A building energy-saving control method and system based on energy consumption parameters

Abstract

The present invention discloses an energy-saving control method and system for buildings based on energy consumption parameters, which relates to the field of building energy conservation. The energy-saving control method for buildings based on energy consumption parameters includes the following steps: S1, obtaining parameters and constructing a building energy consumption database; S2, extracting the historical overall energy consumption parameters and historical block energy consumption parameters of the building energy consumption; S3, constructing an energy consumption prediction model and training and optimizing the energy consumption prediction model; S4, predicting the future overall energy consumption trend and future block energy consumption trend; S5, presetting an energy consumption control database and an energy consumption matching rule, and matching the energy consumption control strategies in the energy consumption control database; S6, implementing the energy consumption control strategies and updating the building energy consumption database. By using the historical energy consumption data of the building, the present invention realizes the accurate prediction of the future energy consumption trend, predicts the overall energy consumption in advance and details it to the energy consumption of each specific area, providing data support for precise control.

Description

Technical Field

[0001] The present invention relates to the field of building energy conservation, and in particular to a building energy conservation control method and system based on energy consumption parameters. Background Art

[0002] With the continuous progress of society, buildings are constantly developing and expanding. Buildings have a certain impact on carbon emissions in time and space, which in turn will have a chain effect on the environment. The carbon emissions generated by the construction industry account for a high proportion of the total carbon dioxide emissions, and are far higher than those of the transportation industry. This makes building energy conservation important in multiple dimensions, covering multiple fields such as the environment, economy, society and technology.

[0003] Building energy consumption parameters are the core link between physical space and digital management. They not only reveal the essential laws of energy flow, but also drive the whole chain of actions from technological improvement to policy formulation. Building energy consumption is the core indicator for quantifying the efficiency of building energy use, just like the key indicators in the physical examination report. Energy consumption parameters can locate high-energy-consuming links in the building, thereby optimizing the adjustment of building energy consumption, saving energy consumption when the building is in use, improving the efficiency of building use, and ultimately achieving scientific energy conservation with data.

[0004] However, the existing building energy-saving control methods based on energy consumption parameters do not take into account the block analysis of energy consumption in the building when they are used, resulting in that the existing building energy-saving control methods only perform energy-saving control by analyzing the overall data of building energy consumption when they are used, which greatly affects the accuracy of building energy-saving control and makes it impossible for the building energy-saving control methods to perform energy-saving control on the building's energy consumption in a timely and effective manner when they are used.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes a building energy-saving control method and system based on energy consumption parameters to overcome the above technical problems existing in the existing related art.

[0007] In order to achieve the above object, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a building energy-saving control method based on energy consumption parameters is provided, comprising the following steps:

[0009] S1. Obtain building specification parameters and real-time building energy consumption parameters, and build a building energy consumption database;

[0010] S2. Based on the building energy consumption database, extract the historical overall energy consumption parameters and historical block energy consumption parameters;

[0011] S3. Based on the historical overall energy consumption parameters and historical block energy consumption parameters, construct and train an optimized energy consumption prediction model;

[0012] As a preferred solution, the constructing and training of the optimized energy consumption prediction model based on the historical overall energy consumption parameters and historical block energy consumption parameters includes the following steps:

[0013] S31. Standardize the historical overall energy consumption parameters and historical block energy consumption parameters;

[0014] S32. Extract the overall time series feature values and block time series feature values of the processed historical overall energy consumption parameters and historical block energy consumption parameters to construct an energy consumption prediction model;

[0015] S33. Input the overall time series feature values and block time series feature values into the energy consumption prediction model for training, and optimize the energy consumption prediction model using the Bayesian optimization algorithm;

[0016] S34. Perform cross-validation on the optimized energy consumption prediction model, and evaluate the performance parameters of the energy consumption prediction model after validation;

[0017] S35. Preset a performance grading standard, grade the performance parameters based on the performance grading standard, and make a usage judgment on the energy consumption prediction model according to the performance grading results. If it meets the performance grading standard, use the energy consumption prediction model; if it does not meet the performance grading standard, return to S33 for re-optimization.

[0018] S4. Input the real-time overall energy consumption parameters and real-time block energy consumption parameters into the optimized energy consumption prediction model to predict the future overall energy consumption trend and future block energy consumption trend;

[0019] As a preferred solution, the inputting of the real-time overall energy consumption parameters and real-time block energy consumption parameters into the optimized energy consumption prediction model to predict the future overall energy consumption trend and future block energy consumption trend includes the following steps:

[0020] S41. Perform data cleaning and normalization on the real-time overall energy consumption parameters and real-time block energy consumption parameters;

[0021] S42. Input the processed real-time overall energy consumption parameters and real-time block energy consumption parameters into the energy consumption prediction model for multi-modal trend prediction to obtain the future overall energy consumption trend and future block energy consumption trend;

[0022] S43. Use the fuzzy algorithm to compare the change intervals of the future overall energy consumption trend and future block energy consumption trend;

[0023] As a preferred solution, the using of the fuzzy algorithm to compare the change intervals of the future overall energy consumption trend and future block energy consumption trend includes the following steps:

[0024] S431. Extract the overall trend characteristic parameters and block trend characteristic parameters of the future overall energy consumption trend and the future block energy consumption trend, and perform data cleaning and integration;

[0025] S432. Define the fuzzy membership degree criteria for the overall trend characteristic parameters and block trend characteristic parameters, and construct a fuzzy rule base;

[0026] S433. Calculate the overlap rate of the future overall energy consumption data and the future block energy consumption data through fuzzy inference based on the fuzzy membership degree criteria;

[0027] As an optimal solution, calculating the overlap rate of the future overall energy consumption data and the future block energy consumption data through fuzzy inference based on the fuzzy membership degree criteria includes the following steps:

[0028] S4331. Preset a sliding window mechanism to align the time series, and perform normalization processing on the future overall energy consumption data and the future block energy consumption data;

[0029] S4332. Define the fuzzy membership degree functions of the future overall energy consumption data and the future block energy consumption data, and construct fuzzy inference rules;

[0030] S4333. Calculate the overlap rate of the future overall energy consumption data and the future block energy consumption data based on the fuzzy membership degree functions and fuzzy inference rules.

[0031] As an optimal solution, the calculation formula for calculating the overlap rate of the future overall energy consumption data and the future block energy consumption data based on the fuzzy membership degree functions and fuzzy inference rules is:

[0032] ;

[0033] Among them, W is the overlap rate of the future overall energy consumption data and the future block energy consumption data;

[0034] is the membership degree function of the overall energy consumption data;

[0035] is the membership degree function of the block energy consumption data;

[0036] X is the independent variable of the energy consumption data;

[0037] d is the integral variable of the fuzzy membership degree function over the entire domain.

[0038] S434. Preset an overlap difference rule and a fuzzy clustering algorithm database, and match the overlap rate with the fuzzy clustering algorithm database according to the overlap difference rule;

[0039] S435. Analyze the outliers of the future overall energy consumption trend and the future block energy consumption trend using a matching fuzzy clustering algorithm, and add annotations to the outliers.

[0040] S44. Optimize and adjust the future overall energy consumption trend and the future block energy consumption trend according to the comparison result of the change interval, and verify the prediction result after adjustment.

[0041] S5. Preset an energy consumption control database and an energy consumption matching rule, and match the predicted future overall energy consumption trend and the future block energy consumption trend with corresponding energy consumption control strategies according to the energy consumption matching rule;

[0042] As an optimal solution, the preset energy consumption control database and the energy consumption matching rule, matching the predicted future overall energy consumption trend and the future block energy consumption trend with corresponding energy consumption control strategies according to the energy consumption matching rule, include the following steps:

[0043] S51. Preset an energy consumption control database to store energy consumption control strategies and classify them according to equipment type and energy consumption intensity;

[0044] S52. Set an energy consumption matching rule to divide the energy consumption matching threshold, and set a building time weight value and a building equipment weight value based on building specification parameters and real-time building energy consumption parameters;

[0045] S53. Calculate the future overall energy consumption value and the future block energy consumption value of the future overall energy consumption trend and the future block energy consumption trend according to the building time weight value and the building equipment weight value;

[0046] As an optimal solution, calculating the future overall energy consumption value and the future block energy consumption value of the future overall energy consumption trend and the future block energy consumption trend according to the building time weight value and the building equipment weight value includes the following steps:

[0047] S531. Standardize the data of the building time weight value and the building equipment weight value;

[0048] S532. Construct a composite weight matrix based on the standardized building time weight value and building equipment weight value;

[0049] S533. Extract the characteristic parameters of the overall energy consumption trend and the future block energy consumption trend, and calculate the future overall energy consumption value and the future block energy consumption value according to the composite weight matrix;

[0050] S534. Verify and adjust the future overall energy consumption value and the future block energy consumption value, and output the verified and adjusted future overall energy consumption value and future block energy consumption value.

[0051] S54. Match the future overall energy consumption value and the future block energy consumption value with the energy consumption matching threshold, and obtain the overall energy consumption control strategy and the block energy consumption control strategy based on the matching result;

[0052] S55. Analyze the conflict parameters between the overall energy consumption control strategy and the block energy consumption control strategy, optimize the overall energy consumption control strategy and the block energy consumption control strategy according to the conflict parameters, and verify and output the optimized overall energy consumption control strategy and block energy consumption control strategy.

[0053] S6. Execute the matched energy consumption control strategy, calculate the virtual energy consumption prediction value, compare the real-time collected electricity energy consumption with the virtual energy consumption prediction value, optimize and iterate the energy consumption prediction model based on the comparison result, and update the building energy consumption database.

