Air conditioner load demand response potential evaluation method based on Internet of Things

Abstract

The invention discloses an air conditioner load demand response potential evaluation method based on the Internet of Things, relates to the field of air conditioner load response analysis, and aims to avoid the uncomfortable problem of completely identifying all physical parameters in a traditional first-order equivalent thermal parameter model. The operation state of the air conditioner is decomposed into a steady-state interval and a dynamic interval, and key macroscopic characteristic parameters, namely a steady-state thermal coefficient, a power consumption boundary function and a dynamic time constant, capable of describing system boundaries and inertia are directly identified from data. According to the method, coupling identification of a plurality of internal states and parameters such as indoor temperature, equivalent thermal resistance and equivalent thermal capacity is ingeniously avoided, finally, in a non-intrusive scene only depending on user aggregation power utilization and outdoor temperature data, the macroscopic parameters capable of being accurately identified are utilized to evaluate response potential, and the response potential is evaluated. The accuracy and robustness of air conditioner load demand response potential evaluation are realized.

Description

Technical Field

[0001] This application relates to the field of air conditioning load response analysis, and more specifically, to an Internet of Things-based method for assessing the potential of air conditioning load demand response. Background Technology

[0002] With the rapid development of smart grids, demand response on the grid side to maintain system stability and economic operation has become crucial. Air conditioning loads, as major energy-consuming devices in buildings and a typical temperature-controlled load, have highly concentrated power consumption in both time and space, possessing enormous demand response potential and making them an ideal regulating resource for peak shaving and valley filling in the power grid. Therefore, how to utilize Internet of Things (IoT) technology to automatically, accurately, and cost-effectively assess the demand response potential of widely distributed air conditioning loads has become a key issue that power companies, load aggregators, and other market players urgently need to address. This is of great significance for improving grid flexibility and promoting energy efficiency.

[0003] Currently, the main technical approach to assessing air conditioning load response potential relies on constructing a building heat exchange model. As a temperature-controlled load, the building's heat exchange determines the operating power consumption of the air conditioning load, a process related to factors such as room size and wall insulation. Some studies have constructed coupled thermal inertia models of the heat exchange between various parts of the air conditioning load system and the building to assess air conditioning load response potential. Among these, the first-order equivalent thermal parameter (ETP) model is widely used due to its low complexity. However, existing technologies have significant drawbacks when applying this model. One approach requires intrusive monitoring of the independent operating status of the air conditioner and recording indoor temperature changes, which severely infringes on user privacy and lacks feasibility for large-scale practical application. In recent years, non-intrusive load monitoring technology has flourished, utilizing aggregated energy consumption data collected by user smart meters to clarify the operating status information of major internal energy-consuming equipment. Based on this, some studies have attempted to identify ETP model parameters using only aggregated user electricity consumption data and publicly available outdoor temperature data. However, these methods greedily attempt to identify all parameters of the model, including equivalent thermal resistance, equivalent heat capacity, indoor temperature, and air conditioning energy efficiency coefficient.

[0004] However, in non-intrusive scenarios with only limited input information such as aggregated power and outdoor temperature, the sheer number of unknown parameters to be identified leads to a severe technical bottleneck: redundancy and ill-posedness in parameter identification. Because the input information is insufficient to constrain the model's degrees of freedom, multiple distinct combinations of physical parameters can produce identical power consumption curves, making the model indistinguishable and resulting in non-unique and poorly robust identification results. This redundant parameter introduction, aimed at identifying a perfect physical model, actually significantly reduces the accuracy of identifying critical states such as indoor temperature, ultimately severely impacting the accuracy and reliability of demand response potential assessment and hindering the large-scale commercial application of this technology. Summary of the Invention

[0005] To address the challenges of current technologies, according to one aspect of this application, a method for assessing air conditioning load demand response potential based on the Internet of Things is provided, comprising:

[0006] Acquire the raw aggregated power data stream and the raw outdoor temperature data stream;

[0007] Data preprocessing and air conditioning load decomposition are performed on the raw aggregated power data stream and the raw outdoor temperature data stream to obtain the air conditioning power sequence and the outdoor temperature sequence.

[0008] The air conditioning power sequence and outdoor temperature sequence are divided into operating state intervals to obtain a set of steady-state data points and a set of dynamic data segments.

[0009] Steady-state parameters and dynamic boundaries are identified from the set of steady-state data points to obtain the steady-state thermodynamic coefficient, upper boundary function of power consumption, and lower boundary function of power consumption.

[0010] Dynamic time constants are obtained by identifying dynamic time constants in a set of dynamic data segments.

[0011] Based on the steady-state thermodynamic coefficient, the upper boundary function of power consumption, the lower boundary function of power consumption, and the dynamic time constant, a comprehensive assessment of demand response is conducted on the current air conditioning power and the current outdoor temperature to obtain the potential for reduction, the potential for increase, and the response duration.

[0012] Compared with existing technologies, this application provides an IoT-based method for assessing the demand response potential of air conditioning loads, which avoids the ill-posed problem of completely identifying all physical parameters in traditional first-order equivalent thermal parameter models. To address this, a new data-driven paradigm for modeling the macroscopic behavior of air conditioning loads is proposed. This paradigm decomposes the air conditioning operating state into steady-state and dynamic intervals and directly identifies key macroscopic characteristic parameters describing the system boundary and inertia from the data, namely the steady-state thermodynamic coefficient, power consumption boundary function, and dynamic time constant. This method cleverly bypasses the coupled identification of multiple internal states and parameters such as indoor temperature, equivalent thermal resistance, and equivalent heat capacity, thus fundamentally solving the ill-posed problem caused by parameter redundancy in the prior art. Finally, in a non-intrusive scenario relying solely on aggregated user electricity consumption and outdoor temperature data, these precisely identifiable macroscopic parameters are used to assess the response potential, achieving both accuracy and robustness in assessing the demand response potential of air conditioning loads. Attached Figure Description

[0013] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings.

[0014] Figure 1 This is a flowchart of an IoT-based method for assessing air conditioning load demand response potential according to an embodiment of this application.

[0015] Figure 2 This is a data flow flowchart of an IoT-based method for assessing air conditioning load demand response potential according to an embodiment of this application.

[0016] Figure 3 This is a flowchart of step 5 in the IoT-based air conditioning load demand response potential assessment method according to an embodiment of this application.

[0017] Figure 4 This is a flowchart of step 6 in the IoT-based air conditioning load demand response potential assessment method according to an embodiment of this application. Detailed Implementation

[0018] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. It should be understood that the drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0019] With the development of smart grids, accurate and non-intrusive demand response potential assessment of air conditioning loads is crucial. Current technologies primarily rely on first-order equivalent thermal parameter (ETP) models, as detailed below:

[0020]

[0021] in, , These are the indoor and outdoor temperatures, respectively. , These represent the building's equivalent thermal resistance and equivalent heat capacity, respectively. The energy efficiency coefficient of an air conditioner. This refers to the air conditioning load power. For the power of other indoor heat sources, The power of the air conditioner before it participates in demand response. The indoor temperature before air conditioning participates in demand response. Set the indoor temperature after the air conditioner participates in demand response. Objective: Identify The goal of traditional methods is to identify the power change during the steady-state operation of an air conditioner. This involves solving a complex kinetic equation to calculate the response duration Δt. However, solving this equation heavily relies on the precise identification of a series of microscopic physical parameters, including the building's equivalent thermal resistance, equivalent heat capacity, and the air conditioner's energy efficiency coefficient. More importantly, it requires real-time acquisition or estimation of two variables that are extremely difficult to obtain in non-intrusive scenarios: the time-varying indoor temperature and the power of other indoor heat sources. In IoT applications where only the total power consumption of users and the outdoor temperature are available, attempting to solve for so many unknown parameters with limited inputs constitutes a typical parameter identification redundancy and ill-posedness problem. This greedy modeling approach leads to non-unique and poorly robust evaluation results, severely impacting the accuracy and reliability of demand response potential assessment and representing a core technical bottleneck hindering the large-scale commercial application of this technology.

