execution of the quantum gate operation sequence is completed, the quantum state in the quantum register is measured. The measurement can be a projection measurement or a weak measurement, and the measurement result is output as a classical bit stream to provide input for the subsequent decoding process.

[0078] Step d2, decoding the measurement result to obtain a second energy demand result.

[0079] Specifically, the measurement results are decoded according to the output format and encoding method of the quantum computing platform. The decoded results are post-processed as necessary, such as denoising, smoothing or correction operations, to improve the accuracy of the results. According to the needs of the actual problem, the decoded results can be further optimized or converted to better meet the format and requirements of the energy demand forecast. The decoded and optimized results are output as the second energy demand result.

[0080] Step S104: formulating an energy regulation strategy based on the second energy demand result, and regulating the energy of the smart entity based on the energy regulation strategy.

[0081] Specifically, the second energy demand results are analyzed in depth to understand the peak and trough periods of energy demand, the changing trends of daily and weekly average energy demand, and possible seasonal fluctuations. According to the results of data analysis, corresponding energy regulation strategies are formulated. For example, it is determined to increase energy supply during peak energy demand periods, and reduce supply or perform equipment maintenance during trough periods. At the same time, it is also possible to consider reducing energy consumption by optimizing the operation mode and time of the equipment. When formulating strategies, it is also necessary to consider how to reduce overall energy consumption by improving energy efficiency. For example, the operation mode of the air-conditioning system can be optimized to reduce unnecessary energy consumption. Smart entities are usually equipped with advanced energy management systems, which can realize automatic control of hospital energy equipment through programming. Based on the formulated energy regulation strategies, these strategies are converted into instructions for the automatic control system. In the process of implementing the energy regulation strategy, it is necessary to monitor the energy use of the hospital in real time, and fine-tune the strategy according to the actual situation to ensure the effectiveness of the energy regulation strategy and respond to possible abnormal situations in a timely manner. After the energy regulation strategy is implemented, its effect needs to be evaluated regularly. This includes comparing key indicators such as energy consumption and equipment operation efficiency before and after the implementation of the strategy. According to the evaluation results, problems in the strategy are discovered and adjusted in a timely manner. At the same time, the experience and lessons learned during the implementation process can also be fed back into the strategy formulation process to continuously improve and optimize the energy regulation strategy.

[0082] The comprehensive energy management method provided in this embodiment realizes efficient management and precise regulation of smart energy through comprehensive and real-time data collection and multi-level energy demand forecasting. First, through equipment operation parameter monitoring and internal and external environmental data monitoring, the operation parameters of each device in the smart entity and the internal and external environmental data can be obtained in real time. These data cover key operation parameters such as equipment operation efficiency data, equipment power data, equipment utilization data and equipment heat dissipation efficiency data, as well as important environmental factors such as building characteristics, seasons, temperature, humidity, etc. This all-round data collection mechanism not only ensures the real-time and accuracy of the data, but also provides reliable data support for subsequent energy demand forecasting and strategy formulation. Secondly, by adopting a two-layer prediction, using historical energy data, real-time equipment operation parameters and internal and external environmental data, combined with a hybrid model based on self-attention mechanism and gated recurrent unit, preliminary energy demand forecasting is performed. This model combines the sensitivity of the self-attention mechanism to time series data and the processing ability of the gated recurrent unit to sequence data, so as to more accurately capture the complex relationship between data and improve the accuracy of the prediction. Afterwards, based on the first-layer prediction output, the impact of building characteristic data on energy demand is further considered. By introducing the quantum support vector machine model, the first-layer prediction results can be corrected and improved to generate more accurate second-layer energy demand prediction results. The powerful computing power and high-dimensional data processing capabilities of the quantum support vector machine model can still maintain the stability and accuracy of the prediction when facing a complex and changeable data environment. Finally, according to the second-layer energy demand prediction results, an adjustment strategy that meets the current energy demand is intelligently generated, which not only takes into account the real-time energy demand, but also takes into account the architectural characteristics and environmental changes of the smart entity to ensure the effectiveness and pertinence of the adjustment measures. Finally, in the energy adjustment execution stage, the method adjusts the energy of the smart entity in real time according to the strategy. By accurately controlling the operating parameters of the equipment, optimizing the energy allocation plan, etc., the efficient utilization of energy and the goal of energy conservation and emission reduction are achieved. At the same time, the method also has a real-time feedback mechanism, which can continuously adjust and optimize the adjustment strategy according to the actual effect to ensure the continuity and adaptability of energy management. In summary, the comprehensive energy management method provided in this embodiment effectively solves many drawbacks in traditional energy management methods through real-time data collection, accurate energy demand prediction and intelligent energy adjustment strategy. Especially when facing changes in the smart entity environment and energy demand, this method can respond and adjust quickly to ensure the flexibility and efficiency of energy management, providing strong guarantees for the sustainable development of smart entities.