[0054] As an optimal solution, the step of executing the matched energy consumption control strategy, calculating the virtual energy consumption prediction value, comparing the real-time collected electricity energy consumption with the virtual energy consumption prediction value, optimizing and iterating the energy consumption prediction model based on the comparison result, and updating the building energy consumption database includes the following steps:

[0055] S61. Extract the control strategy parameters in the implementation of the energy consumption control strategy, and substitute the control strategy parameters into the energy consumption prediction model to calculate the virtual energy consumption prediction value;

[0056] S62. Preset an energy consumption error threshold and an error adjustment strategy, and match the comparison result of the real-time collected electricity energy consumption and the virtual energy consumption prediction value with the energy consumption error threshold;

[0057] S63. Select an error adjustment strategy based on the matching result of the energy consumption error threshold, and optimize and iterate the energy consumption model based on the error adjustment strategy;

[0058] S64. Verify the optimized and iterated energy consumption prediction model, and update the building energy consumption database.

[0059] According to another aspect of the present invention, there is provided a building energy-saving control system based on energy consumption parameters, and the system includes:

[0060] A data acquisition and storage module, which acquires building specification parameters and real-time building energy consumption parameters, and constructs a building energy consumption database;

[0061] A data classification and extraction module, which extracts historical overall energy consumption parameters and historical block energy consumption parameters based on the building energy consumption database;

[0062] A building energy consumption model module, which constructs and trains an optimized energy consumption prediction model based on the historical overall energy consumption parameters and historical block energy consumption parameters;

[0063] An energy consumption trend prediction module, which inputs the real-time overall energy consumption parameters and real-time block energy consumption parameters into the optimized energy consumption prediction model to predict the future overall energy consumption trend and future block energy consumption trend;

[0064] The energy consumption control strategy module presets an energy consumption control database and an energy consumption matching rule, and matches the predicted future overall energy consumption trend with the future block energy consumption trend to the corresponding energy consumption control strategy according to the energy consumption matching rule;

[0065] The strategy implementation and update module executes the matched energy consumption control strategy, calculates the virtual energy consumption prediction value, compares the real-time collected power consumption with the virtual energy consumption prediction value, optimizes and iterates the energy consumption prediction model based on the comparison result, and updates the building energy consumption database.

[0066] The beneficial effects of the present invention are as follows:

[0067] 1. By using the historical energy consumption data of the building, the present invention realizes the accurate prediction of the future energy consumption trend, predicts the overall energy consumption in advance and refines it to the energy consumption distribution of each specific area, thereby providing data support for precise control, continuously comparing the real-time energy consumption data with the virtual energy consumption prediction value, and continuously optimizing and iterating the energy consumption model based on the error adjustment strategy to adapt to the changes in the environment and usage patterns. At the same time, by constructing an energy consumption control database and based on the energy consumption matching rule, the building energy consumption trend is matched with the corresponding energy consumption control strategy, providing customized energy-saving control strategies for different types of buildings, different regions and equipment, and making the system accurately consider the different impacts of time periods and equipment on energy consumption by using the building time weight and the building equipment weight.

[0068] 2. In the comparison and adjustment of the future energy consumption trend, the present invention adopts a fuzzy algorithm to process the uncertainty and volatility existing in the energy consumption prediction to solve the error problem in the energy consumption prediction, adjusts the energy consumption control strategy according to the actual situation, reduces the influence brought by the model deviation, marks the outliers, provides a basis for the subsequent model optimization, ensures that the prediction result is more reliable, and can control the energy consumption at the macro and micro levels by comprehensively considering factors such as the specifications of the building, historical energy consumption data, real-time energy consumption parameters, time and equipment weights, refine to each area and each equipment, ensure the comprehensive improvement of the building energy-saving effect, effectively avoid the excessive consumption of energy, and at the same time ensure the reasonable utilization of various energies in the building, thereby achieving the energy-saving goal.

[0069] 3. Through the real-time update and iterative optimization of the energy consumption model, the present invention enables the energy consumption model to automatically adjust according to these changes, thereby always maintaining a high energy-saving effect. At the same time, through cross-validation, performance grading and other links for the energy consumption prediction model, the stability and accuracy of the model in various environments are ensured. Moreover, the building energy consumption control method in the present invention relies on a large amount of historical energy consumption data, real-time energy consumption data, and various parameters of the building environment. Through data analysis, it can provide more scientific and accurate decision-making support for managers, help formulate more reasonable energy-saving measures, automatically generate energy consumption control strategies based on real-time data and predicted data, reduce manual intervention, improve management efficiency, and combine fuzzy algorithms with machine learning optimization to make the energy consumption management of the building more intelligent and capable of coping with complex building environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0071] Figure 1 is a flowchart of a building energy-saving control method based on energy consumption parameters according to an embodiment of the present invention;

[0072] Figure 2 is a system block diagram of a building energy-saving control system based on energy consumption parameters according to an embodiment of the present invention.

[0073] In the figure:

[0074] 1. Data acquisition and storage module; 2. Data classification and extraction module; 3. Building energy consumption model module; 4. Energy consumption trend prediction module; 5. Energy consumption control strategy module; 6. Strategy implementation and update module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0075] The following will further describe in detail the specific embodiments of the present invention in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0076] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0077] According to an embodiment of the present invention, a building energy-saving control method and system based on energy consumption parameters are provided.

[0078] The present invention will be further described in conjunction with the accompanying drawings and specific embodiments. According to an embodiment of the present invention, as Figure 1 shown, the building energy-saving control method based on energy consumption parameters according to the embodiment of the present invention includes the following steps:

[0079] S1. Obtain building specification parameters and real-time building energy consumption parameters, and construct a building energy consumption database;

[0080] Specifically, for the building specification parameters, laser scanning point cloud reconstruction is used to obtain data such as building volume, floor height, and surface area, and an infrared thermal imager is used for on-site detection to verify the heat transfer coefficient. Then, it is entered by scanning the equipment nameplate code or exported through the building automation system configuration to establish an equipment relationship diagram. At the same time, real-time energy consumption parameter collection is carried out. Time series data is collected through smart meters or branch metering devices, and heat and cold equivalent data is collected through heat meters or cold meters. The meteorological API data is connected and spatial interpolation processing is performed.

[0081] When constructing the building energy consumption database, a building information table is set up, including storing static data such as building ID, building area, and building type; a floor information table is set up, including storing data such as floor ID, floor number, and floor height; a region information table is set up, including storing data such as region ID, region volume, and usage type; a metering equipment table is set up, including storing data such as metering equipment ID, equipment type, and accuracy; a energy consumption reading table is set up, including storing dynamic data such as time stamp, energy consumption value, and unit. The database is optimized according to time series. High-frequency data is compressed in blocks using a time series database, and low-frequency data is sliced weekly using a PostgreSQL partition table. The Gorilla compression algorithm is used to process floating-point values to reduce storage space. Quality verification rules are set to ensure that the readings are within the rated value range of the equipment, and missing data is processed using an interpolation method to complete it.

[0082] S2. Based on the building energy consumption database, extract historical overall energy consumption parameters and historical block energy consumption parameters;

[0083] Specifically, for outlier detection of energy consumption data in the database, an existing density-based clustering algorithm is used to identify and remove noise data to ensure data quality. For missing value processing, the time series interpolation method is adopted, such as cubic spline interpolation, to maintain the temporal continuity of the data. Then, the energy consumption parameters are standardized to eliminate the dimension difference. When extracting the historical overall energy consumption parameters, they are divided according to the time dimension, and the energy consumption data is divided by year, quarter, and day / night mode to extract periodic features (such as the peak value of air-conditioning load in summer), and a sliding window mechanism (window length of 24 hours, step size of 1 hour) is used to analyze the hourly energy consumption trend. When extracting the historical block energy consumption parameters, the functional areas are divided according to the spatial dimension (such as office areas, equipment rooms, public areas) to define the block boundaries, and spatial clustering is performed on the area covered by the sensor network to generate an energy consumption heat map, and the energy consumption ratio of each block is calculated. At the same time, the blocks with significant differences in day and night energy consumption are identified.

[0084] S3. Based on the historical overall energy consumption parameters and historical block energy consumption parameters, construct and train an optimized energy consumption prediction model;

[0085] Specifically, the construction and training of the optimized energy consumption prediction model based on the historical overall energy consumption parameters and historical block energy consumption parameters includes the following steps:

[0086] S31. Standardize the historical overall energy consumption parameters and historical block energy consumption parameters;

[0087] Specifically, for the missing values of the historical overall energy consumption parameters and historical block energy consumption parameters, the data segments with continuous missing for more than 3 days are deleted, and the short-term missing data (such as single-point missing) is interpolated by the average value of the previous and subsequent time periods. The obvious error data (such as negative energy consumption, values exceeding the transformer capacity) is removed, and the occasional spike data is smoothed by the sliding median. The original data of different meters are unified and aligned to a granularity of 15 minutes or 1 hour.

[0088] For the standardization of the overall energy consumption parameters, different energy types such as electricity, gas, cooling, and heating are uniformly converted into standard coal equivalent (kg of standard coal) or carbon emissions (tons of CO2), and the degree-day method is used to eliminate the influence of temperature on the energy consumption of heating, ventilation, and air conditioning. The total energy consumption is converted into energy consumption per unit area (kWh / m²·day) or per capita energy consumption (kWh / person·month). If comparing data of different years, it is necessary to classify and analyze them uniformly according to weekdays or holidays.