[0022] Therefore, in order to solve the above-mentioned technical problems, this application proposes a method for assessing the air conditioning load demand response potential based on the Internet of Things. Figure 1 This is a flowchart of an IoT-based method for assessing air conditioning load demand response potential according to an embodiment of this application. Figure 2 This is a data flow diagram of an IoT-based air conditioning load demand response potential assessment method according to an embodiment of this application. Figure 1 and Figure 2As shown, the IoT-based air conditioning load demand response potential assessment method according to an embodiment of this application includes: Step 1, acquiring the original aggregated power data stream and the original outdoor temperature data stream; Step 2, performing data preprocessing and air conditioning load decomposition on the original aggregated power data stream and the original outdoor temperature data stream to obtain the air conditioning power sequence and the outdoor temperature sequence; Step 3, dividing the air conditioning power sequence and the outdoor temperature sequence into operating state intervals to obtain a set of steady-state data points and a set of dynamic data segments; Step 4, identifying steady-state parameters and dynamic boundaries on the set of steady-state data points to obtain the steady-state thermodynamic coefficient, the upper boundary function of power consumption, and the lower boundary function of power consumption; Step 5, identifying the dynamic time constant on the set of dynamic data segments to obtain the dynamic time constant; Step 6, based on the steady-state thermodynamic coefficient, the upper boundary function of power consumption, the lower boundary function of power consumption, and the dynamic time constant, performing a comprehensive demand response assessment on the current air conditioning power and the current outdoor temperature to obtain the potential for reduction, the potential for increase, and the response duration.

[0023] In step 1, the raw aggregated power data stream and raw outdoor temperature data stream are acquired. It should be understood that, in the context of modern smart grids and efficient energy management, assessing the demand response potential of a large number of geographically distributed air conditioning loads is a crucial step in improving grid operational flexibility and optimizing energy utilization. However, traditional assessment methods often rely on direct monitoring of user indoor environmental parameters and the operating status of individual air conditioners, such as accurately acquiring indoor temperature data and individual air conditioner power consumption data. This intrusive data collection method is not only technically complex and costly, but more importantly, it severely infringes on user privacy, making it difficult for such solutions to be widely adopted and applied in practice. To overcome these inherent limitations, building an IoT assessment framework that can both protect user privacy and achieve accurate and efficient assessment has become particularly urgent. Therefore, acquiring the raw aggregated power data stream and raw outdoor temperature data stream is precisely to provide the necessary and compliant basic data support for subsequent air conditioning load demand response potential assessment in a completely non-intrusive scenario.

[0024] In an exemplary operation, the specific process of step 1 is as follows: First, the core of acquiring the raw aggregated power data stream lies in utilizing the existing smart meter infrastructure. Smart meters are installed at the user's power access point and can measure and record the user's total power consumption in real time and automatically over a specific time period. Therefore, the raw aggregated power data stream refers to the raw user total power time series data collected from these smart meters. For example, by interfacing with a smart meter data management platform, the total power consumption reading of a user per minute or every fifteen minutes can be continuously received and stored, in kilowatts (kW). For instance, during a specific time period, a smart meter might record the following sequence of data from 14:00:00 to 14:05:00: [1.2 kW, 2.8 kW, 15.0 kW, 2.75 kW, null, 2.78 kW], thus forming a continuous and detailed aggregated power time series. This acquisition method using smart meters ensures the comprehensiveness, continuity, and non-intrusiveness of the data, providing raw input for subsequent load decomposition work.

[0025] Secondly, the acquisition of raw outdoor temperature data streams primarily relies on mature and widely used public meteorological services. These services are operated by professional meteorological agencies or third-party data providers, whose weather stations across various locations continuously monitor and publish ambient temperature data for their respective areas. Therefore, raw outdoor temperature data streams refer to the time-series data of raw outdoor temperature changes obtained from these public meteorological services. The acquisition process is implemented through API calls or data subscriptions to obtain real-time or historical outdoor temperature records for a specific geographic area. For example, outdoor temperature data for a specific location in a city can be obtained hourly, such as 32.1 degrees Celsius at 14:00, 32.5 degrees Celsius at 15:00, and 32.3 degrees Celsius at 16:00. These public temperature data sources are highly reliable, have broad coverage, and are freely available, providing important external environmental parameters for the thermodynamic analysis of air conditioning loads, while also achieving completely non-intrusive data acquisition for users.

[0026] In step 2, the raw aggregated power data stream and raw outdoor temperature data stream are preprocessed and decomposed into air conditioning load to obtain air conditioning power sequences and outdoor temperature sequences. However, after acquiring the raw, unprocessed aggregated power and outdoor temperature data streams, these data cannot be directly used to accurately assess the demand response potential of a single air conditioning load. The raw aggregated power data is mixed with power consumption information from all appliances in the user's home, forming a complex and noisy signal, which masks the unique operating power consumption characteristics of the air conditioner. Furthermore, the acquisition frequency, timestamps, and data quality (such as abnormal spikes or missing data) of these two data streams are often inconsistent, making direct correlation analysis impossible. To accurately separate the independent power consumption of the air conditioning load from the mixed electrical signal and ensure that it is strictly synchronized with the corresponding outdoor ambient temperature in the time dimension and that the data is clean and error-free, a series of data purification, target signal extraction, and time-series alignment operations are required. Therefore, data preprocessing and air conditioning load decomposition of the raw aggregated power data stream and raw outdoor temperature data stream can construct a high-quality, time-aligned analysis dataset that contains only the two core variables of air conditioning power consumption and outdoor temperature, laying a solid and reliable data foundation for subsequent operation state interval division and parameter identification.

[0027] In an exemplary operation, step 2 is as follows: The original aggregated power data stream obtained above is [1.2 kW, 2.8 kW, 15.0 kW, 2.75 kW, null, 2.78 kW], with a sampling interval of one minute. In this sequence, 15.0 kW is identified as an outlier due to a data acquisition or transmission error, as it far exceeds the normal electricity consumption range of a typical household. This outlier can be identified and removed by setting a reasonable threshold, such as 10 kW, or by using statistical methods such as the three-times-standard-deviation method. After removal, linear interpolation of adjacent values ​​is used to fill the gap, i.e., interpolation is performed based on 2.8 kW at 14:01:00 and 2.75 kW at 14:03:00, resulting in approximately 2.775 kW. The missing value at 14:04:00 in the sequence represents a missing data point, which is filled using linear interpolation. Based on the preceding and following data points of 2.75 kW and 2.78 kW, the interpolation value is 2.765 kW. After anomaly handling and missing value filling, a smoothing filter, such as a moving average filter, is applied to the entire sequence to reduce high-frequency noise introduced by the frequent start-stop of other small appliances, ultimately resulting in a cleaned raw aggregated power data stream. Similarly, a similar data cleaning process is performed on the raw outdoor temperature data stream to ensure data integrity and accuracy, resulting in a cleaned raw outdoor temperature data stream.

[0028] Next is the air conditioning load decomposition stage. The cleaned raw aggregated power data stream is input into a pre-trained non-intrusive load decomposition model. The model's role is to identify and separate the operating power of the air conditioning units from the total power. In this embodiment, a deep learning-based model can be used, such as an architecture based on Long Short-Term Memory (LSTM) networks. This network model consists of an input layer, several LSTM hidden layers, and an output layer. Its network weights and biases are determined through an offline training process: supervised learning is performed on a public dataset containing a large amount of household electricity consumption data (such as UK-DALE), using data pairs that include both total power data and individual power labels for air conditioners. Through repeated iterations using backpropagation algorithms and optimizers such as the Adam optimizer, the model learns the ability to extract air conditioning-specific electricity consumption characteristics, such as high current during startup, stable power plateaus during operation, and periodic fluctuations, from the complex patterns of the aggregated power sequence. When the cleaned original aggregated power data stream in this embodiment, such as [1.2 kW, 2.8 kW, 2.775 kW, 2.75 kW, 2.765 kW, 2.78 kW], is input into this trained model, the model's output becomes a new time series, representing the estimated operating power of the air conditioning equipment at the corresponding time. For example, the model might output the corresponding air conditioning power series as [0.1 kW, 1.6 kW, 1.58 kW, 1.55 kW, 1.57 kW, 1.59 kW]. In other words, it reveals a clear physical process: at 14:00:00, the air conditioner is not running or is in a low-power state of 0.1 kW, with a base load of 1.1 kW (1.2-0.1); at 14:01:00, the air conditioner starts, and its power jumps to 1.6 kW, while the total power also jumps accordingly from 1.2 kW to 2.8 kW, with the base load at this point being 1.2 kW (2.8-1.6). At subsequent time points, the air conditioner power fluctuates slightly between 1.55 kW and 1.59 kW, which is consistent with the stable operation characteristics of inverter air conditioners, and the calculated base load remains stable at around 1.2 kW. This ability to successfully decompose the total power into a stable base load and a dynamic air conditioner load proves the effectiveness of the decomposition result, ultimately yielding a preliminary air conditioner power sequence.

