Non-intrusive air conditioner load monitoring method based on Kalman filtering

Abstract

The invention relates to the technical field of air conditioner load monitoring, and particularly discloses a non-intrusive air conditioner load monitoring method based on Kalman filtering, which utilizes a non-intrusive terminal to collect historical load data of an air conditioner to form a one-dimensional time sequence. By setting delay time and embedding dimension, phase space reconstruction is carried out on the time sequence, and a high-dimensional space trajectory is obtained. Performing sliding sampling on the track by adopting a preset time window, and calculating a local average value of each sampling window to form a local average value sequence; and predicting the air conditioner load at the next moment based on the sequence, calculating Kalman gain in combination with actual measurement data, and correcting a predicted value to obtain an optimized prediction result. In this way, the air conditioner load can be estimated more accurately, prediction errors can be effectively corrected, and the accuracy and real-time performance of load recognition are improved.

Description

Technical Field

[0001] The present application relates to the technical field of air conditioning load monitoring, and more specifically, to a non-intrusive air conditioning load monitoring method based on Kalman filtering. Background Art

[0002] In the past decade, smart thermostats and advanced cooling and heating technologies have been widely developed and deployed to achieve energy conservation and provide grid services. With the rapid improvement of the digitalization and intelligence level of the demand side in the construction of new power systems, and the gradual opening of the power sales market, the interaction between power users and the power grid has become more frequent, and the volatility and uncertainty of loads have increased significantly. Therefore, accurate identification of the operating status of air-conditioning loads and analysis of the user groups with adjustable potential can improve energy efficiency and grid service quality.

[0003] Due to the wide variety of equipment in large commercial buildings and the extremely complex noise generated, it is extremely difficult to monitor and perceive the air conditioning load status. Current mathematical optimization methods for load identification seek the optimal solution in the feature space, usually based on the additivity assumption of load characteristics. These methods perform well when dealing with simple or small loads, but may have limitations when facing complex and diverse load situations because they require clear feature models for optimization. In addition, when applying conventional deep learning models to process air conditioning load data, they are often interfered by noise, which increases the complexity of the data and makes it difficult for the model to learn effective features, thereby affecting the accuracy of load identification.

[0004] Therefore, an optimized air conditioning load monitoring solution is desired. Summary of the invention

[0005] The present application provides a non-intrusive air conditioning load monitoring method based on Kalman filtering, which can more accurately estimate the air conditioning load and effectively correct the prediction error, thereby improving the accuracy and real-time performance of load identification.

[0006] In a first aspect, a non-intrusive air conditioning load monitoring method based on Kalman filtering is provided, comprising:

[0007] Obtain the historical load data of the monitored air conditioner through the non-intrusive measurement terminal;

[0008] Constructing the historical load data of the monitored air conditioner into a one-dimensional time series of air conditioner load;

[0009] Based on the delay time and the embedding dimension, the one-dimensional time series of the air-conditioning load is reconstructed in phase space to obtain a high-dimensional space trajectory of the air-conditioning load series;

[0010] Sliding sampling is performed on the high-dimensional space trajectory of the air-conditioning load sequence based on a predetermined time window to obtain a sequence of sampling windows of the air-conditioning load sequence trajectory;

[0011] Calculating the local average of each air conditioning load sequence trajectory sampling window in the sequence of air conditioning load sequence trajectory sampling windows to obtain a sequence of local average values;

[0012] Based on the sequence of the local average values, calculating the predicted value of the air conditioning load operation at the next moment;

[0013] Calculating the Kalman gain between the predicted value of the air conditioning load operation at the next moment and the actual measurement data;

[0014] The air conditioning load operation prediction value at the next moment is corrected based on the Kalman gain to obtain an optimized air conditioning load operation prediction value at the next moment.