[0089] For the standardization of the block energy consumption parameters, the energy consumption baseline is formulated separately according to the functional area type (such as office areas, data centers, commercial areas). For areas sensitive to personnel density (such as meeting rooms), the energy consumption needs to be divided by the number of people on duty in real time. At the same time, for special blocks such as computer rooms and laboratories, the energy consumption data needs to be associated with the equipment operation rate (such as server load rate), and the lighting or socket energy consumption is separately stripped and compared with the national standards of similar buildings.

[0090] S32. Extract the overall time-series characteristic values and block time-series characteristic values of the processed historical total energy consumption parameters and historical block energy consumption parameters to construct an energy consumption prediction model;

[0091] Specifically, when extracting the overall energy consumption time-series characteristics, extract the daily / weekly / monthly / annual energy consumption mean values, extreme values, and fluctuation ranges. Calculate the linear growth / decline slope of energy consumption through a sliding window (such as a 7-day window), identify the step changes in energy consumption caused by events such as holidays and equipment renovations. Then align the daily energy consumption with local temperature and humidity data, calculate the energy consumption inflection point corresponding to the air-conditioning turning-on threshold temperature, and mark the abnormal energy consumption patterns on extremely hot days (such as consecutive high-temperature days). Associate with the building operation schedule (such as 8:00-18:00 on weekdays), extract the proportion of basic energy consumption during non-working hours, and calculate the energy consumption change rate during the start and stop periods of seasonal equipment (such as winter heating boilers).

[0092] When extracting the block energy consumption time-series characteristics, calculate the position changes of high-energy-consuming blocks in different time periods (such as concentrating in the office area during the day and in the data center at night), analyze the synchrony of energy consumption fluctuations between adjacent blocks (such as the lighting energy consumption linkage between the east office area and the west), extract the energy consumption pulse signals of equipment start and stop for special functional areas (such as kitchens and machine rooms), calculate the correlation between the energy consumption of areas sensitive to personnel flow (such as elevator lobbies) and the number of access card swipes, and strip sub-item data such as lighting / plugs / hvac, extract the time difference between the morning and evening peak times of different systems, and identify the slow upward trend of the energy consumption baseline caused by equipment aging within the block.

[0093] When constructing the prediction model, splice the overall characteristics (such as monthly average energy consumption) and block characteristics (such as the proportion of air-conditioning energy consumption in the east area) into a mixed feature vector, generate virtual variable labels separately for special dates such as holidays. Use a time-series convolutional network to capture local fluctuation patterns for short-term prediction (<7 days), adopt a transformer model to handle cross-quarter dependencies for medium- and long-term prediction, and use a graph neural network to model the regional energy flow for multi-block joint prediction. During use, test the prediction error of the model during the extreme periods of summer / winter, and verify whether the prediction results of high-energy-consuming blocks are significantly better than those of low-energy-consuming blocks.

[0094] S33. Input the overall time-series characteristic values and block time-series characteristic values into the energy consumption prediction model for training, and optimize the energy consumption prediction model using the Bayesian optimization algorithm;

[0095] Specifically, the spatio-temporal features are input in layers. At the overall feature layer, building-level statistics (such as daily average total energy consumption, weekly fluctuation coefficient) are input into the fully connected layer of the model to capture the macro trends. Meanwhile, climate data (such as the moving average of temperature) is injected as the environmental context. At the block feature layer, after splicing the energy consumption sequences of each block with the functional area attributes (such as area / person flow), it is processed through an independent LSTM branch to retain the spatial specific patterns, and a shared weight sub-network is established for the blocks with equipment linkage (such as the air-conditioning main unit and the terminal).

[0096] An attention mechanism is added to the shallow layer of the model to automatically adjust the contribution weights of the two types of features. For example, during the high-temperature period in summer, the weight of the block features of the chilled water plant is increased, and during holidays, the influence of the overall historical mean features is enhanced. First, the model is pre-trained on a quarterly basis to capture the long-term trends, and then fine-tuned with two-week data to improve the short-term fluctuation response ability. For special blocks (such as data centers), double the number of training rounds is used.

[0097] When optimizing the energy consumption prediction model using the Bayesian optimization algorithm, the core optimization objectives are set, such as the average absolute percentage error of the 72-hour prediction on the validation set of the main index, and the auxiliary constraints, including that the error during the peak period (such as 14:00 - 16:00) does not exceed 8% and the model inference speed < 500 ms / run (for real-time control). Then, the parameter search space is designed, such as the model structure class including the number of long short-term memory network layers (1 - 3 layers), the number of attention heads (2 - 8), and the spatio-temporal feature fusion ratio (30% - 70%), etc., and the training control class conditions are set, such as the sliding window length (12 - 168 hours), the batch size (16 - 256), and the learning rate decay period (every 10 - 50 rounds), etc. During optimization, 20 groups of differentiated parameter combinations are generated using Latin hypercube sampling to quickly locate the potential areas. Based on the Gaussian process surrogate model, dense sampling is preferentially carried out in the neighborhood of the parameters with an average absolute percentage error < 15%, and the optimization automatically terminates when the optimization amplitude of 10 consecutive iterations < 1%.

[0098] S34. Cross-validate the optimized energy consumption prediction model and evaluate the performance parameters of the energy consumption prediction model after validation;

[0099] Specifically, cross-validation adopts time dimension validation and space dimension validation. Time dimension validation uses training, prediction, and sliding mechanisms. Initially, the first three months of data are used for training to predict the fourth month, and the window slides monthly, ultimately covering all historical data. Specific seasonal data is deliberately isolated (such as separately retaining summer data as the test set) to verify the robustness of the model during climate mutations. During space dimension validation, 15% of the block data is randomly hidden for training. During testing, the generalization ability of the model for energy consumption in unknown areas is tested by releasing the data, and particular attention is paid to the prediction error migration of blocks with similar functions (such as the error difference between the meeting rooms in Building A and Building B). A separate validation set is divided for key equipment (such as chillers) to ensure the prediction reliability of core energy-consuming units.

[0100] The average absolute percentage error of the basic accuracy index is constructed and calculated for weekdays or holidays, and peak-valley-flat electricity price periods at different times. It is required that the average absolute percentage error during peak periods ≤ 8%, and the overall average absolute percentage error ≤ 12%. The normalized average width of the prediction interval is predicted to evaluate the rationality of the confidence interval of the probability prediction result, and the ideal value should be between 0.2 - 0.35. Business adaptability indicators are set, such as the proportion of the time difference between the actual energy consumption peak and the predicted peak ≤ 30 minutes, and the deviation between the predicted energy-saving plan and the actual implementation is compared, requiring ≤ 5%. Then, Shapley additive explanations are used for the verification result analysis to analyze the dominant changes of features such as temperature characteristics on historical energy consumption characteristics at different times, identify the feature combinations that cause error mutations (such as high temperature + full-load operation), or perform K-means clustering on the prediction errors, including continuously underestimating the training data and lacking extreme working conditions and periodically overestimating without considering the equipment maintenance cycle.

[0101] S35. Preset performance grading criteria, perform performance grading on performance parameters based on the performance grading criteria, and make a usage judgment on the energy consumption prediction model according to the performance grading results. If it meets the performance grading criteria, the energy consumption prediction model is used; if it does not meet the performance grading criteria, return to S33 for re-optimization.

[0102] Specifically, when presetting the performance grading criteria, key performance parameters need to be selected first, including indicators such as average absolute percentage error, root mean square error, coefficient of determination, and peak load prediction accuracy rate. The grading thresholds are set with reference to the error tolerance of the building energy consumption model of the International Energy Agency or the historical data quantile method (such as the 25%, 50%, and 75% quantiles). At the same time, a dynamic adjustment mechanism is established to calibrate the standard regularly. After grading the model performance parameters based on this standard, if the result reaches above the standard level and the stability meets the requirements, the energy consumption prediction model is approved for use; otherwise, the S33 optimization feedback mechanism is triggered to re-optimize for problems such as feature engineering, algorithm limitations, or data quality, and iterative verification is completed within 14 days to ensure that the model performance meets the actual application requirements.

[0103] S4. Input the real-time overall energy consumption parameter and the real-time block energy consumption parameter into the optimized energy consumption prediction model to predict the future overall energy consumption trend and the future block energy consumption trend;

[0104] Specifically, the step of inputting the real-time overall energy consumption parameter and the real-time block energy consumption parameter into the optimized energy consumption prediction model to predict the future overall energy consumption trend and the future block energy consumption trend includes the following steps:

[0105] S41. Conduct data cleaning and normalization processing on the real-time overall energy consumption parameter and the real-time block energy consumption parameter;

[0106] Specifically, when conducting data cleaning on the real-time overall energy consumption parameter and the real-time block energy consumption parameter, first filter out outliers beyond the rated power range of the equipment through threshold filtering, and use the moving window mean method to correct instantaneous spike data. At the same time, for missing values caused by communication interruptions, select linear interpolation or seasonal similarity filling according to the data characteristics of adjacent time periods. After completion of cleaning, conduct normalization processing on the two types of parameters respectively. The overall energy consumption is converted into a unit energy consumption value according to the building area or production capacity scale using the industry standard benchmark method, while the block energy consumption uses min-max normalization to compress the data into the range of 0 to 1, and establish a dynamic adjustment mechanism to update the normalization benchmark in real time according to the change of the equipment load rate to ensure that the energy consumption parameters in different dimensions are comparable and retain the original data distribution characteristics.