[0029] Finally, crucial timescale matching is performed. Since the initial air conditioner power sequence (e.g., one data point per minute) differs significantly in time resolution from the cleaned outdoor temperature data stream (e.g., one data point per hour), unification and alignment are necessary. This process must be differentiated based on the air conditioner type. For fixed-frequency air conditioners, their operating mode exhibits a clear start-up-stable operation-stop cycle. Therefore, a complete operating cycle is considered as an analysis event. Such a cycle is identified from the initial air conditioner power sequence, for example, from 14:10 to 14:30, during which the air conditioner runs continuously. Then, the average of all power readings within this 20-minute period is calculated to obtain an average power value, such as 1.45 kW. Simultaneously, for the outdoor temperature, linear interpolation is needed on the hourly data of the cleaned outdoor temperature data stream to calculate the average outdoor temperature during the time period from 14:10 to 14:30, such as 32.4 degrees Celsius. In this way, a complete operating event is condensed into a single data point pair (…). 1.45 kilowatts, (32.4 degrees Celsius). For inverter air conditioners, whose power varies continuously, event-driven matching is not suitable. In this case, resampling is performed at fixed time intervals. A time window is set, for example, 15 minutes. The initial air conditioner power sequence is segmented into 15-minute intervals, and the power average within each segment is calculated. Similarly, the cleaned raw outdoor temperature data stream is also aligned and averaged in the same 15-minute intervals. For example, for the period from 14:00 to 14:15, the average air conditioner power is calculated to be 1.2 kW, and the average outdoor temperature is 32.2 degrees Celsius. By performing this operation on all historical data, two time-aligned sequences with the same time index are finally generated: the air conditioner power sequence and the outdoor temperature sequence.

[0030] In step 3, the air conditioning power sequence and outdoor temperature sequence are divided into operating state intervals to obtain a set of steady-state data points and a set of dynamic data segments. It is understandable that in the previous step, through data preprocessing and air conditioning load decomposition, a clean air conditioning power sequence strictly time-aligned with outdoor temperature has been successfully extracted from the complex total electricity consumption data. However, this continuous time-series data stream itself mixes various physical states of air conditioning operation. When the air conditioner starts, stops, or significantly adjusts its power, its power consumption changes drastically; this is a dynamic process, mainly affected by the building's heat capacity and the air conditioner's own inertia. When maintaining a stable indoor temperature, its power consumption is relatively stable, fluctuating only slightly with the outdoor temperature; this is steady-state operation, mainly reflecting the building's insulation performance. The thermodynamic mechanisms behind these two states are completely different. Mixing them together for unified parameter identification will lead to model mismatch and serious deviations in evaluation results. To accurately characterize the behavior of the air conditioning load under different states, the data needs to be physically deconstructed first. Therefore, dividing the air conditioning power series and outdoor temperature series into operating state intervals is to decouple the original data that contains different operating characteristics, and separate two types of high-quality, homogeneous datasets that can be used to identify steady-state thermodynamic coefficients and dynamic time constants, respectively, thus laying a key foundation for the pertinence and accuracy of subsequent parameter identification.

[0031] In an exemplary operation, the specific process of step 3 is as follows: To ensure the completeness of the example, the data stream of the previous embodiment is reused and extended to generate a longer time series that includes not only the start-up process of the air conditioner but also its stable operation and eventual shutdown phases. For example, after time-scale matching, a power sequence of the variable frequency air conditioner with a sampling interval of one minute is obtained. The values ​​(in kilowatts) are [0.10, 1.60, 1.58, 1.55, 1.57, 1.59, 1.60, 1.61, 1.59, 1.60, 1.62, 1.60, 1.20, 0.50, 0.10, 0.10, 0.10, 0.10]. Correspondingly, the outdoor temperature sequence after interpolation and alignment... The unit is [32.20, 32.21, 32.22, 32.23, 32.24, 32.25, 32.26, 32.27, 32.28, 32.29, 32.30, 32.31, 32.32, 32.33, 32.34, 32.35, 32.36, 32.37].

[0032] The first step is to calculate the instantaneous rate of change, which aims to quantify the degree of change in air conditioning power at each point in time. This process uses the backward difference method to calculate the instantaneous power change rate and takes its absolute value to obtain a unified index unaffected by the direction of change. The calculation follows the formula: In this formula, Is The absolute rate of change of power at time t; and These are the air conditioning power at the current moment and the previous moment, respectively; This is the time step, which is 1 minute in this example. The physical meaning of this formula is to calculate the magnitude of power change per unit time. Specifically, at time point... The corresponding power is 1.60 kW, and its absolute power change rate is |(1.60-0.10) / 1| = 1.50 kW / min. This is a very large change rate, directly reflecting the air conditioner's startup process. And at the time point... The corresponding power is 1.61 kW, and its absolute power change rate is |(1.61-1.60) / 1| = 0.01 kW / min. This value is very small, indicating that the air conditioner is operating stably. By calculating the power sequence of the entire air conditioner point by point, a sequence of absolute power change rates of the same length as the original sequence can be generated. .

[0033] The second step is to calculate the normalized state discrimination index and label the state. The simple absolute power change rate may be affected by the rated power of the air conditioner itself. For example, a 0.1 kW / min change in a high-power air conditioner might be considered normal fluctuation, while for a low-power air conditioner it might signify a state change. To eliminate this baseline dependence, the change rate needs to be normalized. The formula for calculating the normalized state discrimination index is as follows: , in the formula, yes Normalized state discrimination index at time step. Denominator part. It is the average power of the current moment and the previous moment, used as the normalization benchmark. Using the average power is more robust than using only the power of the previous moment, and can effectively avoid the influence of... This index can cause calculation overflow or unstable results due to values ​​close to zero (e.g., when the air conditioner is not turned on). The unit is (kW / min) / kW, or 1 / min, which physically represents the percentage change in power per unit time, thus becoming a relative change measure independent of the air conditioner's own power. Next, a predefined state discrimination threshold is set. The threshold can be determined based on statistical analysis of a large amount of operating data from different types of air conditioners, selecting a value that can effectively distinguish between steady-state and dynamic ranges, or it can be set based on expert experience. For example, in this embodiment, it can be set to... =0.1 (1 / minute), which means that if the power change within one minute exceeds 10% of its average operating power, it is considered to be in a dynamic state. Subsequently, the calculated value is used to iterate through each time point. and Compare and generate a sequence of state labels. Continuing with the example data: at time point , =1.50, the average power is (1.60+0.10) / 2=0.85, therefore =1.50 / 0.85≈1.76. Since 1.76>0.1, therefore... Marked as dynamic. At a point in time. , =0.01, the average power is (1.61+1.60) / 2=1.605, therefore =0.01 / 1.605≈0.0062. Since 0.0062<0.1, therefore... It is marked as a steady state. The entire sequence is processed in the same way, as in the final state label sequence. The values ​​are: [dynamic, dynamic, dynamic, dynamic, dynamic, steady state, steady state, steady state, steady state, steady state, steady state, dynamic, dynamic, dynamic, steady state, steady state]. It is worth mentioning that the first point has no label, which can be assigned according to the context rules during subsequent processing.