[0015] The present application provides a non-intrusive air conditioning load monitoring method based on Kalman filtering, which uses a non-intrusive terminal to collect historical air conditioning load data to form a one-dimensional time series. By setting the delay time and embedding dimension, the time series is reconstructed in phase space to obtain a high-dimensional space trajectory. The trajectory is sampled by sliding using a predetermined time window, and the local average of each sampling window is calculated to form a local average sequence. Based on this sequence, the air conditioning load at the next moment is predicted, and the Kalman gain is calculated in combination with the actual measurement data, and the predicted value is corrected to obtain an optimized prediction result. In this way, the air conditioning load can be estimated more accurately and the prediction error can be effectively corrected, thereby improving the accuracy and real-time performance of load identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0017] Figure 1 This is a schematic flowchart of a non-intrusive air conditioning load monitoring method based on Kalman filtering according to Embodiment 1 of the present application.

[0018] Figure 2 Schematic diagram of a non-intrusive air conditioning load monitoring method based on Kalman filtering according to Embodiment 1 of the present application.

[0019] Figure 3 This is a schematic flow chart of calculating the predicted value of the air-conditioning load operation at the next moment based on the sequence of the local average values ​​in the second embodiment of the present application.

[0020] Figure 4This is a data flow diagram for calculating the predicted value of the air-conditioning load operation at the next moment based on the sequence of the local average values ​​in Example 2 of the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0022] Example 1

[0023] It should be understood that non-intrusive load monitoring collects power consumption data of equipment through smart meters, sensors and data acquisition systems. These data usually include basic parameters such as voltage, current, power factor, etc., which can be used to infer the operating status of the equipment. Kalman filter (KF) is an efficient recursive filter that can estimate the state of a dynamic system from a series of incomplete and noisy measurements. The core idea of ​​Kalman filter is to use the system model and observation data to optimally estimate the system state in a recursive manner in the sense of minimum mean square error. This filter is suitable for linear systems, and when both system noise and observation noise are Gaussian white noise, Kalman filter can give the best estimation result. In the application of air conditioning load prediction, Kalman filter plays an important role, which is mainly reflected in the following aspects: First, air conditioning load data is often affected by various factors, such as power grid fluctuations, ambient temperature changes, etc., resulting in a large amount of noise in the data. Kalman filter can effectively filter out these noises and improve the accuracy of prediction. Second, air conditioning load changes over time and has strong dynamics and uncertainty. Kalman filtering can update the predicted value in real time according to the new measurement data, and dynamically adjust the prediction model to make the prediction closer to the actual situation. In air conditioning load prediction, it is difficult to establish an accurate physical model due to the complex and changeable load conditions. However, Kalman filtering does not need to rely on a clear feature model. It can automatically identify and learn load characteristics through self-learning and adaptive mechanisms to improve the accuracy of load identification. Kalman filtering can capture the different time scale characteristics of load data, including long-term trends and short-term fluctuations, which helps to improve the comprehensiveness and accuracy of load prediction. In the process of air conditioning load prediction based on Kalman filtering, the load value at the next moment is first predicted based on historical data, and then the Kalman gain is calculated in combination with the actual measurement data to correct the predicted value to obtain an optimized prediction result. This process effectively improves the accuracy and reliability of the prediction.