[0107] S42. Input the processed real-time overall energy consumption parameter and the real-time block energy consumption parameter into the energy consumption prediction model for multi-modal trend prediction to obtain the future overall energy consumption trend and the future block energy consumption trend;

[0108] Specifically, when inputting the cleaned and normalized real-time overall energy consumption parameter and the block energy consumption parameter into the energy consumption prediction model, a multi-modal fusion architecture is adopted for collaborative analysis. The overall energy consumption data captures the global electricity consumption cycle law through a long short-term memory network, while the block-level data analyzes the energy consumption correlation between equipment clusters through a spatio-temporal graph convolutional network. The model dynamically weights the contribution degrees of the two types of parameters through an attention mechanism, and generates the future 24-hour overall energy consumption trend curve and the block-level energy consumption heat map respectively at the output layer. Among them, the overall prediction focuses on the fluctuation pattern of the total load, while the block prediction marks the positions and conduction paths of high-energy-consuming units. The prediction results are updated every 15 minutes, and when it is detected that the deviation between the actual value and the prediction exceeds the threshold, the model is automatically triggered for fine-tuning to ensure that the trend prediction reflects both the overall evolution law of the system and the local energy consumption characteristics.

[0109] S43. Use a fuzzy algorithm to compare the change intervals of the future overall energy consumption trend and the future block energy consumption trend;

[0110] Specifically, the step of using a fuzzy algorithm to compare the change intervals of the future overall energy consumption trend and the future block energy consumption trend includes the following steps:

[0111] S431. Extract the overall trend characteristic parameters and block trend characteristic parameters of the future overall energy consumption trend and the future block energy consumption trend, and perform data cleaning and integration;

[0112] Specifically, when extracting characteristic parameters from the future overall energy consumption trend, key indicators of the trend curve need to be captured, including the daily average load rate, peak-valley difference, proportion of continuously high-load periods, and fluctuation coefficient. At the same time, the main periodic components are decomposed through Fourier transform. For the block trend, characteristic parameters such as the maximum load density, energy consumption growth slope, and spatial correlation intensity of each unit in the heat map are extracted. After feature extraction, a cross-source verification mechanism is used to clean the data, removing outliers caused by instantaneous model errors, and aligning features with different time and space synchronizations. Finally, redundant features are compressed through principal component analysis to form a standardized feature set containing the overall trend stability index and the block trend spatial heterogeneity coefficient, providing a structured input for subsequent energy efficiency optimization decisions.

[0113] S432. Define the fuzzy membership standards for the overall trend characteristic parameters and block trend characteristic parameters, and construct a fuzzy rule base;

[0114] Specifically, when defining the fuzzy membership standards, first divide the overall trend characteristic parameters into three core dimensions: load volatility (divided into "stable, medium, severe" through the standard deviation membership function), periodic regularity (distinguishing "strong period, weak period, irregular" using the trapezoidal membership function), and trend directionality (describing the three states of "decreasing, stable, increasing" with the S-shaped membership function); for the block trend characteristic parameters, establish fuzzy standards for spatial heterogeneity (quantifying "uniform, gradient, mutation" distributions with the Gaussian membership function), load concentration (evaluating "dispersed, general, aggregated" with the triangular membership function), and conduction delay (judging "instantaneous, lagging, blocked" based on the semi-peak function).

[0115] When constructing the fuzzy rule base, a dual-driven mode of expert experience and historical data is adopted to form an if-then type rule chain. For example, "if the overall load volatility is medium and the block spatial heterogeneity is gradient, then the energy efficiency optimization priority is level two", and the rule weights are dynamically corrected through an adaptive neuro-fuzzy system. Finally, an intelligent decision-making system containing multiple core rules is formed to support multi-dimensional energy consumption trend analysis in an uncertain environment.

[0116] S433. Calculate the overlap rate of the future overall energy consumption data and the future block energy consumption data through fuzzy reasoning based on the fuzzy membership standards;

[0117] Specifically, calculating the overlap rate of the future overall energy consumption data and the future block energy consumption data through fuzzy reasoning based on the fuzzy membership standards includes the following steps:

[0118] S4331. Align time series using a preset sliding window mechanism, and normalize future total energy consumption data and future block energy consumption data;

[0119] Specifically, when using the preset sliding window mechanism, a dynamic window adjustment strategy is adopted. For future total energy consumption data, an adaptive window based on load fluctuation characteristics is set (initially defaulting to 24 hours), and it automatically expands to a 72-hour window when a seasonal pattern is detected to capture long-term patterns. For block energy consumption data, a fixed 4-hour window is used and a spatial sliding mechanism is superimposed to ensure that the energy consumption data of adjacent blocks are strictly aligned in time stamps.

[0120] After completing the window alignment, hierarchical normalization processing is performed on the two types of data. The total energy consumption data is dynamically scaled according to the industry benchmark value (using the median energy consumption of the same type of facilities in the current month as the benchmark unit), and the block-level data uses a local normalization method based on the extreme values within the sliding window, and a decay factor is introduced to exponentially reduce the weight of early data over time. The finally output normalized data retains the spatio-temporal correlation characteristics of the original sequence, while eliminating the dimensional differences between different monitoring units, providing standardized input for subsequent fusion prediction.

[0121] S4332. Define the fuzzy membership functions for future total energy consumption data and future block energy consumption data, and construct fuzzy inference rules;

[0122] Specifically, define the fuzzy membership functions. The fuzzy membership functions are used to map the energy consumption data to the membership values of a fuzzy set. For future total energy consumption and future block energy consumption, some common fuzzy sets and membership functions are defined. First, divide the total energy consumption and block energy consumption. For example, the total energy consumption is divided into three levels: low energy consumption, medium energy consumption, and high energy consumption. Low energy consumption means that the energy consumption is at a relatively low level and the system load is light. Medium energy consumption means that the energy consumption is at a medium level and the system load is normal. High energy consumption means that the energy consumption is at a relatively high level and the system load is heavy. And the block energy consumption is divided into three states: low, medium, and high, similar to the division of total energy consumption. Specifically, it can be judged according to the size and load of the block.

[0123] For each energy consumption level, design a membership function to reflect the degree to which the data point belongs to this category. And the membership function is usually continuous and can take forms such as triangle, trapezoid, normal distribution, etc. For example, for the low energy consumption membership function, the lower the energy consumption, the higher the membership. When the energy consumption is high, the membership is close to 0. For the medium energy consumption membership function, when the energy consumption is within a certain range, the membership is high and reaches the maximum when approaching this range. For the high energy consumption membership function, the higher the energy consumption, the higher the membership, and it remains unchanged after reaching the maximum value as the energy consumption increases.

[0124] Fuzzy inference rules are used to make inferences based on the input fuzzy data and obtain fuzzy results. In this context, fuzzy inference rules are used to speculate on the behavior of the system or predict the change in energy consumption based on the relationship between future total energy consumption and future block energy consumption. For example, if the total energy consumption is low and the block energy consumption is low, the future system energy consumption may be low; if the total energy consumption is medium and the block energy consumption is low, the future system energy consumption may be medium; if the total energy consumption is high and the block energy consumption is high, the future system energy consumption may be high; if the total energy consumption is medium and the block energy consumption is high, the future system energy consumption may be high. In these inference rules, the "if" part is the input condition, and the "then" part is the output result. These input fuzzy sets are mapped to a fuzzy output set through fuzzy inference, and finally a specific output value is obtained through the defuzzification process.

[0125] S4333. Calculate the overlap rate between the future total energy consumption data and the future block energy consumption data based on the fuzzy membership function and fuzzy inference rules.

[0126] Specifically, the calculation formula for the overlap rate between the future total energy consumption data and the future block energy consumption data based on the fuzzy membership function and fuzzy inference rules is as follows:

[0127] ;

[0128] where W is the overlap rate between the future total energy consumption data and the future block energy consumption data;

[0129] is the membership function of the total energy consumption data;

[0130] is the membership function of the block energy consumption data;

[0131] X is the independent variable of the energy consumption data;

[0132] d is the integration variable of the fuzzy membership function over the entire domain.

[0133] S434. Preset the overlap difference rule and the fuzzy clustering algorithm database, and match the overlap rate with the fuzzy clustering algorithm database according to the overlap difference rule;

[0134] Specifically, when presetting the overlap difference rule and the fuzzy clustering algorithm database, first establish a three-level overlap rate determination standard. Low overlap (<30%) triggers forced data re-sampling, medium overlap (30%-70%) activates the fuzzy compensation mechanism, and high overlap (>70%) directly enables the historical optimal clustering template.

[0135] The corresponding fuzzy clustering database is divided into three categories: industrial, commercial, and residential according to the energy consumption scenarios. Each category stores the typical clustering center vectors and membership matrices after feature distillation. When performing matching, a cascaded screening strategy is adopted. First, the spatio-temporal coverage similarity between the current data window and each clustering prototype is calculated through the overlapping difference rule. Then, the three most matching clustering templates are dynamically loaded according to the preset membership threshold. Finally, a weighted fuzzy C-means algorithm is used to fuse and generate a hybrid clustering scheme adapted to the current overlapping state. At the same time, the newly formed effective clustering patterns are updated as incremental data to the algorithm database, forming a closed-loop self-optimizing rule knowledge base collaborative system.

[0136] S435. Analyze the outliers of the future overall energy consumption trend and the future block energy consumption trend using a matching fuzzy clustering algorithm, and add annotations to the outliers.