[0034] The third step is data aggregation and output set generation. After obtaining the state label for each data point, the original time-series data needs to be reorganized based on these labels, and then assigned to the steady-state data point set and the dynamic data segment set respectively. For this purpose, an empty steady-state data point set is initialized. An empty collection of dynamic data fragments and a temporary dynamic data fragment buffer. Then, starting from the first point in the time series, the state label sequence is traversed in chronological order. When traversing to a time point labeled as steady state... First check Is it empty? If not empty, it means a dynamic process has just ended. At this point, all consecutive dynamic data points stored in the buffer are stored as a complete segment (a time series list) and then stored in... Then, immediately clear the buffer. Subsequently, extract the data pairs from the current steady-state point. Add it to In the middle. When traversing to a time point where the label is dynamic. If so, extract the complete data point including the timestamp. and add it to In the middle. After the entire sequence traversal is completed, it is necessary to check again. If data is still present, this indicates that the entire time series ends with a dynamic process, therefore the remaining data in the buffer needs to be stored as a complete segment. In the example data: the first five points of the sequence are marked as dynamic, and their data... They are stored sequentially into the buffer. Upon encountering the first steady-state label, the five data points in the buffer are stored as a startup dynamic fragment. And clear the buffer, then The data pair (1.60, 32.26) is stored in... The next... arrive All are in a steady state, and their data pairs are stored one by one. .exist When a dynamic tag is encountered again, its data is stored in a buffer. and The same applies. Upon encountering a steady-state label, this will trigger another buffer dump operation, storing the shutdown dynamic fragment containing three data points. And clear the buffer, then The data pair (0.10, 32.35) was stored in Subsequent points are all in a steady state, and their data pairs are also added. After the iteration is complete, the buffer is empty. Ultimately, this section successfully outputs two structured data sets. By analyzing all time points, a final set of steady-state data points is obtained. This includes all data points in the form of [(1.60,32.26),(1.61,32.27),(1.59,32.28),(1.60,32.29),(1.62,32.30),(1.60,32.31),(0.10,32.35),(0.10,32.36),(0.10,32.37)…], which reflect the power consumption characteristics of the air conditioner during stable operation under different outdoor temperatures. This is a dynamic data fragment set. This is a list containing two independent sublists. One sublist details the complete power change process of the air conditioner from startup to stabilization, while the other sublists detail its complete operation from startup to shutdown. Specifically, [(t1,1.60,32.21),(t2,1.58,32.22),(t3,1.55,32.23),(t4,1.57,32.24),(t5,1.59,32.25)],[(t 12 ,1.20,32.32),(t 13 ,0.50,32.33),(t 14 ,0.10,32.34)...).

[0035] In a preferred exemplary operation, step 3, dividing the air conditioner power sequence and outdoor temperature sequence into operating state intervals to obtain a steady-state data point set and a dynamic data segment set, includes: step 31, determining a state label sequence based on the air conditioner power sequence and outdoor temperature sequence; step 32, performing data aggregation and output set generation on the air conditioner power sequence and outdoor temperature sequence based on the state label sequence to obtain the steady-state data point set and the dynamic data segment set. Specifically, the implementation process of step 32 is the same as the third step in the exemplary step 3 above, and will not be described here. The detailed implementation of step 31 will be described here.

[0036] It is understandable that in the initial method of deconstructing air conditioner operating data to separate steady-state and dynamic ranges, the core criterion relies solely on the rate of change of the air conditioner's power. The fundamental flaw of this mechanism lies in its lack of awareness of key physical context; specifically, it ignores the strong physical coupling between the air conditioner's operating power and the outdoor temperature. This mechanism, relying solely on the rate of change of power, is a closed, autoregressive analytical paradigm, thus failing to effectively distinguish between two fundamentally different types of power changes: one is a smooth power adjustment driven by outdoor temperature fluctuations and fully conforming to thermodynamic laws, essentially belonging to the steady-state operation category; the other is a power surge dominated by the equipment's internal control logic (such as compressor start / stop, sudden changes in inverter speed, or remote temperature adjustment by the user), inconsistent with the current outdoor temperature trend—this is the true dynamic process that needs to be accurately captured.

[0037] Because it cannot distinguish between these two types of changes, this mechanism is highly prone to misjudgment. A typical scenario is that when the outdoor environment changes rapidly, such as a sudden drop in temperature due to an afternoon thunderstorm, the air conditioner's power will also rapidly and adaptively decrease. Although this is a perfectly normal steady-state response, its instantaneous power change rate may be large, causing the original mechanism to incorrectly label it as a dynamic event. Such misjudgments severely pollute the dynamic dataset used to identify the system's thermal inertia, and also incorrectly remove valuable sample points from the steady-state dataset.

[0038] To address the aforementioned technical deficiencies, this application proposes a state partitioning optimization technique based on physical residual analysis. It should be understood that, in the process of seeking a solution, a possible improvement approach is as follows:

[0039]

[0040] in, In the first The average energy consumption of air conditioning load over a time interval In the first Average outdoor temperature over a time interval For the first The duration of each time interval. This method attempts to overcome the deficiency of the basic method, which completely ignores temperature. Instead of simply analyzing the rate of change of power itself, it calculates the rate of change of the ratio of air conditioning load power consumption to outdoor temperature over time. The logic is that the ratio can, to some extent, approximate the overall thermal conductivity of a building. Therefore, when this equivalent thermal conductivity changes drastically, the system is considered to have entered a dynamic state.

[0041] However, in-depth research revealed that while the above method represents an improvement in approach, it still suffers from significant technical bottlenecks: the ratio of air conditioning load operating power consumption to outdoor temperature is not a strictly physical invariant. It is affected by various factors such as indoor set temperature and air conditioning energy efficiency coefficient, and is not an ideal state indicator. When the outdoor temperature is low, the denominator of this ratio is small, and even small power or temperature noise can cause drastic fluctuations in the ratio, leading to numerous state misjudgments. Its core logic remains detecting the rate of change, failing to achieve prediction and comparison. It cannot truly isolate the power change component that should be caused by temperature changes. For example, in regions where temperature and power are nonlinearly coupled, even if they change completely synchronously and in coordination (essentially a steady state), the rate of change of their ratio may exceed a threshold due to the nonlinear relationship, leading to misjudgments.

[0042] Therefore, this application uses a preferred technique of determining the state label sequence by performing state division based on physical residual analysis on the air conditioning power sequence and the outdoor temperature sequence. The core of this technique is to introduce a module that can adaptively estimate the expected power change. By quantifying the deviation between the actual power change and the expected power change, a more physically insightful state discrimination index is constructed.

[0043] Based on this, in an exemplary operation, step 31, determining the state label sequence based on the air conditioning power sequence and the outdoor temperature sequence, includes: step 311, performing dynamic thermal sensitivity coefficient adaptive estimation on the air conditioning power sequence and the outdoor temperature sequence to obtain the dynamic thermal sensitivity coefficient sequence; step 312, calculating the expected power change and residual power change on the air conditioning power sequence and the outdoor temperature sequence based on the dynamic thermal sensitivity coefficient of the previous moment to obtain the residual power change sequence; step 313, calculating the residual index and marking the state based on the residual power change sequence and the air conditioning power sequence to obtain the state label sequence.

[0044] Step 311: Establish a dynamic parameter capable of adaptively tracking the recent thermodynamic characteristics of the system, providing crucial information for accurate prediction of temperature-driven power changes. Unlike a fixed, global steady-state thermodynamic coefficient, this dynamic thermal sensitivity coefficient... It is a local parameter that updates iteratively over time. It reflects the change in air conditioning power caused by a unit change in outdoor temperature in real time through an exponential smoothing averaging mechanism. The advantage of this is that its value can effectively suppress the interference of short-term data noise and can slowly and adaptively track system sensitivity drift caused by factors such as seasonal changes, equipment aging, and even changes in user window opening and closing habits, thereby obtaining a dynamic thermal sensitivity coefficient sequence that is both robust and reflects the recent characteristics of the system. In an exemplary operation, step 311, performing adaptive estimation of the dynamic thermal sensitivity coefficients on the air conditioning power sequence and the outdoor temperature sequence to obtain the dynamic thermal sensitivity coefficient sequence, includes: performing adaptive estimation of the dynamic thermal sensitivity coefficients on the air conditioning power sequence and the outdoor temperature sequence using the following formula:

[0045]

[0046] in, for The dynamic thermal sensitivity coefficient at any given time. for Air conditioner power at all times for outdoor temperature at any time This is a smoothing factor, ranging from 0 to 1. This formula is a typical first-order low-pass filter or exponential moving average. Wherein, This is the dynamic thermal sensitivity coefficient updated at time ti. The first term... The calculation measures the instantaneous thermal sensitivity within the most recent time step, reflecting the latest system response. The second term... This represents historical sensitivity. Smoothing factor. It is a preset value between 0 and 1, used to weigh the importance of new information against historical information. The value is set based on a trade-off between system response speed and noise immunity, and is set to a small value, such as 0.1, to ensure the stability of the coefficient. The initial value of this coefficient... It can be obtained through linear regression of a small segment of data at the beginning of the data sequence, or it can be set as an industry experience value. For example, in At any given time, the existing dynamic thermal sensitivity coefficient It is 0.155 kilowatts per degree Celsius. In At that moment, observed =1.60 kilowatts, =32.5 degrees Celsius, and =1.52 kilowatts, =32.0 degrees Celsius. Let the smoothing factor be... =0.1. First, calculate the instantaneous thermal sensitivity: (1.52-1.60) / (32.0-32.5) = 0.16 kW / °C. Then update the dynamic thermal sensitivity coefficient: =0.1×0.16+(1-0.1)×0.155=0.1555 kW / degree Celsius.