[0024] Based on this, this application proposes a non-intrusive air conditioning load monitoring method based on Kalman filtering. Figure 1 FIG. 1 is a schematic flow chart of a non-intrusive air conditioning load monitoring method based on Kalman filtering according to an embodiment of the present application. Figure 1 As shown, the non-intrusive air conditioning load monitoring method based on Kalman filtering includes: S1, obtaining historical load data of the monitored air conditioner through a non-intrusive measurement terminal; S2, constructing the historical load data of the monitored air conditioner into a one-dimensional time series of air conditioning load; S3, based on the delay time and the embedding dimension, reconstructing the one-dimensional time series of the air conditioning load in phase space to obtain a high-dimensional space trajectory of the air conditioning load sequence; S4, sliding sampling the high-dimensional space trajectory of the air conditioning load sequence based on a predetermined time window to obtain a sequence of air conditioning load sequence trajectory sampling windows; S5, calculating the local average values ​​of each air conditioning load sequence trajectory sampling window in the sequence of air conditioning load sequence trajectory sampling windows to obtain a sequence of local average values; S6, calculating the air conditioning load operation prediction value at the next moment based on the sequence of local average values; S7, calculating the Kalman gain between the air conditioning load operation prediction value at the next moment and the actual measurement data; S8, correcting the air conditioning load operation prediction value at the next moment based on the Kalman gain to obtain an optimized air conditioning load operation prediction value at the next moment. In this way, there is no need to rely on clear feature models, and complex and diverse load conditions can be handled more flexibly, and load characteristics can be automatically learned and identified, thereby improving the accuracy of load identification. At the same time, by capturing load characteristics at different scales, key features can be obtained, effectively suppressing the impact of noise on load identification, thereby effectively solving the limitations of conventional deep learning models in dealing with noise interference and complex load data, and improving the noise resistance and prediction accuracy of load identification.

[0025] In the above-mentioned non-invasive air conditioning load monitoring method based on Kalman filtering, the S1 obtains the historical load data of the monitored air conditioner through a non-invasive measurement terminal. It should be understood that the non-invasive measurement terminal will not cause any physical changes to the existing electrical system, avoiding the safety risks and system failures that may be caused by traditional invasive measurements, and ensuring the normal operation of the system. At the same time, the non-invasive measurement terminal usually only needs to be installed on the power line, without the need to disassemble or modify the air conditioning equipment, which greatly simplifies the installation process and reduces installation costs and time. Because it is non-invasive, once the installation is completed, subsequent maintenance and upgrades are relatively easy, and there is no need to shut down or interrupt the operation of the air conditioner. Non-invasive measurement will not directly access the internal circuit of the air conditioner, reducing the risk of data leakage and improving data security.

[0026] Optionally, in one embodiment of the present application, the historical load data of the monitored air conditioner is obtained through a non-invasive measurement terminal, including: the non-invasive measurement terminal measures the voltage and current signals on the power line through an induction coil or a clamp-on current transformer. The terminal device calculates the power factor by measuring the phase difference between the voltage and the current. The instantaneous power is calculated based on the voltage, current and power factor. The measured voltage, current, power factor and instantaneous power data are recorded according to the timestamp to form historical load data. The measurement terminal temporarily stores the collected data in a local memory. The data is regularly transmitted to the central processing system or cloud server through a wireless or wired communication module. At the same time, it is necessary to ensure the timeliness and integrity of data transmission to avoid data loss or delay.

[0027] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S2 constructs the historical load data of the monitored air conditioning into a one-dimensional time series of the air conditioning load. It should be understood that by constructing the historical load data into a one-dimensional time series, the data can be arranged in chronological order, which is convenient for time series analysis. Time series data can clearly show the changing trend and periodic characteristics of the load, which helps to identify the laws and patterns of the load.

[0028] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S3, based on the delay time and the embedding dimension, performs phase space reconstruction on the one-dimensional time series of the air conditioning load to obtain the high-dimensional space trajectory of the air conditioning load series. It should be understood that the delayed coordinate embedding method is applied to reconstruct the phase space of the time series data. Specifically, the optimal delay time is determined by the mutual information method, and the embedding dimension is determined by the improved false nearest neighbor method. These parameters are used to reconstruct the time series to form a multi-dimensional state space.

[0029] Optionally, in one embodiment of the present application, based on the delay time and the embedding dimension, the phase space of the one-dimensional time series of the air-conditioning load is reconstructed to obtain a high-dimensional space trajectory of the air-conditioning load sequence, including: determining an initial delay time, and delaying the one-dimensional time series of the air-conditioning load based on the initial delay time to obtain a one-dimensional delayed time series of the air-conditioning load; calculating the mutual information between the one-dimensional time series of the air-conditioning load and the one-dimensional delayed time series of the air-conditioning load, and selecting the first minimum value as the delay time based on the mutual information; constructing a phase space based on the one-dimensional time series of the air-conditioning load and the delay time, and determining the embedding dimension based on the false adjacent points of the phase space; reconstructing the phase space of the one-dimensional time series of the air-conditioning load based on the delay time and the embedding dimension to obtain a high-dimensional space trajectory of the air-conditioning load sequence.

