Multi-mode public building air conditioning system flexible load prediction method based on parameter configuration

Abstract

The invention discloses a parameter configuration-based multi-mode public building air conditioning system flexible load prediction method, which comprises the following steps of: S1, acquiring historical energy consumption data and indoor and outdoor environment factor data of a public building air conditioning system, and preprocessing the historical energy consumption data and the indoor and outdoor environment factor data; s2, constructing an energy consumption prediction model based on a long short-term memory (LSTM) network, and training to generate a preliminary energy consumption prediction value; s3, injecting configuration parameters through an external parameter interface, and dynamically adjusting input characteristics or loss weight of the energy consumption prediction model so as to switch among three flexible load prediction modes, namely an adjustable flexible load prediction mode, a transferable flexible load prediction mode and an interruptible flexible load prediction mode; s4, on the basis of an indoor temperature prediction sub-model of the GRU, predicting indoor temperature change for the environment sequence in the interruptible mode, and verifying a prediction result according to a preset comfort interval; and S5, based on historical data, selecting a historical day similar to the prediction day in feature as a similar day, calculating deviation and constructing a correction model, and correcting the prediction value after mode switching to output a final load prediction value.

Description

Technical Field

[0001] This invention belongs to the field of air conditioning load forecasting and energy management technology, specifically involving a flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration. It is applicable to flexible load forecasting, optimized scheduling, and demand response of air conditioning systems in public buildings (such as office buildings, shopping malls, libraries, etc.). Background Technology

[0002] In public buildings such as office buildings, shopping malls, and libraries, air conditioning systems typically account for 40% to 60% of total energy consumption, making them a key challenge in urban energy management. With the increasing penetration of renewable energy and the development of smart grids, the flexible load of air conditioning systems—the adjustable, transferable, and interruptible portion that can be controlled—shows great potential in demand response and peak-valley regulation. However, existing air conditioning load forecasting technologies face multiple limitations: traditional physical models rely on building thermodynamic parameters and meteorological data, but struggle to capture nonlinear coupling effects, time-series dependencies, and the impact of sudden events, leading to a significant decrease in forecast accuracy in dynamic environments; data-driven methods, while incorporating machine learning to process historical data, are often limited to single load types and cannot flexibly adapt to different scenarios, such as temperature fine-tuning in adjustable modes, cold storage shifts in transferable modes, or short-term shutdowns in interruptible modes; furthermore, these methods are prone to errors when dealing with high-dimensional environmental factors (such as indoor and outdoor temperature and humidity, solar radiation, occupancy rates, and grid signals). Current technologies suffer from overfitting or underfitting issues, especially under conditions of high data noise or imbalanced samples, where prediction errors can exceed 15%. Furthermore, current techniques neglect verification of indoor comfort after interrupted regulation, potentially leading to decreased user experience and energy waste. Insufficient real-time rolling prediction capabilities prevent precise intraday scheduling, further hindering the integrated application of air conditioning systems in comprehensive energy systems. Overall, these problems increase peak grid pressure, reduce energy efficiency, and impede the achievement of carbon reduction targets. Therefore, a high-precision prediction framework integrating multiple modes and considering comfort constraints is urgently needed to optimize energy management and grid interaction in public buildings. Summary of the Invention

[0003] This invention aims to address the problems of single air conditioning load forecasting modes, neglect of comfort, and poor flexibility in existing technologies, and provides a flexible load forecasting method for public building air conditioning systems based on parameter configuration and multiple modes. This method uses LSTM as its core, parameterizes three modes, and integrates a GRU temperature sub-model to ensure accurate forecasting and user comfort.

[0004] The objective of this invention is achieved through the following technical solution: A flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration includes: S1. Data Acquisition and Preprocessing: Collect historical energy consumption data and indoor and outdoor environmental factor data of public building air conditioning systems, and preprocess the collected data; S2. Energy consumption prediction model construction: Construct an energy consumption prediction model based on the Long Short-Term Memory (LSTM) network, train it using preprocessed data, and generate preliminary energy consumption prediction values; S3. External parameter configuration and mode switching: Configuration parameters are injected through the external parameter interface, and the input features or loss weights of the energy consumption prediction model are dynamically adjusted according to the configuration parameters to switch between three flexible load prediction modes: adjustable, transferable, and interruptible. S4. Indoor temperature prediction and comfort verification: Based on the indoor temperature prediction sub-model of the gated cycle unit GRU, the indoor temperature change is predicted for the environmental sequence in the interruptible mode in step S3, and the prediction results are verified according to the preset comfort range. S5. Similar Day Correction and Deviation Optimization: Based on historical data, select historical days with similar characteristics to the predicted day as similar days, calculate the deviation and build a correction model to correct the predicted value after mode switching to output the final load prediction value.