[0137] Specifically, when analyzing the outliers of the future overall energy consumption trend and the block energy consumption trend, a three-level anomaly detection mechanism based on a matching fuzzy clustering algorithm is adopted. The membership of the real-time energy consumption data stream is matched with the pre-stored clustering template library, and the standardized distance of each data point relative to the nearest clustering center is calculated. For the overall trend, when the membership of three consecutive cycle data points is lower than 0.3 and the Euclidean distance exceeds twice the standard deviation, it is determined as a systematic anomaly and marked with a red alert. For block-level data, combined with spatial correlation testing, when the energy consumption deviation of a single block exceeds three times the interquartile range of the mean of adjacent blocks, it is marked as an orange spatial anomaly even if it meets the clustering membership requirements.

[0138] All annotation information is attached with four-dimensional feature fingerprints (time offset, spatial location, deviation direction, influence intensity), and is output through a visual heat map and a structured log double channel. At the same time, an adaptive clustering center adjustment mechanism is triggered to ensure that the anomaly annotation is synchronized with the model update. For complex anomalies with a duration exceeding the preset threshold, the system will automatically generate enhanced annotations including speculation on the root cause, providing hierarchical early warning support for operation and maintenance decisions.

[0139] S44. Optimize and adjust the future overall energy consumption trend and the future block energy consumption trend according to the comparison results of the change intervals, and verify the adjusted prediction results.

[0140] Specifically, when performing energy consumption optimization and adjustment based on the comparison results of the trend prediction change intervals, a global-local collaborative optimization strategy is adopted. First, a two-level feedback mechanism is established. For the overall energy consumption trend, when the predicted value exceeds the upper and lower limits of the confidence interval, the load shifting strategy is automatically triggered. Through the redistribution of peak and valley periods and the adjustment of the basic load rate, the overall curve is returned to the green operation band.

[0141] For block-level trends, a spatial compensation algorithm is adopted to perform dynamic weight rebalancing in the predicted abnormal areas, prioritize the scheduling of redundant capacity in adjacent blocks for local compensation, and evaluate the effect through a three-dimensional verification system after optimization. In the time dimension, a sliding window mean square error test is used to examine the improvement degree of prediction deviation. In the spatial dimension, the load similarity index between blocks is calculated. In the system dimension, it is verified whether the voltage volatility of key grid nodes meets the standard. Finally, an optimization report including the adjustment amplitude, energy-saving benefit, and stability coefficient is generated, and a 72-hour deduction verification is carried out through a digital twin platform to ensure that the optimization strategy meets the requirements of short-term accuracy and long-term sustainability simultaneously.

[0142] S5. Preset an energy consumption control database and energy consumption matching rules, and match the predicted future overall energy consumption trend and future block energy consumption trend with corresponding energy consumption control strategies according to the energy consumption matching rules;

[0143] Specifically, the preset energy consumption control database and energy consumption matching rules, and the matching of the predicted future overall energy consumption trend and future block energy consumption trend with corresponding energy consumption control strategies according to the energy consumption matching rules include the following steps:

[0144] S51. The preset energy consumption control database stores energy consumption control strategies and classifies them according to equipment type and energy consumption intensity;

[0145] Specifically, when constructing the energy consumption control database, a two-dimensional classification storage architecture of equipment type and energy consumption intensity is adopted. According to the international energy efficiency standard, equipment is divided into three categories: high-energy-consuming equipment (such as central air conditioners, industrial furnaces), medium-energy-consuming equipment (such as elevators, lighting systems), and low-energy-consuming equipment (such as sensors, security devices). Each category of equipment has a subclass identifier (such as cooling type / heating type). Then, for energy consumption intensity, a five-level classification is adopted, such as special level (>500 kW·h / d), first level (200 - 500 kW·h / d), second level (100 - 200 kW·h / d), third level (50 - 100 kW·h / d), fourth level (<50 kW·h / d). The strategy storage adopts a dynamic label system, and each control strategy is associated with four sets of metadata, such as applicable conditions (such as temperature threshold, load rate interval), control actions (such as variable frequency speed regulation, time period disabling), energy-saving expectation (marked percentage of baseline energy consumption reduction), and priority weight (automatically calculated according to equipment criticality). The database has a built-in self-learning index mechanism. When new equipment or energy consumption patterns change, it automatically matches similar strategy templates and generates derivative versions. At the same time, it automatically upgrades or downgrades equipment classification through monitoring of the energy consumption intensity change rate, and all strategy versions record the revision track to support comparing the control effects at different stages by time backtracking.

[0146] S52. Set energy consumption matching rules to divide energy consumption matching thresholds, and set building time weight values and building equipment weight values based on building specification parameters and real-time building energy consumption parameters;

[0147] Specifically, for the division of energy consumption matching thresholds, the benchmark threshold sets the initial energy consumption standard according to building types (such as commercial, industrial, residential). For example, the daily average energy consumption threshold per square meter of commercial buildings is set at 2.5 kW·h, that of industrial buildings is set at 4.0 kW·h, and that of residential buildings is set at 1.2 kW·h. The floating threshold combines external variables such as seasons and weather to form a dynamic adjustment range of ±20% on the basis of the benchmark value. For example, when the air-conditioning load increases in summer, the threshold of commercial buildings rises to 3.0 kW·h. When the abnormal threshold is that the real-time energy consumption exceeds 1.5 times the floating threshold, an abnormal alarm is triggered and an emergency control strategy is started. When setting the building time weight value, the peak-valley period weights are divided according to the electricity price policy into peak (weight 1.5), flat (weight 1.0), and valley (weight 0.8). For example, 8:00-12:00 on weekdays is set as the peak period, and priority regulation is carried out when the energy consumption exceeds the limit. The seasonal correction factor increases the weight by 20% in summer or winter due to temperature control requirements and decreases it by 10% in the transitional seasons (spring or autumn), and a separate weight (such as 0.7) is set for holidays to avoid misjudging the normal low-consumption state as abnormal.

[0148] For the setting of the building equipment weight value, the equipment criticality is classified. The weight of core equipment (such as the UPS in the data center) is set at 1.2, and the weight of non-critical equipment (such as landscape lighting) is set at 0.6. The energy consumption contribution correction is dynamically adjusted according to the real-time energy consumption ratio. For example, when the current energy consumption of an air-conditioning unit accounts for 30% of the total building energy consumption, its weight is automatically increased to 1.3. In the same time period of the linkage priority, the equipment with a higher weight participates in the load adjustment first (such as reducing lighting first and then air-conditioning) to ensure that key services are not affected. At the same time, the system compares the real-time energy consumption with the threshold every hour, generates a control instruction according to the time / equipment weight (such as turning off non-critical equipment with a weight of 0.6), and records the frequency of over-limit events for optimizing the threshold setting. And the system automatically analyzes the impact of building specification parameters (such as area or insulation level) on energy consumption every month, dynamically updates the weight distribution logic, and forms a closed-loop optimization mechanism.

[0149] S53. Calculate the future overall energy consumption value and the future block energy consumption value of the future overall energy consumption trend and the future block energy consumption trend according to the building time weight value and the building equipment weight value;

[0150] Specifically, the calculation of the future overall energy consumption value and the future block energy consumption value of the future overall energy consumption trend and the future block energy consumption trend according to the building time weight value and the building equipment weight value includes the following steps:

[0151] S531. Standardize the data of the building time weight value and the building equipment weight value;

[0152] S532. Construct a composite weight matrix based on the standardized building time weight value and building equipment weight value;

[0153] Specifically, for the preprocessing of weight normalization, the time weight normalization scales the building time weight values (such as 1.5 for peak, 1.0 for normal period, and 0.8 for low valley) to the range of 0 to 1 in proportion to the maximum weight (1.5). For example, the peak period is normalized to 1.0, the normal period is 0.67, and the low valley is 0.53. Then, the equipment weight is normalized. Similarly, the equipment weights (such as 1.2 for core equipment and 0.6 for non-critical equipment) are scaled to the range of 0 to 1 according to the highest value (1.2). For example, the core equipment is normalized to 1.0, and the non-critical equipment is 0.5. A spatio-temporal equipment cross matrix is constructed. The row dimension (time weight) is divided into rows according to the normalized time periods (peak, normal period, low valley), and each row represents the regulation priority for different time periods. The column dimension (equipment weight) is divided into columns according to the normalized equipment types (core or non-critical), and each column represents the energy consumption impact degree of different equipment. The composite weight value of each cell in the matrix is assigned as the product of the time weight and the equipment weight. For example, for the peak period + core equipment, 1.0 (time) × 1.0 (equipment) = 1.0 (highest regulation priority), and for the normal period + non-critical equipment: 0.67 × 0.5 = 0.34 (low priority, can be preferentially reduced). If a certain equipment frequently exceeds the limit in a specific period, the weight of its corresponding cell is automatically increased (such as +0.1), and vice versa. Set the policy mapping rules. For example, according to the matrix values, three-level policies are divided: >0.8 (red interval), immediately execute strong intervention (such as forced load reduction); 0.4 - 0.8 (yellow interval), flexible regulation (such as frequency modulation operation); <0.4 (green interval), only monitor without intervention. This matrix is updated every hour, generates dynamic regulation policies in combination with the real-time data of building energy consumption, and optimizes the weight allocation logic through historical data regression analysis to ensure that the policies accurately match the actual energy consumption scenarios.