[0083] This solution can also be applied to smart entities in different industries, such as oil parks, communication plants, thermal power plants, etc. Taking the oil park as an example, when the target smart entity is the oil park, the historical energy data includes oil consumption data, natural gas usage data, electricity consumption data, cooling and heating energy consumption data and regional energy consumption distribution data. The equipment operation parameters include the operating efficiency data of the oil equipment, the power data of the equipment, the utilization data of the equipment and the heat dissipation efficiency data of the equipment. The internal and external environmental data include the geographical characteristics data of the oil park, seasonal data, the internal temperature data of the oil park and the internal humidity data. The historical energy data, equipment operation parameters and internal and external environmental data of the target oil park are obtained. The internal and external environmental data include building characteristic data, seasonal data, building internal temperature data and building internal humidity data. The historical energy data, equipment operation parameters and internal and external environmental data are input into the trained first regulation model to output the first energy demand result. The first regulation model is a hybrid model based on the self-attention mechanism and the gated recurrent unit. The building characteristic data and the first energy demand result are input into the trained second regulation model to output the second energy demand result. The second regulation model is a quantum support vector machine model. An energy regulation strategy is formulated based on the second energy demand result, and the energy of the target intelligent entity is adjusted based on the energy regulation strategy.

[0084] Specifically, the oil park is a complex industrial environment that involves the consumption of multiple energy sources, including oil, natural gas, and electricity. Energy management in the park is crucial to improving operational efficiency, reducing costs, and ensuring environmental safety. However, since the energy consumption of the oil park is affected by many factors, such as the operating status of the equipment, changes in the external environment, and the geographical characteristics of the park, energy management faces great challenges. Collect historical energy consumption data of the oil park, including oil consumption, natural gas usage, and electricity consumption. These data reflect the energy usage of the park in different time periods and under different conditions. Collect data such as the operating efficiency, power, utilization, and heat dissipation efficiency of the oil equipment. These data help to understand the performance status and energy utilization efficiency of the equipment. Collect data such as the geographical characteristics, seasonal changes, internal temperature, and humidity of the park. These data are crucial for predicting energy demand because they directly affect the operating efficiency and energy consumption of the equipment. The collected historical energy data, equipment operating parameters, and internal and external environmental data are input into a hybrid model based on the self-attention mechanism and gated recurrent units. This model can capture the time series dependencies and long-term trends in the data, thereby outputting the first energy demand results. The self-attention mechanism enables the model to focus on key information in the data, while the gated recurrent unit helps to deal with long-term dependencies in time series data. The building characteristic data and the first energy demand results are input into the quantum support vector machine model. This model uses the principles of quantum computing to handle complex nonlinear relationships and further improve the accuracy of energy demand forecasting. The quantum support vector machine model has strong classification and regression capabilities and can handle high-dimensional data and complex pattern recognition problems, thereby providing more accurate energy demand forecasts. Based on the second energy demand results, targeted energy regulation strategies are formulated. These strategies may include adjusting the operating time of equipment, optimizing energy allocation plans, and making energy reserves in advance. By implementing these strategies, efficient, economical and environmentally friendly regulation of energy in the oil park can be achieved, improving the overall operational efficiency and sustainability of the park. This solution combines the actual scenarios of the oil park and achieves high-precision prediction and intelligent regulation of energy demand by comprehensively using advanced machine learning models and multiple data sources. This not only helps to improve the energy efficiency of the oil park, reduce costs and reduce environmental pollution, but also provides strong support for the sustainable development of the park.