[0005] Furthermore, in step S1, the historical energy consumption data includes hourly total energy consumption, baseline load, and flexible portion sequence; indoor and outdoor environmental factor data includes outdoor temperature, indoor temperature, relative humidity, solar radiation intensity, occupancy rate, holiday indicators, grid demand response signals, and personnel density. The preprocessing includes: removing impulse noise using a median filtering algorithm; reducing high-frequency sampled data to one sampling point every 15 minutes; using the Pearson correlation coefficient to filter factors with a correlation threshold greater than 0.5; and using the Min-Max normalization formula to scale the feature values ​​of all data to the [0,1] interval.

[0006] Furthermore, the energy consumption prediction model adopts a two-layer LSTM structure, with 128 hidden units in the first layer and 64 hidden units in the second layer, and uses a unidirectional structure; the energy consumption prediction model is trained using the Adam optimizer, combined with a quantization loss function. The formula is:

[0007] in, This is the quantile parameter, which defaults to 0.3 to capture the lower quantile deviation. i is the sample index. This is the actual value. These are predicted values.

[0008] Furthermore, in step S3, the parameter configuration logic for the three flexible load forecasting modes is as follows: Adjustable mode: Configuration parameters include temperature float threshold, cycle time, and rolling step size; the energy consumption prediction model calculates load change according to the formula: Adjustable load = Baseline load - (Temperature offset × Sensitivity coefficient); where, the adjustable load represents the load value in adjustable mode; the baseline load represents the reference load when not adjusted; the temperature offset represents the temperature adjustment amount; and the sensitivity coefficient represents the load sensitivity to temperature changes, calculated through regression of historical data. Transferable mode: Configuration parameters include cold storage ratio, time offset, and storage efficiency; the energy consumption prediction model integrates energy storage logic: Transfer load = storage capacity × efficiency - release delay, where transfer load represents the load value in transferable mode; storage capacity represents cold storage capacity; efficiency represents storage efficiency; release delay represents the delay loss during release; interruptible mode predicts zero energy consumption after shutdown; Interruptible mode: Configuration parameters include interrupt duration, recovery threshold, and safety buffer; in this mode, the predicted energy consumption after shutdown is zero. The configuration parameters are dynamically injected via JSON files or API interfaces, allowing users to customize them to suit different building types.

[0009] Furthermore, step S4 includes: S401. For the operation state in interruptible mode in step S3, obtain the corresponding post-interruption environment sequence, the environment sequence including indoor and outdoor temperature and humidity data within the prediction step; S402. Input the environmental sequence into a pre-trained indoor temperature prediction sub-model based on a gated recurrent unit (GRU), and output the indoor temperature sequence for a future preset time period; S403. Compare the indoor temperature sequence with the preset comfort range. If the predicted temperature exceeds the comfort range, trigger the feedback mechanism and return to step S3 to adjust the interruption duration parameter in this mode until the generated indoor temperature sequence meets the comfort requirements.

[0010] Furthermore, hidden units 128-64 are set up and trained using the MSE loss function, outputting indoor temperature sequences at 15-minute intervals; the comfort range is 23-26℃, and the feedback mechanism includes triggering an alarm or automatically adjusting the interruption duration. When the interruption duration is automatically adjusted, the system returns to step S3 to adjust the interruption duration parameter in the interruption mode.

[0011] Furthermore, partial data of the predicted day is collected, and the root mean square error (RMSE) between the predicted day and each segmented subset of historical days in the historical dataset is calculated. The historical day with the smallest RMSE is selected as the similar day. Constructing a correction model , where t represents the time variable; Represents the historical load function for similar days; This represents the overall deviation; and the prediction time... satisfy ;in This indicates the latest data collection time; k represents the loop count coefficient, ranging from 3 to 7.

[0012] Preferably, the present invention also provides a flexible load forecasting device for a multi-mode public building air conditioning system based on parameter configuration, comprising: The data acquisition and preprocessing unit is used to collect historical energy consumption data and indoor and outdoor environmental factor data of the air conditioning system of public buildings, and to preprocess the collected data. The energy consumption prediction model building unit is used to build an energy consumption prediction model based on the Long Short-Term Memory (LSTM) network. It is trained using preprocessed data to generate preliminary energy consumption prediction values. An external parameter configuration and mode switching unit is used to inject configuration parameters through an external parameter interface and dynamically adjust the input features or loss weights of the energy consumption prediction model according to the configuration parameters, so as to switch between three flexible load prediction modes: adjustable, transferable, and interruptible. The indoor temperature prediction and comfort verification unit is used to predict indoor temperature changes for the environmental sequence in the interruptible mode in step S3 based on the indoor temperature prediction sub-model of the gated cycle unit GRU, and to verify the prediction results according to the preset comfort range. The similar day correction and deviation optimization unit is used to select historical days with similar characteristics to the predicted day based on historical data, calculate the deviation and build a correction model to correct the predicted value after mode switching and output the final load prediction value.