[0154] S533. Extract the characteristic parameters of the overall energy consumption trend and the future energy consumption trend of the block, and substitute them to calculate the future overall energy consumption value and the future block energy consumption value according to the composite weight matrix;

[0155] Specifically, for the extraction of the overall energy consumption trend characteristics, the historical baseline parameters calculate the average energy consumption value, standard deviation, and daily ring ratio volatility (such as ±8%) of the same type of time period in the past 30 days as the benchmark anchor points for trend prediction. Capture the slope of the current energy consumption curve (such as rising 2 kW·h per minute), peak offset (advance / delay duration compared to the historical mean), and load mutation points (such as a sudden increase of 10% within 5 minutes) for short-term trend fine-tuning. Integrate environmental parameters such as temperature and humidity (such as a 1°C increase in temperature resulting in a +5% increase in air-conditioning energy consumption) and real-time feedback data on building occupancy density. For the extraction of the energy consumption trend characteristics of each block, spatial differentiation parameters are calculated. The energy consumption proportion (such as the data center accounting for 40% of the total) and the energy consumption intensity per unit area (such as 0.8 kW·h / m² in the office area) are separately counted for each functional area (such as the office area, data center, lighting area). Equipment cluster characteristics: Identify highly correlated equipment groups within the same block (such as the coordinated energy consumption of air conditioners and fresh air systems), and record their associated energy consumption fluctuation range (such as ±15%). Time period sensitivity markers, for example, the data center still maintains a 90% load at night, while the office area only has a 10% load at night.

[0156] With the intervention of the composite weight matrix, the overall prediction adds the historical baseline parameter (such as 1200 kW·h) to the real-time slope correction value (+2 kW·h / min × 30 min = +60 kW·h), and then multiplies it by the time weight of the current time period (such as 1.0 during peak hours) to output a 1-hour future prediction value of 1260 kW·h. For the office area, use its energy consumption intensity (0.8 kW·h / m²) × area × equipment weight (0.5), combined with the time weight during off-peak hours (0.67), to calculate the 30-minute future prediction value as 0.335 of the total energy consumption of the area. Compare the prediction value with the actual value every 15 minutes. If the error exceeds 5%, trigger the rebalancing of the weight matrix (such as increasing the equipment weight near the mutation point by 10%). The system generates a 24-hour energy consumption heat map, marks high, medium, and low-risk blocks, and automatically pushes control suggestions (such as "the data center may exceed the threshold in the next 2 hours, it is recommended to activate the standby cooling"), and at the same time feedback the prediction results to the energy management platform for global optimization.

[0157] S534. Verify and adjust the future overall energy consumption value and the future energy consumption value of each block, and output the verified and adjusted future overall energy consumption value and the future energy consumption value of each block.

[0158] Specifically, for real-time data comparison and error analysis, during the overall energy consumption verification, the predicted overall energy consumption value (e.g., 1260 kW·h) is compared with the actual monitored value (e.g., 1280 kW·h) every hour, and the absolute error (20 kW·h) and relative error (1.6%) are calculated. If the error exceeds the preset threshold (e.g., 5%), the adjustment mechanism is triggered. When conducting block energy consumption verification: Compare the predicted values (e.g., the predicted value for the office area is 300 kW·h and the actual value is 310 kW·h) separately by functional area (such as the office area, data center), identify abnormal blocks (such as the data center exceeding the expectation by more than 5%), and then attribute the error and correct the weight. For example, for time weight adjustment, if there is a continuously high error during a certain period (such as peak hours), recalibrate its time weight (such as from 1.0 to 0.95) to reduce the prediction dependence during that period. For equipment weight adjustment, for equipment clusters that frequently exceed expectations (such as the air-conditioning group), increase their equipment weights in the composite weight matrix (such as from 0.5 to 0.6) to enhance their influencing factors. For external factor compensation, if the error is caused by a sudden change in environmental temperature, introduce a temperature compensation coefficient (such as +3% energy consumption / °C) to correct subsequent predictions. For short-term correction (at the 15-minute level), based on the latest actual data, use the sliding window method to recalculate the energy consumption for the next hour. For example, if the original prediction for the office area in the next 30 minutes was 150 kW·h and the actual measurement was 160 kW·h, then adjust the subsequent prediction upward to 170 kW·h according to the error ratio. For long-term correction (at the 24-hour level), combine the historical error distribution (such as the prediction deviation is smaller at night), automatically reduce the prediction weight for low-confidence periods (such as early morning), and increase the prediction proportion for high-confidence periods (such as during the day on weekdays). When outputting the overall energy consumption, mark the key risk points (such as "the predicted value from 14:00 to 16:00 is reduced by 2% compared to the initial value") on the corrected future 24-hour curve, and attach a confidence rating (such as A / B / C level). When outputting the block energy consumption, generate a dynamic heat map by functional area, mark the over-limit blocks in red (such as the predicted value for the data center at 14:00 is +8%), and push the control strategy (such as "it is recommended to start the standby power supply 10 minutes in advance"), and record the adjustment logic each time (such as "increase the air-conditioning weight due to temperature mutation") for subsequent training of machine learning models (such as LSTM) to continuously optimize the prediction algorithm.

[0159] S54. Match the future overall energy consumption value and the future block energy consumption value with the energy consumption matching threshold, and obtain the overall energy consumption control strategy and the block energy consumption control strategy based on the matching result;

[0160] Specifically, for setting the energy consumption matching threshold, the overall energy consumption threshold is based on the historical peak value (such as 1500 kW·h) and the grid carrying capacity. The safety threshold (1300 kW·h), warning threshold (1200 kW·h), and optimization threshold (1000 kW·h) are set, and the energy consumption risk levels (red, yellow, green) are divided. The energy consumption threshold of each block is dynamically set according to the functional area. For example, the single-hour threshold of the data center is 400 kW·h (when exceeded, the refrigeration backup is triggered), and the lighting threshold of the office area is 80 kW·h (when exceeded, the intelligent dimming is started). When matching the predicted value with the threshold, if the predicted value for the next 1 hour (such as 1250 kW·h) exceeds the warning threshold (1200 kW·h) but is lower than the safety threshold (1300 kW·h), it is marked as a yellow warning, and intervention is required but no emergency response is needed. When matching for each block, if the predicted value of the data center (such as 420 kW·h) exceeds the single-hour threshold (400 kW·h), it is marked as a red overlimit, and the load reduction strategy needs to be immediately executed. A hierarchical response mechanism is set. In the green interval (<1000 kW·h), normal operation is maintained, and renewable energy (such as photovoltaic energy storage) is preferentially used. For the yellow warning (1000 - 1200 kW·h), demand-side management is started, such as delaying the start time of non-critical equipment and raising the air-conditioning temperature setting by 1℃. For the red overlimit (>1300 kW·h), forced load reduction is carried out, non-essential loads are cut off (such as stopping some elevators), and a diesel generator is enabled to supplement the power supply. Global optimization combines the time-of-use electricity price policy (such as peak-valley electricity price), and automatically switches to energy storage power supply during peak hours to reduce the dependence on the grid.

[0161] The energy consumption control strategy for each block generates high-energy-consuming blocks (such as data centers). The short-term strategy is to enable standby refrigeration units and migrate some computing tasks to low-load periods. The long-term strategy is to optimize the server cluster scheduling algorithm to reduce the energy use efficiency value below 1.2. For low-priority blocks (such as office areas), lighting control is based on occupancy sensors to dynamically turn off the lighting in unoccupied areas, and the brightness is reduced to 70%. For air-conditioning control: the temperature floating range is set for each area (such as 26±2℃) to avoid redundant energy consumption due to overcooling / overheating. The execution effect of the strategy is verified every 5 minutes (such as the actual energy consumption of the data center drops to 390 kW·h). If the expected value is not reached, the measures are upgraded (such as further restricting the server performance). The strategy iteration records the effectiveness of the strategy (such as "raising the air-conditioning temperature by 1℃ saves 5% of the energy consumption"), and updates it to the strategy knowledge base for the AI model (such as reinforcement learning) to optimize future decisions.

[0162] S55. Analyze the conflict parameters of the overall energy consumption control strategy and the energy consumption control strategy for each block, optimize the overall energy consumption control strategy and the energy consumption control strategy for each block according to the conflict parameters, and verify and output the optimized overall energy consumption control strategy and the energy consumption control strategy for each block.

[0163] Specifically, the conflict parameters identify and analyze global and local target conflicts. The overall strategy (such as reducing the total energy consumption by 10%) requires shutting down some equipment, but the block strategy (such as the data center needs to maintain 99.9% availability) prohibits forced power-off, resulting in execution contradictions. The time priority conflict is that the overall strategy requires load reduction during peak electricity consumption (such as 14:00), while the block strategy (such as laboratory equipment) needs to maintain high-energy consumption operation due to the continuity of experiments. The resource allocation conflict is that the overall strategy relies on the energy storage system to supply power first, but the block strategy (such as the medical area) requires the energy storage to be used as an emergency backup, leading to resource contention.

[0164] When optimizing the conflict parameters, dynamically adjust the weights. Assign higher priority weights to the conflicting blocks (such as the data center), allowing it to exempt some strategies during the overall load reduction (such as only reducing the load of non-core servers). Automatically increase the overall strategy weight during non-critical periods (such as at night) to enforce the unified implementation of energy consumption reduction measures. The hierarchical compromise mechanism: If the overall strategy and the block strategy cannot be fully compatible, adopt a compromise plan (such as the data center reduces the load by 5% instead of the overall required 10%), and increase the energy consumption reduction efforts in other blocks (such as the office area) for compensation. Execute the strategy at off-peak times. For example, reduce the energy consumption of the office area 15 minutes before the overall load reduction period to reserve a buffer for high-priority blocks (such as the production workshop).