[0085] The above is an embodiment of the integrated energy management system provided by the present application. The following describes other embodiments of the integrated energy management provided by the present application. Please refer to the following for details.

[0086] In this embodiment, a comprehensive energy management system is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the system described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0087] This embodiment provides a comprehensive energy management system, such as Figure 2 As shown, including: The acquisition module 201 is used to acquire the historical energy data, equipment operation parameters and internal and external environmental data of the target smart entity, and the internal and external environmental data include building characteristic data, seasonal data, internal temperature data of the building and internal humidity data of the building.

[0088] The first output module 202 is used to input historical energy data, equipment operating parameters and internal and external environmental data into a trained first regulation model, and output a first energy demand result. The first regulation model is a hybrid model based on a self-attention mechanism and a gated recurrent unit.

[0089] The second output module 203 is used to input the building characteristic data and the first energy demand result into the trained second adjustment model, and output the second energy demand result. The second adjustment model is a quantum support vector machine model.

[0090] The adjustment module 204 is used to formulate an energy adjustment strategy based on the second energy demand result, and adjust the energy of the target smart entity based on the energy adjustment strategy.

[0091] In some optional implementations, when the target smart entity is a hospital, the historical energy data includes medical gas usage data, electricity consumption data, cooling and heating energy consumption data, and energy consumption distribution data of each department, the equipment operation parameters include medical equipment operation efficiency data, medical equipment power data, medical equipment utilization data, and medical equipment heat dissipation efficiency data, the internal and external environment data include building characteristic data, season data, building internal temperature data, and building internal humidity data, and the first output module 202 includes: An embedding vector generating unit, used for obtaining a parameter embedding vector based on historical energy data, equipment operation parameters and internal and external environmental data; A linear transformation unit, used to perform a linear transformation on the parameter embedding vector to obtain a first embedding vector, a second embedding vector and a third embedding vector, respectively, wherein the first embedding vector is a search vector, the second embedding vector is an identification vector, and the third embedding vector is an information vector; An attention weight calculation unit, configured to calculate an attention weight based on the first embedding vector and the second embedding vector using a scaled dot product attention mechanism; A weighting unit, configured to perform a weighted summation on the third embedding vector based on the attention weight to obtain a self-attention layer output value; The first energy demand result output unit is used to determine the target hidden state based on the reset gate, update gate and self-attention layer output value of the first adjustment model, and output the self-attention layer output value of the target hidden state as the first energy demand result.

[0092] In some optional implementations, the embedding vector generation unit includes: a first feature extraction subunit, configured to normalize the medical gas usage data, the power consumption data, the cooling and heating energy consumption data, the energy consumption distribution data of each department, the operation efficiency data of the medical equipment, the power data of the medical equipment, the utilization rate data of the medical equipment, and the heat dissipation efficiency data of the medical equipment, extract features, and obtain a first feature set; The second feature extraction subunit is used to perform category feature encoding conversion on the building characteristic data and the season data to obtain a second feature set; A third feature extraction subunit is used to normalize the building internal temperature data and the building internal humidity data, extract features, and obtain a third feature set; A weight acquisition subunit, used to acquire a first feature weight, a second feature weight, and a third feature weight, wherein the first feature weight, the second feature weight, and the third feature weight correspond to the first feature set, the second feature set, and the third feature set, respectively, and the first feature weight, the second feature weight, and the third feature weight are obtained by training the first adjustment model; The feature fusion subunit is used to use the attention mechanism to fuse the first feature set, the second feature set and the third feature set based on the first feature weight, the second feature weight and the third feature weight to obtain a parameter embedding vector.