[0013] Preferably, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the flexible load prediction method for the multi-mode public building air conditioning system based on parameter configuration.

[0014] Preferably, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration.

[0015] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: 1. This invention, through an external parameter configuration mechanism, achieves seamless switching between three load types: adjustable, transferable, and interruptible. This effectively solves the problem of single-mode operation in existing technologies, improving the model's adaptability and flexibility by over 30%. In diverse scenarios of public buildings, such as peak occupancy periods in office buildings and peak weekend traffic in shopping malls, parameters can be dynamically adjusted, avoiding model retraining and saving 50% of computing resources. Simultaneously, this parameter configuration mechanism, as a core innovation, avoids the black-box problem of deep learning models, ensuring full transparency. Combined with physical models, it better handles nonlinear relationships, exhibiting significantly stronger robustness than purely data-driven methods. This fills the patent gap in multi-mode and comfort verification, enhancing the overall technological advancement.

[0016] 2. Significantly improves prediction accuracy, robustness, and computational efficiency: This invention integrates the time-series processing capabilities of LSTM and the quantile optimization of Quantile Loss, achieving a MAE lower than 8% of the high energy consumption of central air conditioning, an RMSE lower than 10%, and a R... 2 With a prediction performance greater than 0.85, CV-RMSE less than 25%, and NMBE less than 10%, it surpasses traditional GBDT or SVM methods by 15%-20%. Under noisy data or extreme weather conditions, the similarity day correction mechanism further reduces the error by 5%-10%, ensuring prediction stability. In addition, the LSTM-GRU combination reduces the number of parameters by 20%, Adam optimization accelerates training by 30%, supports rolling prediction with a step size of 1-4 hours, is suitable for edge device deployment, and comprehensively optimizes computational and operational efficiency.

[0017] 3. Comfort Guarantee and User Experience Optimization: This invention features an original integrated GRU indoor temperature sub-model that verifies the temperature sequence in interruptible mode in real time, avoiding deviations of more than 1°C within the comfort range (23-26°C) and reducing user complaint rates by 20%. Compared to technologies without verification, this model can automatically adjust the interruption duration, balancing energy conservation and humanistic care, ensuring that the user experience is not affected, and improving the sustainability of energy management in public buildings.

[0018] 4. Practicality, Economic and Environmental Benefits: This invention is applicable to various public buildings, such as office buildings and shopping malls, supports grid demand response, and has great potential for carbon emission reduction; the parametric design facilitates integration with BMS or IoT systems, making it easy to commercialize and promote; economically, it reduces peak-valley difference by 10%-20%, reduces investment in reserve capacity, and tests show that the energy-saving cost payback period is shortened to 2-3 years in office building applications; environmentally, it promotes the integration of renewable energy, contributes to dual carbon targets, and achieves a comprehensive improvement in energy utilization efficiency.

[0019] 5. Diverse data sources (BMS / EMS / smart meters / weather stations); employing median filtering to remove impulse noise, high-frequency data reduction to 15-minute sampling, Pearson correlation screening, linear interpolation to fill missing values, and Min-Max normalization to [0,1], among other data acquisition and preprocessing methods. Noise reduction and normalization reduce high-frequency interference and lower training time (estimated to reduce training time by 20–30%); correlation screening removes redundant features, reduces dimensionality, mitigates overfitting, and improves generalization; normalization accelerates gradient descent convergence, improves training stability, and thus improves prediction accuracy (indirectly supporting an increase in MAE / RMSE metrics); addressing issues such as excessive noise, missing values, and redundant variables in the original data leading to model overfitting, slow convergence, and unstable predictions.

[0020] 6. Collect partial observations for the forecast date (n=12 points), and a historical date set of m=30 days. Select similar days based on segmented RMSE and construct a bias correction function Q(t), iteratively correcting k times (k=3–7). This addresses the problem of systematic bias in single-model predictions caused by sudden weather, events, or anomalies. Matching similar days for correction reduces the impact of noise and extreme events, improving prediction robustness (reducing error by 5–10%). Using recent observations for online correction enhances short-term prediction stability. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the principle of the method of the present invention.