[0165] Verify and output the optimized strategy. When conducting simulation verification, simulate the optimized strategy in the digital twin system, check whether the overall energy consumption meets the standard (such as a total energy consumption reduction of 8.5%) and whether the block requirements (such as the availability of the data center) are not damaged. Implement the optimized strategy in stages. For example, only adjust the lighting strategy of the office area on the first day, and superimpose the data center load reduction on the next day. Monitor in real time whether the conflict is alleviated (such as the total energy consumption drops by 7% and there is no block alarm).

[0166] S6. Execute the matched energy consumption control strategy, calculate the virtual energy consumption prediction value, compare the real-time collected power consumption with the virtual energy consumption prediction value, optimize and iterate the energy consumption prediction model based on the comparison result, and update the building energy consumption database.

[0167] Specifically, the steps of the execution of the matched energy consumption control strategy, calculating the virtual energy consumption prediction value, comparing the real-time collected power consumption with the virtual energy consumption prediction value, optimizing and iterating the energy consumption prediction model based on the comparison result, and updating the building energy consumption database include the following:

[0168] S61. Extract the control strategy parameters in the implementation of the energy consumption control strategy, and substitute the control strategy parameters into the energy consumption prediction model to calculate the virtual energy consumption prediction value;

[0169] Specifically, when presetting the framework of the energy consumption control strategy, a dynamic parameter hierarchical extraction mechanism is adopted. For the overall control target (such as the power saving rate at the park level), an adaptive parameter group based on load characteristics is set (initially default including three core parameters: time period division, load reduction ratio, and equipment priority, etc.). When the energy consumption mode mutates (such as the peak season of production), the parameter expansion mechanism is automatically triggered to append long-term control rules (such as continuous multi-day stepped energy consumption reduction). For the block-level strategy (such as workshops or floors), a fixed parameter template is used to overlay the space linkage rules to ensure that the control thresholds of adjacent areas (such as temperature set values, lighting brightness) are dynamically synchronized within the time window.

[0170] After the parameter extraction is completed, hierarchical mapping processing is performed on the strategy parameters. The overall-level parameters are converted into weight coefficients recognizable by the model according to the industry standard library (such as mapping "load reduction by 15% during peak hours" to the energy consumption reduction factor for the corresponding time period). The block-level parameters generate local control instructions through extreme value constraints within the sliding window (such as converting "the upper limit of air conditioner temperature is 26°C" into the piecewise correction value of the temperature-energy consumption relationship curve), and an aging attenuation factor is introduced to gradually reduce the influence weight of the early strategy as the prediction time progresses. The finally generated parameter set retains the spatio-temporal constraint logic of the original strategy and eliminates the instruction conflicts between different control levels, providing standardized control input for the virtual energy consumption prediction model.

[0171] In the model calculation stage, the mapped strategy parameters are embedded in a two-layer prediction architecture. The overall layer loads long-term strategies (such as 72-hour seasonal temperature adjustment rules) through an adaptive window to drive the global deformation of the baseline energy consumption curve. The block layer relies on the sliding alignment mechanism of the fixed window to convert the space linkage parameters (such as synchronous switching of lighting between adjacent floors) into local energy consumption correction amounts. When the model outputs, the strategy intervention nodes are marked (such as the energy consumption depression during the load reduction period), and the deviation between the virtual prediction value and the strategy target is verified through the reverse verification module (such as whether the actual power saving rate matches the preset 15%). If the deviation exceeds the threshold, the parameter feedback adjustment is automatically triggered (such as shortening the load reduction period or relaxing the temperature tolerance). The finally generated virtual energy consumption curve synchronously outputs the strategy impact assessment report, quantifying the actual contribution degree of each parameter (such as 60% of the power saved by time period control and 30% by temperature adjustment), providing a data closed-loop for strategy iteration.

[0172] S62. Preset the energy consumption error threshold and error adjustment strategy, and match the result of comparing the real-time collected power consumption with the virtual energy consumption prediction value with the energy consumption error threshold;

[0173] Specifically, a hierarchical error threshold mechanism is adopted. For overall energy consumption data (such as parks or building complexes), a static error threshold is set based on the historical prediction deviation distribution (the initial default is ±10%), and a dynamic floating interval is superimposed (such as relaxed to ±15% during peak load periods). For block-level data (such as production lines or floors), a sliding window extreme value method is used to automatically generate local thresholds (the 90% quantile of the prediction error under the same working conditions in the past 7 days is taken). At the same time, spatial coupling rules are introduced to ensure that the thresholds of adjacent blocks are adjusted in a time-series linkage (for example, when the error of workshop A exceeds the limit, the threshold check of workshop B is triggered synchronously). When the real-time collected energy consumption is compared with the virtual prediction value, execution The three-level error matching strategy, such as instantaneous error detection, immediately marks it as an abnormal point and triggers a real-time alarm if the current data point deviation exceeds the threshold, and freezes the prediction model output to avoid error diffusion. The window cumulative verification calculates the average deviation with a rolling window of 1 hour. If the cumulative error of three consecutive windows exceeds 80% of the threshold, it is judged as a systematic deviation, and the hot update of model parameters (such as refitting the load curve coefficient) is started. Long-term backtracking, daily summary of error data, and seasonal exponential smoothing method are used to correct the threshold baseline (such as the overall increase of 2% in the summer cooling season threshold), and generate an error tracing report (distinguishing between equipment failure, strategy failure or external interference factors).

[0174] For random errors (such as single instrument noise), the data stream is directly smoothed through sliding mean filtering. For trend errors (such as continuous high deviations), the control strategy parameter library is linked to automatically reduce the load reduction ratio or relax the temperature control constraints until the real-time data returns to the threshold range. All adjustment operations are recorded in the error decision log, including the threshold trigger time, adjustment range and correction effect, to support subsequent strategy optimization and model retraining.

[0175] S63, selecting an error adjustment strategy based on the energy consumption error threshold matching result, and optimizing and iterating the energy consumption model based on the error adjustment strategy;

[0176] Specifically, the error threshold matching result will trigger the hierarchical strategy selection engine, which automatically adapts and adjusts the solution according to the deviation type and duration. When there is an instantaneous overlimit (short-term random error), the data cleaning strategy is adopted to replace the abnormal points with the moving window mean or perform spatial interpolation repair based on the data of adjacent blocks to avoid noise interfering with the model input. Meanwhile, short-term cache isolation is enabled to temporarily store the data during the abnormal period in the verification queue. If the data returns to normal in the subsequent 3 consecutive cycles, it will be released and marked as an occasional event. When there is a continuous overlimit (systematic deviation), if the error accumulates beyond the threshold and lasts for more than 2 hours, it is determined that the model is mismatched, triggering dynamic parameter recalibration. By using a rolling time window (such as 24 hours), the relationship curve of the load environment variables is refitted. First, the weight coefficients of sensitive factors such as temperature and humidity are adjusted preferentially, and then based on spatial correlation analysis, the energy consumption characteristics of blocks under the same working conditions are clustered and compensated (for example, when the error of production line A exceeds the limit, the similar working condition parameters of production line B are borrowed for temporary coverage). For the failure of seasonal patterns (such as predicting winter data in summer), the long-term pattern library of the same historical period is forcibly called and injected into the model input layer for trend compensation.

[0177] When the number of daily error alarms exceeds 5 times, the strategy backtracking evaluation is started. By comparing the difference distribution between the real-time data and the virtual prediction, the overconstrained points of the control strategy are identified (such as the actual energy consumption rebounds due to the too low temperature set value), and the strategy relaxation suggestions are automatically generated (such as adjusting the upper limit of the air conditioner temperature from 26°C to 27°C), which are pushed to the strategy execution end after manual confirmation.

[0178] S64. Verify the optimized and iterated energy consumption prediction model and update the building energy consumption database.

[0179] According to another embodiment of the present invention, as Figure 2 shown, a building energy-saving control system based on energy consumption parameters is provided, and the system includes:

[0180] The data acquisition and storage module 1 acquires the building specification parameters and real-time building energy consumption parameters and constructs a building energy consumption database;

[0181] The data classification and extraction module 2 extracts the historical overall energy consumption parameters and historical block energy consumption parameters based on the building energy consumption database;

[0182] The building energy consumption model module 3 constructs and trains an optimized energy consumption prediction model based on the historical overall energy consumption parameters and historical block energy consumption parameters;

[0183] The energy consumption trend prediction module 4 inputs the real-time overall energy consumption parameters and real-time block energy consumption parameters into the optimized energy consumption prediction model to predict the future overall energy consumption trend and future block energy consumption trend;

[0184] The energy consumption control strategy module 5 presets an energy consumption control database and an energy consumption matching rule, and matches the predicted future overall energy consumption trend with the future block energy consumption trend according to the energy consumption matching rule to obtain the corresponding energy consumption control strategy;

[0185] The strategy implementation and update module 6 executes the matched energy consumption control strategy, calculates the virtual energy consumption prediction value, compares the real-time collected power consumption with the virtual energy consumption prediction value, optimizes and iterates the energy consumption prediction model based on the comparison result, and updates the building energy consumption database.