[0030] Specifically, in the embodiment of the present application, two time series A and B of power load are defined, where B is obtained after A is delayed by t. The mutual information M(t) between the two time series is a function of the delay time, and the magnitude of the mutual information represents the certainty magnitude of time series B when time series A is known. It can be expressed as follows:

[0031] M(t)=H(A)+H(B)-H(A,B)

[0032] Among them, H(A), H(B), and H(A,B) are the information entropy and joint information entropy of their respective time series respectively:

[0033]

[0034] Among them, P is the corresponding probability.

[0035] The first minimum value of M(t) represents the maximum possibility that the two time series are uncorrelated, so the first minimum value of M(t) is used as the final delay time.

[0036] Based on the time series and delay time of power load, the phase space can be constructed as follows:

[0037] y i (d)=(x(i),…,x(i+(d-1t)))

[0038] Here, d is the embedding dimension.

[0039] False adjacent points refer to two non-adjacent points in a high-dimensional phase space that are projected onto a one-dimensional space and sometimes become adjacent points. The present invention uses the maximum model number to measure the distance between two points.

[0040]

[0041] Through iterative calculation, when the dimension d is greater than a certain value d c , No longer changes, select d c +1 for the embedding dimension.

[0042] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S4, S5 and S6 perform sliding sampling on the high-dimensional space trajectory of the air conditioning load sequence based on a predetermined time window to obtain a sequence of air conditioning load sequence trajectory sampling windows, and then calculate the local average values ​​of each air conditioning load sequence trajectory sampling window in the sequence of air conditioning load sequence trajectory sampling windows to obtain a sequence of local average values, and calculate the air conditioning load operation prediction value at the next moment based on the sequence of local average values. It should be understood that a suitable time window is selected, and the local average values ​​of each time point in the window are calculated based on the reconstructed state space data, and the most likely load at the next moment is predicted by analyzing these local average values.

[0043] Optionally, in one embodiment of the present application, 2n+1 Sigma points are selected from the load forecast value to approximately describe the distribution of the state variable, ensuring that the mean and covariance of the state can be more accurately reflected under nonlinear transformation. According to the local average prediction method, each Sigma point in the point set can be processed to obtain the load operation forecast value of the corresponding next moment.

[0044] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S7 calculates the Kalman gain between the predicted value of the air conditioning load operation at the next moment and the actual measurement data. Optionally, in one embodiment of the present application, the actual measurement data is compared with the prediction result to calculate the covariance and Kalman gain using the following formula:

[0045]

[0046] Among them, K k is the Kalman gain; is the cross covariance of state and power; is the power covariance; is the state prediction value; Estimated power.

[0047] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S8 corrects the predicted value of the air conditioning load operation at the next moment based on the Kalman gain to obtain an optimized predicted value of the air conditioning load operation at the next moment. Further, the following formula is executed for prediction correction:

[0048]

[0049] Optionally, in one embodiment of the present application, the non-intrusive air conditioning load monitoring method based on Kalman filtering can also be used as follows: Figure 2The method is described as follows: first, the power data at time T is obtained, then the phase space is reconstructed, then the Sigma point is selected, then the composite prediction at time T+1 is performed, then the Coleman gain is calculated, and finally, the prediction correction is performed to obtain the operating state at time T+1.

[0050] In the embodiment of the present application, the evaluation index F for the load identification result is maxro , whose expression is as follows:

[0051]

[0052]

[0053] The non-intrusive air conditioning load monitoring method based on Kalman filtering in the above embodiment is used to identify the load. macro It is 0.958.