[0022] Figure 2 This is a flowchart illustrating the external parameter configuration and mode switching process.

[0023] Figure 3 This is a flowchart illustrating the similarity date correction and deviation optimization process. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0025] The technical solution of the present invention will be clearly and completely described and explained below with reference to the accompanying drawings of the embodiments of this application, but it is not intended to limit the scope of protection of this application.

[0026] Example 1 This embodiment provides a flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration. (See...) Figure 1The system comprises five parts: data acquisition and preprocessing, construction of a unified LSTM prediction framework, configuration of external parameters and mode switching, indoor temperature prediction and comfort verification, and correction and deviation optimization based on similar days. After preprocessing historical energy consumption data and environmental factors of public building air conditioning systems, the system performs initial prediction using a unified LSTM framework, implements mode switching through parameter configuration, verifies comfort using a temperature sub-model, and optimizes deviations based on similar days. These five parts progressively achieve multi-mode prediction of public building air conditioning systems based on flexible loads. Specifically: S1. Data Acquisition and Preprocessing: Collect total load data from the energy management system (EMS) or smart meters installed in the air conditioning systems of public buildings (such as office buildings or shopping malls). The data includes hourly total energy consumption (kW / h), baseline load and flexible load sequence, as well as indoor and outdoor environmental factors, such as outdoor temperature (range -10℃ to 40℃), indoor temperature (initial setting 23-26℃), relative humidity (20%-80%), and solar radiation intensity (0-1000 W / m²). 2 ), occupancy rate (0-100%), holiday indicators (binary variables), power grid demand response signals (peak and valley time indicators), and personnel density (people / m²). 2 Due to equipment malfunction or environmental interference, the raw data contains noise and outliers, which can lead to overfitting or decreased accuracy if used directly for model training. This section preprocesses the raw data sequence. First, median filtering is used to remove significant impulse noise. Then, to reduce computational complexity and dimensionality, high-frequency sampling data is reduced to one sampling point every 15 minutes while preserving the data's trend, generating a smooth data sequence suitable for subsequent prediction. Furthermore, Pearson correlation coefficient analysis is used to screen factors with a correlation threshold >0.5 and remove redundant variables. Missing values ​​are filled using linear interpolation, and Min-Max normalization is applied. This method scales all feature values ​​to the [0,1] interval, improving model convergence speed. Here, x' represents the normalized value; x represents the original value; min(x) represents the minimum feature value; and max(x) represents the maximum feature value. Through data reduction, this method effectively improves subsequent computational efficiency, reducing training time by 20%-30% compared to unprocessed data. This part effectively filters noise and optimizes the data structure, thus providing a high-quality input foundation for the prediction framework.

[0027] S2. Construction of a Unified LSTM Prediction Framework: An energy consumption prediction model based on Long Short-Term Memory (LSTM) networks is constructed. This model performs preliminary predictions on the preprocessed total load energy consumption sequence to capture long-range dependencies in time-series data. The energy consumption prediction model uses a two-layer LSTM structure: the first layer has 128 hidden units, and the second layer has 64, employing a unidirectional structure to reduce computational complexity. Each LSTM layer includes an input gate, a forget gate, and an output gate, as shown in the following formulas: Forgotten Gate: ;in, Forgetting threshold, It is the Sigmoid activation function. Here is the forget gate weight matrix. The hidden state of the previous time step. Enter the current time step. Offset the forget gate.

[0028] Input Gate: , ;in, For the input gate value, The input gate weight matrix, To offset the forget gate, Candidate cell state, It is the hyperbolic tangent function. This is the candidate cell state weight matrix. Bias for candidate cell states.

[0029] Cell status update: ;in, The current cell state, For element-wise multiplication, This represents the cell state at the previous time step.

[0030] Output gate: , ;in, To output the gate value, This is the output gate weight matrix. For output gate bias, It is currently hidden.

[0031] The input dimension is the selected number of factors (approximately 8-10), and the output is the predicted energy consumption value for the next time period; training uses the Adam optimizer (initial learning rate 0.001, beta1=0.9, beta2=0.999), combined with the quantile loss function: ,in, This is the quantile parameter, which defaults to 0.3 to capture the lower quantile deviation. i is the sample index. This is the actual value. The value is the predicted value; the batch size is 32, the training iterations are 200, and early stopping is supported (patience threshold 10); to prevent gradient explosion, the gradient is normalized using Clipnorm=1.0.