[0186] In summary, by means of the above technical solutions of the present invention, the present invention realizes the accurate prediction of the future energy consumption trend by using the historical energy consumption data of the building, predicts the overall energy consumption in advance and details the energy consumption distribution in each specific area, so as to provide data support for precise control, continuously compares the real-time energy consumption data with the virtual energy consumption prediction value, and continuously optimizes and iterates the energy consumption model based on the error adjustment strategy, so as to adapt to the changes in the environment and usage patterns. At the same time, by constructing an energy consumption control database and based on the energy consumption matching rule, the building energy consumption trend is matched with the corresponding energy consumption control strategy, providing customized energy-saving control strategies for different types of buildings, different regions and equipment, and using the building time weight and building equipment weight to make the system accurately consider the different impacts of time periods and equipment on energy consumption.

[0187] In addition, in the comparison and adjustment of the future energy consumption trend, the present invention adopts a fuzzy algorithm to process the uncertainty and volatility in energy consumption prediction to solve the error problem in energy consumption prediction, adjusts the energy consumption control strategy according to the actual situation, reduces the impact caused by model deviation, marks the outliers, provides a basis for subsequent model optimization, ensures that the prediction result is more reliable, and can perform energy consumption control at the macro and micro levels by comprehensively considering factors such as the specifications of the building, historical energy consumption data, real-time energy consumption parameters, time and equipment weights, details to each area and each equipment, ensures the comprehensive improvement of the building energy-saving effect, effectively avoids excessive energy consumption, and at the same time ensures the reasonable utilization of various energies in the building, so as to achieve the energy-saving goal.

[0188] In addition, through the real-time update and iterative optimization of the energy consumption model, the present invention enables the energy consumption model to automatically adjust according to these changes, thereby always maintaining a high energy-saving effect. At the same time, through cross-validation, performance grading and other links of the energy consumption prediction model, the stability and accuracy of the model in various environments are ensured. Moreover, the building energy consumption control method in the present invention relies on a large amount of historical energy consumption data, real-time energy consumption data and various parameters of the building environment. Through data analysis, it can provide more scientific and accurate decision-making support for managers, help formulate more reasonable energy-saving measures, automatically generate energy consumption control strategies according to real-time data and predicted data, reduce manual intervention, improve management efficiency, and combine fuzzy algorithms with machine learning optimization to make the energy consumption management of the building more intelligent and capable of coping with complex building environments.

[0189] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A building energy-saving control method based on energy consumption parameters, characterized in that, It includes the following steps: S1. Obtain building specification parameters and real-time building energy consumption parameters, and construct a building energy consumption database; S2. Based on the building energy consumption database, extract historical overall energy consumption parameters and historical block energy consumption parameters; S3. Based on the historical overall energy consumption parameters and historical block energy consumption parameters, construct and train an optimized energy consumption prediction model; S4. Input the real-time overall energy consumption parameters and real-time block energy consumption parameters into the optimized energy consumption prediction model to predict the future overall energy consumption trend and future block energy consumption trend, including: S41. Perform data cleaning and normalization processing on the real-time overall energy consumption parameters and real-time block energy consumption parameters; S42. Input the processed real-time overall energy consumption parameters and real-time block energy consumption parameters into the energy consumption prediction model for multi-modal trend prediction to obtain the future overall energy consumption trend and future block energy consumption trend; S43. Use a fuzzy algorithm to compare the change intervals of the future overall energy consumption trend and future block energy consumption trend, specifically including: Define the fuzzy membership degree standard for the overall and block energy consumption trend characteristics, and construct a fuzzy rule base; Based on the fuzzy membership degree and inference mechanism, calculate the overlap rate of the future overall and block energy consumption trends; Preset an overlap difference rule and a fuzzy clustering algorithm database, and match the overlap rate with the fuzzy clustering algorithm database according to the overlap difference rule; Use the matching fuzzy clustering algorithm to analyze the outliers of the future overall and future block energy consumption trends and add annotations; S44. According to the change interval comparison result, optimize and adjust the future overall energy consumption trend and future block energy consumption trend, and verify the adjusted prediction result; S5. Preset an energy consumption control database and an energy consumption matching rule, and match the predicted future overall energy consumption trend and future block energy consumption trend with corresponding energy consumption control strategies, including: S51. Preset the energy consumption control database to store energy consumption control strategies and classify them according to equipment type and energy consumption intensity; S52. Set the building time weight value and building equipment weight value based on the building specification parameters and real-time building energy consumption parameters, specifically including: Construct a composite weight matrix based on the building time weight value and building equipment weight value; Extract the characteristic parameters of the future overall and future block energy consumption trends, and substitute them into the composite weight matrix to calculate the future overall and future block energy consumption values; Verify and adjust the output of the future overall and future block energy consumption values; S54. Set an energy consumption matching threshold, match the future overall and future block energy consumption values with the energy consumption matching threshold, and obtain the overall energy consumption and block energy consumption control strategies based on the matching result; S55. Analyze the conflict parameters of the overall energy consumption and block energy consumption control strategies, and optimize and output the overall energy consumption and block energy consumption control strategies according to the conflict parameters; S6. Execute the matched energy consumption control strategy, calculate the virtual energy consumption prediction value, compare the real-time collected electricity energy consumption with the virtual energy consumption prediction value, and optimize and iterate the energy consumption prediction model based on the comparison result, and update the building energy consumption database.

2. The building energy-saving control method based on energy consumption parameters according to claim 1, wherein, The constructing and training an optimized energy consumption prediction model based on the historical overall energy consumption parameters and historical block energy consumption parameters includes the following steps: S31. Perform standardization processing on the historical overall energy consumption parameters and historical block energy consumption parameters; S32. Extract the overall time-series characteristic values and block time-series characteristic values of the processed historical overall energy consumption parameters and historical block energy consumption parameters to construct an energy consumption prediction model; S33. Input the overall time-series characteristic values and block time-series characteristic values into the energy consumption prediction model for training, and optimize the energy consumption prediction model using the Bayesian optimization algorithm; S34. Conduct cross-validation on the optimized energy consumption prediction model and evaluate the performance parameters of the verified energy consumption prediction model; S35. Preset a performance grading standard, grade the performance parameters based on the performance grading standard, and make a usage judgment on the energy consumption prediction model according to the performance grading result. If it meets the performance grading standard, use the energy consumption prediction model; if it does not meet the performance grading standard, return to S33 for re-optimization.

3. The building energy-saving control method based on energy consumption parameters according to claim 1, characterized in that The calculation of the overlap rate of future overall energy consumption data and future block energy consumption data through fuzzy inference based on the fuzzy membership degree standard includes the following steps: S4331. Preset a sliding window mechanism to align time series, and perform normalization processing on future overall energy consumption data and future block energy consumption data; S4332. Define the fuzzy membership degree functions of future overall energy consumption data and future block energy consumption data, and construct fuzzy inference rules; S4333. Calculate the overlap rate of future overall energy consumption data and future block energy consumption data based on the fuzzy membership degree functions and fuzzy inference rules.

4. The building energy-saving control method based on energy consumption parameters according to claim 3, characterized in that, The calculation formula for calculating the overlap rate of future overall energy consumption data and future block energy consumption data based on the fuzzy membership degree functions and fuzzy inference rules is: ； Where W is the overlap rate of future overall energy consumption data and future block energy consumption data; is the membership function of the overall energy consumption data; is the membership function of the energy consumption data of the block; X is the independent variable of the energy consumption data; d is the integral variable of the fuzzy membership degree function over the entire domain.

5. A building energy-saving control method based on energy consumption parameters according to claim 1, characterized in that The execution of the matched energy consumption control strategy, calculation of the virtual energy consumption prediction value, comparison of the real-time collected power consumption energy with the virtual energy consumption prediction value, optimization and iteration of the energy consumption prediction model based on the comparison result, and update of the building energy consumption database include the following steps: S61. Extract the control strategy parameters in the implemented energy consumption control strategy, and input the control strategy parameters into the energy consumption prediction model to calculate the virtual energy consumption prediction value; S62. Preset an energy consumption error threshold and an error adjustment strategy, and match the result of comparing the real-time collected power consumption energy with the virtual energy consumption prediction value with the energy consumption error threshold; S63. Select an error adjustment strategy based on the matching result of the energy consumption error threshold, and optimize and iterate the energy consumption model based on the error adjustment strategy; S64. Verify the optimized and iterated energy consumption prediction model and update the building energy consumption database.

6. An energy-saving control system for buildings based on energy consumption parameters is used to implement the energy-saving control method for buildings based on energy consumption parameters described in any one of claims 1-5, characterized in that, The system includes: A data acquisition and storage module, which acquires building specification parameters and real-time building energy consumption parameters and constructs a building energy consumption database; A data classification and extraction module, which extracts historical overall energy consumption parameters and historical block energy consumption parameters based on the building energy consumption database; A building energy consumption model module, which constructs, trains, and optimizes an energy consumption prediction model based on the historical overall energy consumption parameters and historical block energy consumption parameters; An energy consumption trend prediction module, which inputs the real-time overall energy consumption parameters and real-time block energy consumption parameters into the optimized energy consumption prediction model to predict the future overall energy consumption trend and future block energy consumption trend; The energy consumption control strategy module presets an energy consumption control database and an energy consumption matching rule, and matches the predicted future overall energy consumption trend with the future block energy consumption trend according to the energy consumption matching rule to the corresponding energy consumption control strategy; The strategy implementation and update module executes the matched energy consumption control strategy, calculates the virtual energy consumption prediction value, compares the real-time collected power consumption with the virtual energy consumption prediction value, optimizes and iterates the energy consumption prediction model based on the comparison result, and updates the building energy consumption database.

Images