[0047] Step 312: Decompose the total, observed power change into two parts: those explainable by physical laws and those not, thereby accurately isolating the signal that truly characterizes the state transition. Specifically, it utilizes the thermal sensitivity coefficient obtained in the previous step, which represents the recent physical characteristics of the system, from the previous moment. To predict the expected power change driven entirely by changes in outdoor temperature within the current time step. Subsequently, the expected portion is subtracted from the actual observed total power change; the difference is the residual power change. The physical meaning of this residual is that it represents power fluctuations that cannot be explained by changes in the external ambient temperature, and is an unexpected signal of deviation from the expected physical baseline. A very small residual value strongly indicates that the power change is highly consistent with the temperature change, and the system is in a steady state; conversely, a large residual value suggests that a dynamic event dominated by the internal control logic of the device has occurred. In an exemplary operation, step 312, based on the dynamic thermal sensitivity coefficient of the previous moment, calculates the expected power change and residual power change of the air conditioning power sequence and the outdoor temperature sequence to obtain the residual power change sequence, including: calculating the expected power change and residual power change of the air conditioning power sequence and the outdoor temperature sequence using the following formula:

[0048]

[0049]

[0050] in, for The expected change in power at any given time. for The dynamic thermal sensitivity coefficient at any given time. for outdoor temperature at any time for The change in residual power at time t. for The air conditioner power at any given time. Continuing with the example above, in... At any given moment, it's necessary to determine the state. (Use...) Sensitivity coefficient at time =0.155 kW / degree Celsius. Outdoor temperature variation is... =32.0 - 32.5 = -0.5 degrees Celsius. Calculate the expected power change: =0.155 × (-0.5) = -0.0775 kW. This means that, according to the physical model, the expected power of the air conditioner would decrease by 0.0775 kW. The actual observed power change is... =1.52-1.60=-0.08 kW. Calculate the residual power change: =-0.08-(-0.0775)=-0.0025 kW. This residual value is very small. Let's look at the next time step. Situations such as when the air conditioner suddenly turns off: =0.10 kilowatts, =31.9 degrees Celsius. Use at this time. Sensitivity coefficient at time =0.1555 kW / °C. Expected power change: =0.1555×(31.9-32.0)=-0.01555 kW. Actual power change: =0.10-1.52=-1.42 kW. Residual power change: =-1.42-(-0.01555)=-1.40445 kilowatts. This residual value is extremely large.

[0051] Step 313: Transform the physically meaningful but still absolute residual power change obtained in the previous step into a standardized, power-reference-independent discrimination index, so that a unified threshold can be used for robust state partitioning. In specific implementation, first, the residual power change Δ... Take the absolute value, and then normalize it using the average power of the current time step and the previous time step to obtain the final normalized residual index. This indicator physically quantifies the relative significance of power fluctuations not explained by temperature changes relative to the current operating power level of the air conditioner. Compared to the indicator in the original mechanism, which is based solely on the rate of change of total power, this residual indicator has stronger physical interpretability, is less sensitive to external environmental disturbances, and is more robust. The formula for calculating this indicator is:

[0052]

[0053] Finally, this metric is compared to a preset state discrimination threshold. This threshold can be set by statistically analyzing a large amount of typical air conditioning operation data labeled with steady-state and dynamic states to find a value that maximizes the discrimination; for example, it could be set to 0.2. Continuing with the example above: time: =0.0025 kW; average power The result is (1.52 + 1.60) / 2 = 1.56 kW. Normalized residual index. =0.0025 / 1.56≈0.0016. Since 0.0016 < state discrimination threshold (0.2), therefore The time point was correctly labeled as the steady state. time: =1.40445 kW; average power is (0.10+1.52) / 2=0.81 kW. Normalized residual index =1.40445 / 0.81≈1.73. Since 1.73 > the state discrimination threshold of 0.2, therefore The time points were correctly labeled as dynamic. Through this analysis method based on physical residuals, normal power regulation driven by temperature and true equipment state switching were successfully distinguished, greatly improving the accuracy of state classification and laying a solid foundation for the reliability of subsequent parameter identification.

[0054] In step 4, steady-state parameters and dynamic boundaries are identified on the steady-state data point set to obtain the steady-state thermodynamic coefficient, upper boundary function of power consumption, and lower boundary function of power consumption. It should be understood that after successfully dividing the air conditioner operating data into steady-state and dynamic sets, a pure steady-state data point set is obtained, which depicts the intrinsic relationship between the air conditioner's energy consumption and the ambient temperature when its power is stable. However, these discrete data points themselves cannot directly constitute a generalized model that can be used for real-time evaluation. Due to the influence of various random factors such as user-set temperature and indoor occupant activity, even at the same outdoor temperature, the steady-state power consumption of the air conditioner will fluctuate within a range rather than being a fixed value. To quantify demand response potential, it is necessary not only to understand the average trend of its power consumption changing with temperature, but more importantly, to define the power adjustment range within which it can operate stably at any given temperature. Without accurately characterizing this operating boundary, it is impossible to determine how much upward or downward adjustment space the current power has. Therefore, the purpose of this application to identify steady-state parameters and dynamic boundaries of the steady-state data point set is to extract macroscopic parameters describing the core correlation between air conditioner power consumption and outdoor temperature from discrete steady-state data points, and to define the power adjustment range for stable operation at any temperature, thereby providing a mathematical model and decision-making basis for directly quantifying the power potential that can be reduced and increased.

[0055] In an exemplary operation, step 4 follows this process: Before identifying the thermodynamic parameters, it is noted that the set contains two distinct physical states. Data points with power around 1.60 kW correspond to the steady-state cooling state where the air conditioner compressor is running, while data points with power of only 0.10 kW correspond to the non-cooling steady-state cooling state where the air conditioner is in standby or only supplying air. The latter's power consumption does not follow the thermodynamic laws of cooling in relation to the outdoor temperature. Therefore, the data is first filtered to ensure that only data points reflecting actual cooling behavior are used for modeling. Here, a power threshold of 0.3 kW is set, and only data points with power greater than this value are retained. This power threshold is pre-set based on the basic physical operating characteristics of the air conditioning equipment and combined with industry experience, aiming to distinguish whether the compressor is in an effective cooling working state. After filtering, the set of valid steady-state data points for subsequent analysis is updated to [(1.60,32.26),(1.61,32.27),(1.59,32.28),(1.60,32.29),(1.62,32.30),(1.60,32.31)…].

[0056] The first step is to perform linear regression fitting to identify the steady-state thermodynamic coefficient. Using outdoor temperature as the independent variable and air conditioning power as the dependent variable from the selected set of effective steady-state data points, the least squares method is applied to fit the linear relationship between the two. This linear relationship is expressed by the following formula: In this formula, This is the steady-state power of the air conditioner. That corresponds to the outdoor temperature. It is the slope of the fitted straight line, i.e., the steady-state thermodynamic coefficient, and its unit is kilowatts per degree Celsius. Physically, it is the average increase in power required by the air conditioner to maintain a stable indoor temperature for every degree increase in outdoor temperature. This is the intercept of the fitted line. Specifically, the least squares method determines this by solving an optimization problem. and The value is to make the actual power of all data points equal. The power predicted by this linear model ( The sum of squared residuals between the two is minimized. This solution process has an exact mathematical analytical solution. Steady-state thermodynamic coefficient The calculation formula is as follows: First, calculate the covariance of the outdoor temperature series and the air conditioning power series; then calculate the variance of the outdoor temperature series; finally, divide the two to obtain the result. The intercept... This is achieved by subtracting the average value from the power points. The value is determined by multiplying by the average of all temperature points. Linear regression is then performed on the selected data points. Due to the small sample size and narrow temperature range, to make the example more general, it is stated here that the calculation result is obtained by combining similar data points from a longer historical period. For example, the final fitted parameter is: steady-state thermodynamic coefficient. =0.15 kW / °C, intercept = -3.24 kW. This represents the core thermodynamic response characteristics of the air conditioning-building system, i.e., its power consumption central trend line is... .