[0032] In this invention, the energy consumption prediction model does not use the mean squared error (MSE) as the loss function during training. Instead, a quantile loss function is introduced. This is because air conditioning load prediction in public buildings exhibits significant uncertainty and asymmetric risk characteristics under flexible control scenarios. For flexible load modes such as adjustable or interruptible loads, the impact of load prediction deviations on system operation is asymmetric in different directions: when the load is underestimated, it may lead to insufficient control or exceed indoor comfort limits; while when the load is overestimated, it mainly affects the accuracy of flexible potential assessment. The mean squared error loss function symmetrically penalizes positive and negative prediction errors, making it difficult to reflect the aforementioned differentiated risk characteristics. The quantile loss function, by introducing a quantile parameter τ, allows the model to be optimized for specific quantiles of the load distribution. When τ is less than 0.5, the model focuses more on the conservative lower bound of load prediction during training, which helps reduce the comfort or operational safety risks caused by prediction deviations during flexible load release. Therefore, adopting the quantile loss function enables the prediction model to better adapt to the application requirements of public building air conditioning systems in demand response and flexible load control scenarios, improves the robustness and engineering applicability of the model under uncertain conditions, and thus enhances the overall technical solution and practical value of the present invention.

[0033] Implemented in PyTorch and supported by GPU acceleration, this model can detect and predict the periodicity and trend changes in load. The algorithm effectively handles non-stationary time-series data, improving prediction accuracy by 15%-20% compared to traditional ARIMA or SVM methods, thus providing reliable preliminary predictions for subsequent mode switching. This lays a solid foundation for multi-mode flexible load forecasting.

[0034] S3. External Parameter Configuration and Mode Switching: For the initial forecast values ​​generated by the energy consumption prediction model, external parameter configuration enables switching between three flexible load prediction modes: adjustable, transferable, and interruptible. External parameters include: in adjustable mode, the temperature rise threshold (1-2℃), cycle time (15min), and rolling step size (1h / 2h / 3h / 4h); in transferable mode, the cold storage ratio (0-1), time offset (1h), and storage efficiency (0.8-0.95); and in interruptible mode, the interruption duration (1h), recovery threshold (temperature recovery rate), and safety buffer (5min). Parameters are dynamically injected via JSON files or API interfaces, supporting user customization to adapt to different building types.

[0035] Specifically, the algorithm modifies the input features or loss weights of the LSTM energy consumption prediction model through parameter injection. When switching to adjustable mode, it predicts energy consumption changes under temperature fine-tuning, using the formula: Adjustable Load = Baseline Load - (Temperature Offset × Sensitivity Coefficient), where the adjustable load represents the load value in adjustable mode; the baseline load represents the reference load when not adjusted; the temperature offset represents the temperature adjustment amount; and the sensitivity coefficient represents the load sensitivity to temperature changes, calculated through regression analysis of historical data. The transferable mode integrates energy storage logic: Transfer Load = Storage Capacity × Efficiency - Release Delay, where the transfer load represents the load value in transferable mode; the storage capacity represents the cold storage capacity; the efficiency represents the storage efficiency; and the release delay represents the delay loss during release. The interruptible mode predicts zero energy consumption after shutdown, but requires monitoring of peak rebound during the recovery period. This method, through parameterization, effectively overcomes the limitations caused by a single mode and significantly improves the model's flexibility. Partial processes based on external parameter configuration and mode switching are as follows: Figure 2 As shown, this section can seamlessly switch modes and record relevant predicted values ​​for subsequent verification and optimization, thereby improving the overall adaptability of the forecast. Compared to traditional single-mode forecasting methods, this section has higher scenario applicability and can effectively handle different flexible load types, thus improving the accuracy of the forecast. This provides a good data foundation for comfort verification.

[0036] S4. Indoor Temperature Prediction and Comfort Verification: This section addresses the interruptible mode by integrating a Gated Recurrent Unit (GRU) network as the indoor temperature prediction sub-model to simplify the LSTM gate structure and improve computational efficiency. The indoor temperature prediction sub-model has 128-64 hidden units and is trained using the MSE loss function, similar to the LSTM training process. The GRU formula is as follows: Update Gate: ;in, To update the gate value, To update the gate weight matrix; Reset Door: ;in, To reset the gate value, To reset the gate weight matrix; Candidate hidden state: ;in, In the candidate hidden state, The candidate hidden state weight matrix; Hidden state: ; It is currently hidden.