[0093] In some optional implementations, the linear transformation unit includes: A preset data acquisition subunit is used to acquire a first weight matrix, a first bias item, a second weight matrix, a second bias item, a third weight matrix and a third bias item, wherein the first weight matrix and the first bias item correspond to the first embedding vector, the second weight matrix and the second bias item correspond to the second embedding vector, and the third weight matrix and the third bias item correspond to the third embedding vector; A first linear transformation subunit, configured to perform a linear transformation on the parameter embedding vector using a first weight matrix and a first bias term to obtain a first embedding vector; A second linear transformation subunit, configured to perform a linear transformation on the parameter embedding vector using a second weight matrix and a second bias term to obtain a second embedding vector; The third linear transformation subunit is used to perform a linear transformation on the parameter embedding vector using a third weight matrix and a third bias term to obtain a third embedding vector.

[0094] In some optional implementations, the attention weight calculation unit includes a target vector inner product determining subunit, configured to determine a target vector inner product based on the first embedding vector and the second embedding vector; a target scaling factor determining subunit, configured to determine a target scaling factor based on the second embedding vector; The attention weight determination subunit is used to determine the attention weight based on the target vector inner product and the target scaling factor using a normalized exponential function.

[0095] In some optional implementations, the building characteristic data includes building energy consumption data, building information model data and heat transfer efficiency data, and the second output module 203 includes: A first correction feature extraction unit is used to extract features from building energy consumption data and heat conduction efficiency data to obtain a first correction feature; A second correction feature extraction unit is used to perform encoding conversion on the building information model data to obtain a second correction feature; A third correction feature extraction unit, used for extracting features from the first energy demand result to obtain a third correction feature; a fused correction feature generating unit, configured to generate a fused correction feature based on the first correction feature, the second correction feature and the third correction feature; A mapping unit, used for performing quantum feature mapping on the fusion correction feature by using quantum bits and quantum gate operations to obtain a quantum state fusion correction feature; The second energy demand result generating unit is used to input the quantum state fusion correction feature into the trained quantum support vector machine model to obtain the second energy demand result.

[0096] In some optional implementations, the second energy demand result generating unit includes: A quantum gate operation subunit, used to input the quantum state fusion correction feature into the quantum register of the quantum computing platform, and use the quantum computing platform to perform a predefined quantum gate operation on the quantum state fusion correction feature to obtain a measurement result; The decoding subunit is used to decode the measurement result to obtain a second energy demand result.