[0057] The second step is to calculate the residuals and their standard deviations. To quantify the fluctuation range of the actual running points around the central trend line, it is necessary to calculate the residuals for each data point. That is, the difference between the actual power and the model-predicted power: Taking the data point (1.60, 32.26) as an example, its predicted power is... The residual at this point is 0.15 × 32.26 - 3.24 = 1.599 kilowatts. =1.60-1.599=0.001 kW. Taking data point (1.62, 32.30) as an example, its predicted power is 0.15×32.30-3.24=1.605 kW, and its residual is 1.62-1.605=0.015 kW. The residuals for all six data points are calculated using the same method to obtain the residual sequence. Then, the standard deviation of this residual sequence is calculated. Calculations yielded the following results: It is approximately 0.012 kilowatts. This value measures the typical magnitude of power fluctuations caused by other random factors such as user behavior, after excluding the linear effect of outdoor temperature.

[0058] The third step is to construct the upper and lower power consumption boundary functions. These two functions together define the power path of the air conditioner under steady-state operation at any outdoor temperature. Upper power consumption boundary function Defined based on an upward offset of the central trend line by twice the standard deviation of the residuals, this statistically covers approximately 95% of normal operating points. Its functional form is: Substituting the identified parameters, we get: .

[0059] Power consumption lower boundary function Similarly, it is defined based on a downward offset of the central trend line by twice the standard deviation of the residuals: Substituting the parameters, we get: .

[0060] These two boundary functions clearly define the conditions under a given arbitrary outdoor temperature. The power range within which the air conditioner can operate stably.

[0061] In step 5, the dynamic time constant is identified from the set of dynamic data segments to obtain the dynamic time constant. Accordingly, after identifying the steady-state operating characteristics of the air conditioning system, the power regulation range within which it can operate stably under any outdoor temperature can be defined. However, a complete assessment of demand response requires not only knowing how much power can be adjusted (i.e., the reduction / increase potential), but also knowing how long this adjustment can be sustained safely and effectively (i.e., the response duration). When the air conditioning power is reduced, the building interior gradually warms up due to continuous heat infiltration, and the duration of the response is directly limited by the rate at which the indoor temperature rises to the upper limit of user comfort. This rate of temperature rise is not constant, but is dominated by thermal inertia determined by factors such as the building's insulation performance, space size, and air circulation. The previously separated dynamic data segments—the complete process of the air conditioning system starting up and increasing power or shutting down and decreasing power—are precisely a direct manifestation of this thermal inertia under power changes. Therefore, the dynamic time constant is identified from the set of dynamic data segments to extract and quantify the key macroscopic parameter that represents the unique thermal inertial characteristic of the building-air conditioning system, namely the dynamic time constant. This parameter is the core and fundamental element for accurately determining the response duration and ensuring that demand response events do not affect user comfort.

[0062] Figure 3 This is a flowchart of step 5 in the IoT-based air conditioning load demand response potential assessment method according to an embodiment of this application. In one exemplary operation, such as Figure 3As shown, step 5, which involves identifying the dynamic time constant of the dynamic data segment set to obtain the dynamic time constant, includes: step 51, filtering effective dynamic segments from the dynamic data segment set and the steady-state data point set to obtain a set of fitable segments; step 52, performing model transformation and linear regression fitting on the set of fitable segments to obtain a list of candidate time constants; and step 53, aggregating candidate values ​​from the list of candidate time constants to obtain the dynamic time constant.

[0063] In the above exemplary operation, the specific process of step 5 is as follows: Step 51: First, each dynamic data segment in the dynamic data segment set is screened. Two core screening criteria are set to ensure the quality of the segments used for fitting. The first is the minimum duration, which aims to exclude false dynamic segments caused by signal noise or brief power disturbances. This threshold can be set empirically, for example, 4 minutes for minute-level sampling data. The second is the minimum power change amplitude, to ensure that the selected segment corresponds to a significant, physically meaningful state switching process, rather than a small power fluctuation in steady-state operation. This threshold can be set to an absolute value, such as 1 kilowatt, or a percentage relative to the rated power of the air conditioner. For the first dynamic segment in this embodiment, its duration is 5 time steps (i.e. - The duration is 5 minutes, satisfying the minimum duration of 4 minutes. Its power change needs to be judged in conjunction with the steady-state conditions before and after it. It starts from approximately 0.1 kW in the off state and increases to approximately 1.6 kW in the running state, a change of approximately 1.5 kW, which is much greater than the minimum power change of 1.0 kW. Therefore, this segment passes the screening. For the second dynamic segment, its duration is 3 time steps (i.e., - The segment (3 minutes) does not meet the minimum duration requirement of 4 minutes, therefore it will be discarded. Next, boundary parameters are extracted from the selected segments. For a valid dynamic segment, its power change process can be well described by a first-order exponential model, but this model requires knowledge of the initial steady-state power of its change process. and the final steady-state power Therefore, it is necessary to use a set of steady-state data points to determine these boundary values. For the first segment, its start time is... The end time is By searching within the complete original time series or nearby steady-state data points, it is possible to determine... The previous steady-state power was 0.10 kilowatts, therefore =0.10 kW. Then find the nearest neighbor in the set of steady-state data points. The subsequent steady-state point, the first steady-state point to appear is at The time interval is (1.60, 32.26), therefore it can be determined. =1.60 kW. If no corresponding steady-state boundary points can be found before or after a dynamic segment, for example, if the beginning or end of the data sequence is a dynamic process, then this segment cannot be used for fitting and needs to be discarded. After performing the above judgment on all segments, the detailed power time series of this segment, along with its boundary parameters, is encapsulated into a data structure containing {initial steady-state power, ending steady-state power, and time series data of the dynamic segment}, and stored in the set of fitable segments. Over long-term data accumulation, this set will contain multiple start-up and shutdown segments that meet the conditions. Finally, the data structure is encapsulated. The extracted boundary parameters... , And the power time series of that segment itself, i.e., from - The power sequence [1.60, 1.58, 1.55, 1.57, 1.59] is encapsulated into a structured data object. After performing the above judgment on all segments in the dynamic data segment set, the final output is a set of fitable segments, such as: { 0.10, :1.60, Power Time Series:[( ,1.60),( ,1.58),...,( ,1.59)]…}.

[0064] Step 52: The change in air conditioning power during the dynamic process can be approximated by a first-order response model, the original nonlinear exponential form of which is: In this formula, It is the air conditioning power at time t; This is the start time of the dynamic segment; It is the dynamic time constant to be solved, which characterizes the change in power from Change to The time elapsed when 63.2% of the required time is the core indicator for measuring system inertia. Directly fitting this nonlinear model is complex and unstable. Therefore, this application employs a method combining model transformation and linear regression. Through simple algebraic transformation, the above formula can be linearized as:

[0065]

[0066] The transformed formula reveals a key linear relationship: (Logarithmic power difference) and relative time The two variables are linearly related, with the form y = b + kx, where the dependent variable y = Independent variable x= The slope is This transforms a complex nonlinear fitting problem into a simple, efficient, and robust linear regression problem. Next, the following operations are performed for each segment in the set of fitable segments. First, the dataset required for the linear regression task is constructed. Then, each data point in its power time series is iterated over. To avoid computation... Time Too close The mathematical problem that leads to ln(0) can be solved by... Make minor adjustments, such as using the average of the next few steady-state points, or ignoring the closest values ​​during the calculation. The data points. In this embodiment, for simplicity, we continue to use... =1.60, and note =1.60, this point cannot be calculated. Therefore, from Begin building the dataset. For example... for The corresponding time point is 0. (Relative time x = 1 minute): y = ln|1.58 - 1.60| = ln(0.02) ≈ -3.91; (Relative time x = 2 minutes): y = ln|1.55 - 1.60| = ln(0.05) ≈ -3.00; (Relative time x = 3 minutes): y = ln|1.57 - 1.60| = ln(0.03) ≈ -3.51; (Relative time x = 4 minutes): y = ln|1.59 - 1.60| = ln(0.01) ≈ -4.61. This yields a dataset for linear regression: {(1, -3.91), (2, -3.00), (3, -3.51), (4, -4.61)}. Then, least squares linear regression is performed on this dataset, fitting a straight line y = kx + b. The slope is calculated to be approximately -0.2. Finally, inverse kinematics is performed. Based on the relationship between the slope k and the time constant τ, k = -1 / τ, the estimated time constant for this segment can be obtained. τ1 = -1 / k = -1 / (-0.2) = 5.0 minutes. This calculated value is added to the candidate time constant list. After processing all fitable segments, and after accumulating data over several days, the candidate time constant list is [5.0, 5.2, 4.9, 15.0, 5.3, 4.8].