[0037] For the interruptible operation mode in step S3, the corresponding post-interruption environmental sequence is obtained. This sequence includes indoor and outdoor temperature and humidity data within the prediction step size. The input to the indoor temperature prediction sub-model is the corresponding post-interruption environmental sequence, and the output is the indoor temperature sequence (step size 15 minutes). The comfort range is constrained to 23-26℃. If the prediction exceeds this range, an alarm is triggered or the process returns to step S3 to automatically adjust the interruption duration. Specifically, the algorithm calculates the temperature rise after the interruption to determine if it falls within the comfort range. If the rise exceeds 2℃, the interruption is shortened to 45 minutes. This indoor temperature prediction sub-model effectively avoids a decline in user experience by introducing comfort verification. This part can accurately predict temperature changes and provides safety assurance for the interruption mode. Compared to prediction methods without verification, it improves the comfort rate by more than 95%, thus completing a reliable assessment of flexible loads. Experimental verification shows that the average prediction accuracy of this sub-model is better than 92%.

[0038] To further illustrate the applicability and practical effect of the method of the present invention in different types of public buildings, the application of the flexible load prediction mode is explained below in two typical scenarios: office buildings and commercial buildings.

[0039] (1) Application instructions for adjustable mode in office building scenario In a typical office building scenario, the building's primary function is daytime office work, resulting in relatively stable occupancy rates and a highly predictable operating pattern for the air conditioning system. After training an energy consumption prediction model based on the building's historical energy consumption and environmental data, the "adjustable mode" described in this invention is activated without altering the air conditioning system's operating time, and the indoor air conditioning set temperature is uniformly increased by 1°C from the original setting.

[0040] Comparative analysis results show that, under the above parameter configuration conditions, the average absolute error between the predicted load and the actual operating load of the air conditioning system remains at a low level, and the prediction results can accurately reflect the load change trend. At the same time, compared with the baseline operating condition without temperature regulation, the overall energy consumption level of the building air conditioning system shows a significant decrease.

[0041] The results show that in public buildings with stable occupancy characteristics, such as office buildings, the method of the present invention can achieve effective flexible load release through small temperature adjustments while ensuring prediction accuracy, and has good energy-saving potential and engineering feasibility.

[0042] (2) Application instructions for interruptible mode in commercial building scenarios In commercial building scenarios, due to significant changes in occupancy density over time, air conditioning load is highly sensitive to adjustments in operating strategies. For such buildings, when the "interruptible mode" is enabled, candidate interruption duration schemes are first generated based on external parameter configurations, and the indoor temperature prediction sub-model of the gated cyclic unit (GRU) is used to predict the indoor temperature changes corresponding to the interruption scheme in advance.

[0043] Under a set of typical parameter configurations, the temperature prediction results show that if the initial interruption duration parameter is followed, the indoor temperature at the end of the interruptible period tends to approach or exceed the upper limit of the comfort range (26℃). In response to this prediction result, the system automatically adjusts the interruption duration parameter according to the feedback mechanism described in step S403 and re-verifies the temperature prediction.

[0044] After parameter correction, the predicted indoor temperature sequence can be stably maintained within the preset comfort range, thereby avoiding comfort risks that may be caused by improper interruption parameter settings.

[0045] This application description shows that, in scenarios with large load fluctuations, such as commercial buildings, the method of this invention can reasonably constrain the operation scheme in interruptible mode through a closed-loop mechanism of "prediction-verification-parameter correction," thereby improving the safety and reliability of flexible load applications.

[0046] S5. Similarity Day Correction and Bias Optimization: For predicted values ​​after mode switching, this section extracts similar days from historical data for bias correction to improve robustness. Similarity day correction is set to collect partial data for the predicted day (n=12 points, 1-hour interval), read the historical dataset (m=30 days), and calculate the root mean square error (RMSE) of the segmented subsets: Select the historical date with the lowest RMSE as the similar date; construct a modified model. ,in For overall deviation; prediction time satisfy ;in The last data collection time is represented by ; k represents the loop count coefficient, ranging from 3 to 7. These correction parameters together constitute an optimization vector that can adapt to dynamic changes. This method effectively reduces the impact of noise through similar day matching. Part of the process based on similar day correction and deviation optimization is as follows: Figure 3As shown in the figure, this section can distinguish between different environmental fluctuations and provide accurate bias correction for the final output. Practical testing has verified that the average prediction efficiency of this method is better than 90%.