[0097] The comprehensive energy management system provided in this embodiment realizes efficient management and precise regulation of smart energy through comprehensive and real-time data collection and multi-level energy demand forecasting. First, through equipment operation parameter monitoring and internal and external environmental data monitoring, the operation parameters of each device in the smart entity and the internal and external environmental data can be obtained in real time. These data cover key operation parameters such as equipment operation efficiency data, equipment power data, equipment utilization data and equipment heat dissipation efficiency data, as well as important environmental factors such as building characteristics, seasons, temperature, humidity, etc. This all-round data collection mechanism not only ensures the real-time and accuracy of the data, but also provides reliable data support for subsequent energy demand forecasting and strategy formulation. Secondly, by adopting a two-layer prediction, using historical energy data, real-time equipment operation parameters and internal and external environmental data, combined with a hybrid model based on self-attention mechanism and gated recurrent unit, preliminary energy demand forecasting is performed. This model combines the sensitivity of the self-attention mechanism to time series data and the processing ability of the gated recurrent unit to sequence data, so that it can more accurately capture the complex relationship between data and improve the accuracy of the prediction. Afterwards, based on the first-layer prediction output, the impact of building characteristic data on energy demand is further considered. By introducing the quantum support vector machine model, the first-layer prediction results can be corrected and improved to generate more accurate second-layer energy demand prediction results. The powerful computing power and high-dimensional data processing capabilities of the quantum support vector machine model can still maintain the stability and accuracy of the prediction when facing a complex and changeable data environment. Finally, according to the second-layer energy demand prediction results, an adjustment strategy that meets the current energy demand is intelligently generated, which not only takes into account the real-time energy demand, but also takes into account the architectural characteristics and environmental changes of the smart entity to ensure the effectiveness and pertinence of the adjustment measures. Finally, in the energy adjustment execution stage, the method adjusts the energy of the smart entity in real time according to the strategy. By accurately controlling the operating parameters of the equipment, optimizing the energy allocation plan, etc., the efficient utilization of energy and the goal of energy conservation and emission reduction are achieved. At the same time, the method also has a real-time feedback mechanism, which can continuously adjust and optimize the adjustment strategy according to the actual effect to ensure the continuity and adaptability of energy management. In summary, the comprehensive energy management method provided in this embodiment effectively solves many drawbacks in traditional energy management methods through real-time data collection, accurate energy demand prediction and intelligent energy adjustment strategy. Especially when facing changes in the smart entity environment and energy demand, this method can respond and adjust quickly to ensure the flexibility and efficiency of energy management, providing strong guarantees for the sustainable development of smart entities.

[0098] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

Claims

1. A comprehensive energy management method, applied to a smart entity, characterized in that: The method comprises: Acquire historical energy data, equipment operation parameters, and internal and external environmental data of the target smart entity, wherein the internal and external environmental data include building characteristic data, seasonal data, internal temperature data of the building, and internal humidity data of the building; Inputting the historical energy data, the equipment operating parameters, and the internal and external environment data into a trained first regulation model, and outputting a first energy demand result, wherein the first regulation model is a hybrid model based on a self-attention mechanism and a gated recurrent unit; Inputting the building characteristic data and the first energy demand result into a trained second regulation model, and outputting a second energy demand result, wherein the second regulation model is a quantum support vector machine model; An energy regulation strategy is formulated based on the second energy demand result, and the energy of the target smart entity is regulated based on the energy regulation strategy.

2. The method according to claim 1, characterized in that When the target smart entity is a hospital, the historical energy data includes medical gas usage data, electricity consumption data, cooling and heating energy consumption data, and energy consumption distribution data of each department; the equipment operation parameters include medical equipment operation efficiency data, medical equipment power data, medical equipment utilization data, and medical equipment heat dissipation efficiency data; the internal and external environment data include building characteristic data, seasonal data, building internal temperature data, and building internal humidity data. The inputting the historical energy data, the equipment operation parameters and the internal and external environment data into the trained first adjustment model and outputting the first energy demand result comprises: Obtaining a parameter embedding vector based on the historical energy data, the equipment operating parameters and the internal and external environmental data; Performing a linear transformation on the parameter embedding vector to obtain a first embedding vector, a second embedding vector and a third embedding vector, respectively, wherein the first embedding vector is a search vector, the second embedding vector is an identification vector, and the third embedding vector is an information vector; The attention weight is calculated based on the first embedding vector and the second embedding vector using the scaled dot product attention mechanism; Performing weighted summation on the third embedding vector based on the attention weight to obtain a self-attention layer output value; Based on the reset gate, update gate and self-attention layer output value of the first regulation model, a target hidden state is determined, and the self-attention layer output value of the target hidden state is output as a first energy demand result.