[0067] Step 53: In practical applications, after a long period of data accumulation, the candidate time constant list may contain multiple candidate values ​​identified from dozens or even hundreds of valid dynamic segments. These values ​​may fluctuate due to measurement noise, model mismatch, etc. The purpose of this step is to aggregate from this list containing multiple candidate values ​​to obtain a stable and unique dynamic time constant that best represents the thermal inertia of the user's building-air conditioning system. First, outliers are removed from the candidate time constant list to enhance the robustness of the final result. To make this step more general, after several days of data accumulation, the candidate time constant list is [5.0, 5.2, 4.9, 15.0, 5.3, 4.8]. Among them, 15.0 deviates significantly from the other values, which may be a misfit caused by outlier data. A standard statistical method, namely, the method based on the median and interquartile range (IQR), is used to identify and remove outliers. First, the list is sorted: [4.8, 4.9, 5.0, 5.2, 5.3, 15.0]. The median is calculated as (5.0 + 5.2) / 2 = 5.1. The interquartile range (IQR) (the difference between the upper quartile Q3 and the lower quartile Q1) is calculated, yielding IQR = 5.3 - 4.9 = 0.4. A valid interval is then defined as [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR], i.e., [4.85 - 0.6, 5.25 + 0.6] = [4.25, 5.85]. The candidate value 15.0 is not within this interval and is therefore considered an outlier and removed. The cleaned list is [4.8, 4.9, 5.0, 5.2, 5.3]. Finally, a central tendency is calculated. A central tendency value is calculated for the cleaned list as the final result. While the mean is an option, using the median as the final aggregation result is more robust, considering the potential slight skewness in the data distribution. The median of the purified list is calculated as follows: τ = Median([4.8,4.9,5.0,5.2,5.3]) = 5.0 minutes. The final output is the user's unique dynamic time constant τ, which is 5.0 minutes in this example. This value accurately quantifies the building's thermal inertia and serves as a crucial bridge connecting power changes and temperature changes.

[0068] In step 6, a comprehensive demand response assessment is performed based on the steady-state thermodynamic coefficient, upper and lower power consumption boundary functions, and dynamic time constant to obtain the potential for reduction, potential for increase, and response duration. In other words, in the preceding steps, this technical solution has successfully identified a series of key parameters from the raw, non-intrusive data stream that accurately characterize the macroscopic physical properties of a specific user's building-air conditioning system. These parameters include the steady-state thermodynamic coefficient characterizing its thermodynamic sensitivity, the upper and lower power consumption boundary functions defining its power regulation range, and the dynamic time constant quantifying its thermal inertia. However, these parameters are static descriptions of the system's inherent properties. To truly achieve demand response, these static models must be applied to dynamic, real-time operating scenarios. The power grid's demand response command is issued at a specific moment, and the air conditioning's response capability is closely related to its current actual operating state. Therefore, based on the steady-state thermodynamic coefficient, the upper boundary function of power consumption, the lower boundary function of power consumption, and the dynamic time constant, a comprehensive demand response assessment is performed on the current air conditioning power and the current outdoor temperature. This combines the previously identified static characteristic model with the rapidly changing real-time operating data to perform a comprehensive application calculation, thereby outputting three most direct and critical actionable indicators for grid operators or load aggregators: how much power can be reduced at this moment, how much power can be increased, and how long the reduction can be safely sustained.

[0069] Figure 4 This is a flowchart of step 6 in the IoT-based air conditioning load demand response potential assessment method according to an embodiment of this application. In one exemplary operation, such as Figure 4 As shown, step 6, based on the steady-state thermodynamic coefficient, the upper boundary function of power consumption, the lower boundary function of power consumption, and the dynamic time constant, performs a comprehensive demand response assessment of the current air conditioning power and the current outdoor temperature to obtain the reduction potential, the increase potential, and the response duration. This includes: step 61, based on the upper boundary function of power consumption and the lower boundary function of power consumption, performing a real-time state assessment and power potential calculation of the current air conditioning power and the current outdoor temperature to obtain the reduction potential and the increase potential; step 62, based on the reduction potential and the steady-state thermodynamic coefficient, estimating the equivalent steady-state temperature rise to obtain the equivalent steady-state temperature rise; and step 63, based on the dynamic time constant, solving for the dynamic response duration of the equivalent steady-state temperature rise to obtain the response duration.

[0070] In the above exemplary operation, the specific process of step 6 is as follows: First, it is necessary to obtain the instantaneous input data required for real-time evaluation. The current air conditioning power is obtained by decomposing the user's total power in real time through the Non-Intrusive Load Monitoring (NILM) module. This module runs continuously, providing the latest air conditioning power consumption value for evaluation. The current outdoor temperature is obtained in real time from the public meteorological service through the API interface, reflecting the external environmental conditions at the time of evaluation. At the same time, a key preset parameter is also required, namely the user comfort temperature rise tolerance threshold. This value represents the maximum acceptable increase in indoor temperature when the user participates in demand response, and is a key constraint to ensure user experience and avoid complaints. This threshold is jointly agreed upon by the load aggregator and the user when signing the demand response service agreement, or a standard value is set according to different service levels, a typical setting value is 1.5 degrees Celsius. In this embodiment, based on the model parameters identified in the previous steps, these parameters are: steady-state thermodynamic coefficient =0.15 kW / °C; Upper boundary function of power consumption Power consumption lower boundary function The dynamic time constant τ = 5.0 minutes. In a specific evaluation scenario: at 2:30 PM on a summer afternoon, the current outdoor temperature is acquired in real time. 33.0 degrees Celsius, and the current air conditioning power output by the NILM module. 1.720 kW. A user comfort temperature rise tolerance threshold of 1.5 degrees Celsius is set.

[0071] Step 61: First, substitute the current outdoor temperature of 33.0 degrees Celsius into the identified power consumption boundary function to calculate the theoretical lower and upper limits of power consumption for stable operation of the air conditioner at this temperature: upper power consumption boundary value. (33.0) = 0.15 × 33.0 - 3.216 = 1.734 kW; Lower boundary value of power consumption (33.0) = 0.15 × 33.0 - 3.264 = 1.686 kW. These two values ​​define the operating power of the air conditioner to maintain indoor comfort at an outdoor temperature of 33.0 degrees Celsius within the range of [1.686, 1.734] kW. Next, the current real-time power will be... 1.720 kW was compared with these two boundary values ​​to calculate the adjustable potential. Reduceable potential. Defined as the difference between the current power and the lower boundary of power consumption, it represents the maximum amount of power that can be reduced without deviating from the stable operating range. Its calculation formula is: Substitute the values: =max(0,1.720-1.686)=0.034 kW. Potential for increased power. Defined as the difference between the upper limit of power consumption and the current power, it represents the maximum amount of power that can be increased when needed (such as in response to a power grid increase command). Its calculation formula is: Substitute the values: =max(0,1.734-1.720)=0.014 kW. The final reduction potential is 34 kW, and the increase potential is 14 kW.

[0072] Step 62: This step aims to translate the potential for power reduction into the potential for changes in indoor temperature in the physical world. According to the principles of steady-state thermodynamics, the cooling power of an air conditioner is approximately proportional to the temperature difference between indoors and outdoors. The steady-state thermodynamic coefficient we identified in Step 4... This is the macroscopic quantity that describes this relationship. Therefore, when the air conditioner power is reduced... This will inevitably lead to the indoor temperature reaching a new, higher equilibrium point after a sufficiently long period of time. This final temperature rise is called the equivalent steady-state temperature rise. It represents the maximum theoretical increase in indoor temperature that can occur under a specific power reduction. The formula for its calculation is: The physical meaning of this formula lies in the amount of power reduction. Through thermal coefficient This was converted into the amount of reduction in the indoor-outdoor temperature difference that it could compensate for, i.e., the increase in indoor temperature. The reduction potential calculated in step 61 was then used to... =0.034 kW and Substituting = 0.15 kW / °C: =0.034 / 0.15≈0.227 degrees Celsius, which means the equivalent steady-state temperature rise is 0.227 degrees Celsius. In other words, if the air conditioner power is reduced by 34 watts and it is run indefinitely, the indoor temperature will eventually be 0.227 degrees Celsius higher than before.