[0047] Example 2 Based on the same inventive concept, this application also provides a flexible load forecasting device for a multi-mode public building air conditioning system based on parameter configuration, which can be used to implement the method described in the above embodiments, as described in the following embodiments. Since the principle of the flexible load forecasting device for a multi-mode public building air conditioning system based on parameter configuration is similar to that of the flexible load forecasting method for a multi-mode public building air conditioning system based on parameter configuration, the implementation of this device can refer to the aforementioned method implementation, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0048] The embodiments of the present invention provide a specific implementation of a forecasting device capable of realizing a flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration, specifically including the following: The data acquisition and preprocessing unit is used to collect historical energy consumption data and indoor and outdoor environmental factor data of the air conditioning system of public buildings, and to preprocess the collected data. The energy consumption prediction model building unit is used to build an energy consumption prediction model based on the Long Short-Term Memory (LSTM) network. It is trained using preprocessed data to generate preliminary energy consumption prediction values. An external parameter configuration and mode switching unit is used to inject configuration parameters through an external parameter interface and dynamically adjust the input features or loss weights of the energy consumption prediction model according to the configuration parameters, so as to switch between three flexible load prediction modes: adjustable, transferable, and interruptible. The indoor temperature prediction and comfort verification unit is used to predict indoor temperature changes for the environmental sequence in the interruptible mode in step S3 based on the indoor temperature prediction sub-model of the gated cycle unit GRU, and to verify the prediction results according to the preset comfort range. The similar day correction and deviation optimization unit is used to select historical days with similar characteristics to the predicted day based on historical data, calculate the deviation and build a correction model to correct the predicted value after mode switching and output the final load prediction value.

[0049] Preferably, embodiments of this application also provide a specific implementation of an electronic device capable of implementing all steps in the parameter-configured flexible load forecasting method for multi-mode public building air conditioning systems described in the above embodiments. The electronic device specifically includes the following components: Processor, memory, communications interface, and bus; The processor, memory, and communication interface communicate with each other via a bus; the communication interface is used to realize information transmission between server-side devices, metering devices, and user-side devices.

[0050] The processor is used to call the computer program in memory. When the processor executes the computer program, it implements all the steps in the flexible load prediction method for multi-mode public building air conditioning system based on parameter configuration in the above embodiments.

[0051] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the parameter-configurable multi-mode public building air conditioning system flexible load forecasting method in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the parameter-configurable multi-mode public building air conditioning system flexible load forecasting method in the above embodiments.

[0052] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, hardware + program embodiments are relatively simple in description because they are fundamentally similar to method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0053] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0054] While this application provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed in the order shown in the embodiments or drawings or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0055] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0056] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0057] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0058] This invention is not limited to the embodiments described above. The above description of specific embodiments is intended to illustrate and explain the technical solutions of this invention. The specific embodiments described above are merely illustrative and not restrictive. Without departing from the spirit and scope of the claims, those skilled in the art can make many specific modifications based on the teachings of this invention, and these modifications all fall within the scope of protection of this invention.

Claims

1. A flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration, characterized in that, include: S1. Data Acquisition and Preprocessing: Collect historical energy consumption data and indoor and outdoor environmental factor data of public building air conditioning systems, and preprocess the collected data; S2. Energy consumption prediction model construction: Construct an energy consumption prediction model based on the Long Short-Term Memory (LSTM) network, train it using preprocessed data, and generate preliminary energy consumption prediction values; S3. External parameter configuration and mode switching: Configuration parameters are injected through the external parameter interface, and the input features or loss weights of the energy consumption prediction model are dynamically adjusted according to the configuration parameters to switch between three flexible load prediction modes: adjustable, transferable, and interruptible. S4. Indoor temperature prediction and comfort verification: Based on the indoor temperature prediction sub-model of the gated cycle unit GRU, the indoor temperature change is predicted for the environmental sequence in the interruptible mode in step S3, and the prediction results are verified according to the preset comfort range. S5. Similar Day Correction and Deviation Optimization: Based on historical data, select historical days with similar characteristics to the predicted day as similar days, calculate the deviation and build a correction model to correct the predicted value after mode switching to output the final load prediction value.

2. The flexible load forecasting method for multi-mode public building air conditioning systems according to claim 1, characterized in that, In step S1, the historical energy consumption data includes total energy consumption at the hourly level, baseline load, and flexible portion sequence; indoor and outdoor environmental factor data includes outdoor temperature, indoor temperature, relative humidity, solar radiation intensity, occupancy rate, holiday indicators, grid demand response signals, and personnel density. The preprocessing includes: removing impulse noise using a median filtering algorithm; reducing high-frequency sampled data to one sampling point every 15 minutes; using the Pearson correlation coefficient to filter factors with a correlation threshold greater than 0.5; and using the Min-Max normalization formula to scale the feature values ​​of all data to the [0,1] interval.

3. The flexible load forecasting method for multi-mode public building air conditioning systems according to claim 1, characterized in that, The energy consumption prediction model employs a two-layer LSTM structure, with 128 hidden units in the first layer and 64 hidden units in the second layer, and uses a unidirectional structure. The model is trained using the Adam optimizer, combined with a quantization loss function. The formula is: in, This is the quantile parameter, which defaults to 0.3 to capture the lower quantile deviation. i is the sample index. This is the actual value. These are predicted values.