3. The method according to claim 2, characterized in that The obtaining of the parameter embedding vector based on the historical energy data, the equipment operating parameters and the internal and external environmental data includes: Normalizing the medical gas usage data, electricity consumption data, cooling and heating energy consumption data, energy consumption distribution data of each department, operation efficiency data of medical equipment, power data of medical equipment, utilization data of medical equipment, and heat dissipation efficiency data of medical equipment, extracting features, and obtaining a first feature set; Performing category feature coding conversion on the building characteristic data and the season data to obtain a second feature set; Normalizing the building internal temperature data and the building internal humidity data, extracting features, and obtaining a third feature set; Obtaining a first feature weight, a second feature weight, and a third feature weight, wherein the first feature weight, the second feature weight, and the third feature weight correspond one-to-one to the first feature set, the second feature set, and the third feature set, respectively, and the first feature weight, the second feature weight, and the third feature weight are obtained by training the first adjustment model; The first feature set, the second feature set and the third feature set are subjected to feature fusion by using an attention mechanism based on the first feature weight, the second feature weight and the third feature weight to obtain a parameter embedding vector.

4. The method according to claim 3, characterized in that The linear transformation of the parameter embedding vector to obtain a first embedding vector, a second embedding vector and a third embedding vector respectively includes: Obtain a first weight matrix, a first bias item, a second weight matrix, a second bias item, a third weight matrix and a third bias item, wherein the first weight matrix and the first bias item correspond to a first embedding vector, the second weight matrix and the second bias item correspond to a second embedding vector, and the third weight matrix and the third bias item correspond to a third embedding vector; Performing a linear transformation on the parameter embedding vector using the first weight matrix and the first bias term to obtain a first embedding vector; Performing a linear transformation on the parameter embedding vector using the second weight matrix and the second bias term to obtain a second embedding vector; The parameter embedding vector is linearly transformed by using the third weight matrix and the third bias term to obtain a third embedding vector.

5. The method according to claim 4, characterized in that The method of using a scaled dot product attention mechanism to calculate an attention weight based on the first embedding vector and the second embedding vector includes: Determine a target vector inner product based on the first embedding vector and the second embedding vector; determining a target scaling factor based on the second embedding vector; An attention weight is determined based on the target vector inner product and the target scaling factor using a normalized exponential function.

6. The method according to claim 1, characterized in that The building characteristic data includes building energy consumption data, building information model data and heat conduction efficiency data, and the inputting of the building characteristic data and the first energy demand result into the trained second adjustment model to output the second energy demand result includes: Extracting features from the building energy consumption data and heat conduction efficiency data to obtain a first correction feature; Performing encoding conversion on the building information model data to obtain a second correction feature; Performing feature extraction on the first energy demand result to obtain a third corrected feature; Generate a fused correction feature based on the first correction feature, the second correction feature and the third correction feature; Performing quantum feature mapping on the fusion correction feature by using quantum bits and quantum gate operations to obtain quantum state fusion correction features; The quantum state fusion correction feature is input into the trained quantum support vector machine model to obtain a second energy demand result.

7. The method according to claim 6, characterized in that The step of inputting the quantum state fusion correction feature into a trained quantum support vector machine model to obtain a second energy demand result includes: Inputting the quantum state fusion correction feature into a quantum register of a quantum computing platform, and using the quantum computing platform to perform a predefined quantum gate operation on the quantum state fusion correction feature to obtain a measurement result; The measurement result is decoded to obtain a second energy demand result.

8. An integrated energy management system, characterized in that: The system comprises: An acquisition module, used to acquire historical energy data, equipment operation parameters, and internal and external environmental data of a target smart entity, wherein the internal and external environmental data include building characteristic data, seasonal data, internal temperature data of a building, and internal humidity data of a building; a first output module, configured to input the historical energy data, the equipment operation parameters, and the internal and external environment data into a trained first regulation model, and output a first energy demand result, wherein the first regulation model is a hybrid model based on a self-attention mechanism and a gated recurrent unit; A second output module, used for inputting the building characteristic data and the first energy demand result into a trained second regulation model, and outputting a second energy demand result, wherein the second regulation model is a quantum support vector machine model; The adjustment module is used to formulate an energy adjustment strategy based on the second energy demand result, and adjust the energy of the target smart entity based on the energy adjustment strategy.

Images