[0073] Step 63: Calculate the longest duration that the calculated power reduction can be sustained while maintaining user comfort. First, perform a validity check. Compare this to the user comfort temperature rise tolerance threshold. The 1.5°C temperature rise is equivalent to the theoretical maximum temperature rise of 0.227°C calculated in the previous step, which is the same as the steady-state temperature rise. In this scenario, the user comfort temperature rise tolerance threshold is greater than 0.227°C. This means that even if the user participates in demand response indefinitely with reduced power, the rise in indoor temperature (maximum 0.227°C) will never reach a level that the user cannot tolerate (1.5°C). Therefore, in this case, the response duration is theoretically infinite. In engineering practice, this can be set as a maximum allowable response time, such as 4 hours.

[0074] To illustrate the calculation process more fully, consider another scenario: at a certain moment, the calculated reduction potential is relatively large. =0.400 kW. At this point, recalculate step 62 to obtain the new equivalent steady-state temperature rise. =0.400 / 0.15≈2.67 degrees Celsius. In this new scenario, =1.5 degrees Celsius =2.67 degrees Celsius. This means that if the response process continues long enough, the rise in indoor temperature will inevitably exceed the user's comfort threshold. Therefore, it is necessary to calculate the exact temperature rise. The required time. The dynamic change of indoor temperature during the response process follows the exponential growth law of a first-order system, and its functional form is: This formula describes the actual temperature rise. How it evolves with time t, taking τ as a time constant, gradually approaches its theoretical upper limit. The goal is to solve the problem when... Exactly equal to Let t be the time. By inversely solving the above equation, we can obtain the formula for calculating the response duration: Substituting the values ​​from the new scenario: τ = 5.0 minutes, =1.5 degrees Celsius =2.67 degrees Celsius =-5.0*ln(1-(1.5 / 2.67)≈4.125 minutes. The final output response duration is 4.125 minutes, or approximately 4 minutes and 8 seconds. This set of quantitative indicators provides accurate, real-time data support for power grid dispatching decisions, taking into account user experience.

[0075] In summary, the IoT-based method for assessing air conditioning load demand response potential, based on embodiments of this application, is explained. It avoids the ill-posed problem of completely identifying all physical parameters in traditional first-order equivalent thermal parameter models. To address this, a new data-driven paradigm for modeling the macroscopic behavior of air conditioning loads is proposed. This paradigm decomposes the air conditioning operating state into steady-state and dynamic intervals and directly identifies key macroscopic characteristic parameters describing the system boundary and inertia from the data, namely the steady-state thermodynamic coefficient, power consumption boundary function, and dynamic time constant. This method cleverly bypasses the coupled identification of multiple internal states and parameters such as indoor temperature, equivalent thermal resistance, and equivalent heat capacity, thereby fundamentally solving the ill-posed problem caused by parameter redundancy in the background technology. Finally, in a non-intrusive scenario relying solely on user-aggregated electricity consumption and outdoor temperature data, these precisely identifiable macroscopic parameters are used to assess the response potential, achieving accuracy and robustness in assessing the air conditioning load demand response potential.

Claims

1. A method for assessing the demand response potential of air conditioning loads based on the Internet of Things, characterized in that, include: Acquire the raw aggregated power data stream and the raw outdoor temperature data stream; Data preprocessing and air conditioning load decomposition are performed on the raw aggregated power data stream and the raw outdoor temperature data stream to obtain the air conditioning power sequence and the outdoor temperature sequence. The air conditioning power sequence and outdoor temperature sequence are divided into operating state intervals to obtain a set of steady-state data points and a set of dynamic data segments. Steady-state parameters and dynamic boundaries are identified from the set of steady-state data points to obtain the steady-state thermodynamic coefficient, upper boundary function of power consumption, and lower boundary function of power consumption. Dynamic time constants are obtained by identifying dynamic time constants in a set of dynamic data segments. Based on the steady-state thermodynamic coefficient, the upper boundary function of power consumption, the lower boundary function of power consumption, and the dynamic time constant, a comprehensive assessment of demand response is conducted on the current air conditioning power and the current outdoor temperature to obtain the potential for reduction, the potential for increase, and the response duration.

2. The method for assessing the air conditioning load demand response potential based on the Internet of Things according to claim 1, characterized in that, The air conditioning power sequence and outdoor temperature sequence are divided into operating state intervals to obtain a set of steady-state data points and a set of dynamic data segments, including: The status label sequence is determined based on the air conditioner power sequence and the outdoor temperature sequence; Based on the state label sequence, the air conditioning power sequence and outdoor temperature sequence are aggregated and output sets are generated to obtain a set of steady-state data points and a set of dynamic data segments.

3. The method for assessing the air conditioning load demand response potential based on the Internet of Things according to claim 2, characterized in that, Based on the air conditioner power sequence and the outdoor temperature sequence, a status label sequence is determined, including: Dynamic thermal sensitivity coefficient adaptive estimation is performed on the air conditioning power sequence and the outdoor temperature sequence to obtain the dynamic thermal sensitivity coefficient sequence. Based on the dynamic thermal sensitivity coefficient of the previous moment, the expected power change and residual power change are calculated on the air conditioning power sequence and the outdoor temperature sequence to obtain the residual power change sequence. The residual index is calculated and the state is marked based on the residual power change sequence and the air conditioning power sequence to obtain the state label sequence.

4. The method for assessing the potential of air conditioning load demand response based on the Internet of Things according to claim 3, characterized in that, To obtain a dynamic thermal sensitivity coefficient sequence, adaptive estimation of the dynamic thermal sensitivity coefficient is performed on the air conditioning power sequence and the outdoor temperature sequence, including: adaptive estimation of the dynamic thermal sensitivity coefficient of the air conditioning power sequence and the outdoor temperature sequence using the following formula: [Formula omitted for brevity] in, for The dynamic thermal sensitivity coefficient at any given time. for Air conditioner power at all times for outdoor temperature at any time It is a smoothing factor, with a value between 0 and 1.

5. The method for assessing the air conditioning load demand response potential based on the Internet of Things according to claim 3, characterized in that, Based on the dynamic thermal sensitivity coefficient of the previous moment, the expected power change and residual power change are calculated on the air conditioning power sequence and the outdoor temperature sequence to obtain the residual power change sequence. This includes: calculating the expected power change and residual power change on the air conditioning power sequence and the outdoor temperature sequence using the following formula: in, for The expected change in power at any given time. for The dynamic thermal sensitivity coefficient at any given time. for outdoor temperature at any time for The change in residual power at time t. for The air conditioner power at any given time.

6. The method for assessing air conditioning load demand response potential based on the Internet of Things according to claim 1, characterized in that, Dynamic time constants are identified from a set of dynamic data segments to obtain the dynamic time constants, including: Effective dynamic segment selection is performed on the dynamic data segment set and the steady-state data point set to obtain a set of fitable segments; A model transformation and linear regression fitting are performed on the set of fitable segments to obtain a list of candidate time constants; The candidate time constant list is aggregated to obtain the dynamic time constant.

7. The method for assessing air conditioning load demand response potential based on the Internet of Things according to claim 1, characterized in that, Based on the steady-state thermodynamic coefficient, upper boundary function of power consumption, lower boundary function of power consumption, and dynamic time constant, a comprehensive demand response assessment is performed on the current air conditioning power and current outdoor temperature to obtain the potential for reduction, the potential for increase, and the response duration, including: Based on the upper and lower boundary functions of power consumption, real-time state assessment and power potential calculation are performed on the current air conditioning power and the current outdoor temperature to obtain the potential for reduction and the potential for increase. Equivalent steady-state temperature rise is estimated based on the reduction potential and steady-state thermodynamic coefficient to obtain the equivalent steady-state temperature rise. Based on the dynamic time constant, the dynamic response duration of the equivalent steady-state temperature rise is solved to obtain the response duration.

Images