4. The flexible load forecasting method for multi-mode public building air conditioning systems according to claim 1, characterized in that, In step S3, the parameter configuration logic for the three flexible load forecasting modes is as follows: Adjustable mode: Configuration parameters include temperature float threshold, cycle time, and rolling step size; the energy consumption prediction model calculates load change according to the formula: Adjustable load = Baseline load - (Temperature offset × Sensitivity coefficient); where, the adjustable load represents the load value in adjustable mode; the baseline load represents the reference load when not adjusted; the temperature offset represents the temperature adjustment amount; and the sensitivity coefficient represents the load sensitivity to temperature changes, calculated through regression of historical data. Transferable mode: Configuration parameters include cold storage ratio, time offset, and storage efficiency; the energy consumption prediction model integrates energy storage logic: Transfer load = storage capacity × efficiency - release delay, where transfer load represents the load value in transferable mode; storage capacity represents cold storage capacity; efficiency represents storage efficiency; release delay represents the delay loss during release; interruptible mode predicts zero energy consumption after shutdown; Interruptible mode: Configuration parameters include interrupt duration, recovery threshold, and safety buffer; in this mode, the predicted energy consumption after shutdown is zero. The configuration parameters are dynamically injected via JSON files or API interfaces, allowing users to customize them to suit different building types.

5. The flexible load forecasting method for multi-mode public building air conditioning systems according to claim 1, characterized in that, Step S4 includes: S401. For the operation state in interruptible mode in step S3, obtain the corresponding post-interruption environment sequence, the environment sequence including indoor and outdoor temperature and humidity data within the prediction step; S402. Input the environmental sequence into a pre-trained indoor temperature prediction sub-model based on a gated recurrent unit (GRU), and output the indoor temperature sequence for a future preset time period; S403. Compare the indoor temperature sequence with the preset comfort range. If the predicted temperature exceeds the comfort range, trigger the feedback mechanism and return to step S3 to adjust the interruption duration parameter in this mode until the generated indoor temperature sequence meets the comfort requirements.

6. The flexible load forecasting method for multi-mode public building air conditioning systems according to claim 5, characterized in that, Hidden units 128-64 are set up and trained using the MSE loss function. The indoor temperature sequence is output at 15-minute intervals. The comfort range is 23-26℃. The feedback mechanism includes triggering an alarm or automatically adjusting the interruption duration. When the feedback mechanism is to automatically adjust the interruption duration, it returns to step S3 to adjust the interruption duration parameter in the interruption mode.

7. The flexible load forecasting method for multi-mode public building air conditioning systems according to claim 1, characterized in that, Collect partial data of the predicted day, calculate the root mean square error (RMSE) between it and each segmented subset of historical days in the historical dataset, and select the historical day with the smallest RMSE as the similar day. Constructing a correction model , where t represents the time variable; Represents the historical load function for similar days; This represents the overall deviation; and the prediction time... satisfy ;in This indicates the latest data collection time; k represents the loop count coefficient, ranging from 3 to 7.

8. A flexible load forecasting device for multi-mode public building air conditioning systems based on parameter configuration, characterized in that, include: The data acquisition and preprocessing unit is used to collect historical energy consumption data and indoor and outdoor environmental factor data of the air conditioning system of public buildings, and to preprocess the collected data. The energy consumption prediction model building unit is used to build an energy consumption prediction model based on the Long Short-Term Memory (LSTM) network. It is trained using preprocessed data to generate preliminary energy consumption prediction values. An external parameter configuration and mode switching unit is used to inject configuration parameters through an external parameter interface and dynamically adjust the input features or loss weights of the energy consumption prediction model according to the configuration parameters, so as to switch between three flexible load prediction modes: adjustable, transferable, and interruptible. The indoor temperature prediction and comfort verification unit is used to predict indoor temperature changes for the environmental sequence in the interruptible mode in step S3 based on the indoor temperature prediction sub-model of the gated cycle unit GRU, and to verify the prediction results according to the preset comfort range. The similar day correction and deviation optimization unit is used to select historical days with similar characteristics to the predicted day based on historical data, calculate the deviation and build a correction model to correct the predicted value after mode switching and output the final load prediction value.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the flexible load prediction method for multi-mode public building air conditioning systems based on parameter configuration as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the flexible load forecasting method for multi-mode public building air conditioning systems based on parameter configuration as described in any one of claims 1 to 7.

Images