[0054] Embodiment 2

[0055] In particular, in step S6 of the first embodiment, based on the sequence of the local average values, the predicted value of the air-conditioning load operation at the next moment is calculated. The technical concept of the second embodiment of the present application is to use artificial intelligence-based data analysis and processing technology to perform global trend recognition and local trend recognition on the sequence of the local average values, and then perform principal component gradient attenuation propagation on the local time series pattern of the air-conditioning load obtained by the local trend recognition, so as to intelligently predict the air-conditioning load operation value at the next moment based on the multi-scale time series pattern between the first-scale air-conditioning load time series pattern obtained by the global trend recognition and the second-scale air-conditioning load time series pattern obtained by the attenuation propagation. In this way, there is no need to rely on a clear feature model, and complex and diverse load conditions can be handled more flexibly, and load characteristics can be automatically learned and identified, thereby improving the accuracy of load identification. At the same time, by capturing load characteristics at different scales, key features can be obtained, and the influence of noise on load identification can be effectively suppressed, thereby effectively solving the limitations of conventional deep learning models in processing noise interference and complex load data, and improving the noise resistance and prediction accuracy of load identification.

[0056] Figure 3 This is a schematic flow chart of calculating the predicted value of the air-conditioning load operation at the next moment based on the sequence of the local average values ​​in the second embodiment of the present application. Figure 4 This is a data flow diagram for calculating the predicted value of the air conditioning load operation at the next moment based on the sequence of the local average values ​​in the second embodiment of the present application. Figure 3 and Figure 4As shown, based on the sequence of local average values, the air-conditioning load operation prediction value at the next moment is calculated, including: S61, performing global trend recognition on the sequence of local average values ​​to obtain a first-scale air-conditioning load time series pattern implicit representation vector; S62, performing local trend recognition on the sequence of local average values ​​to obtain a sequence of air-conditioning load local time series pattern implicit representation vectors; S63, performing dynamic time series principal component gradient attenuation propagation on the sequence of air-conditioning load local time series pattern implicit representation vectors to obtain a second-scale air-conditioning load time series pattern implicit representation vector; S64, fusing the first-scale air-conditioning load time series pattern implicit representation vector and the second-scale air-conditioning load time series pattern implicit representation vector to obtain a multi-scale air-conditioning load time series pattern implicit representation vector; S65, based on the multi-scale air-conditioning load time series pattern implicit representation vector, obtaining the air-conditioning load operation prediction value at the next moment.

[0057] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S61 performs global trend recognition on the sequence of local average values ​​to obtain an implicit representation vector of the time series pattern of the first-scale air conditioning load. It should be understood that, considering that the sequence of local average values ​​has already included the changes in the air conditioning load within a certain period of time, there is a mutual temporal correlation between these different time spans, and in order to effectively capture the long-term temporal trends and periodic changes in the air conditioning load data to extract higher-level global features, in the technical solution of the present application, the sequence of local average values ​​is input into the global trend identifier based on the forward LSTM model to obtain the implicit representation vector of the time series pattern of the first-scale air conditioning load. That is, the forward LSTM model is a special recurrent neural network (RNN) that is particularly good at processing correlation data with long time intervals, so that it can capture the long-term dependencies in the local data to better understand the global pattern of the air conditioning load.

[0058] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S62 performs local trend recognition on the sequence of local average values ​​to obtain a sequence of implicit representation vectors of the local time series pattern of the air conditioning load. It should be understood that in order to capture and mine the local, short-term air conditioning load time series characteristics in the sequence of local average values, in the technical solution of the present application, the sequence of local average values ​​is input into a local trend identifier based on a 1D-CNN model to slide through different filters (convolution kernels), extract local patterns, and obtain a sequence of implicit representation vectors of the local time series pattern of the air conditioning load, and these local features reflect the change pattern of the air conditioning load in a relatively short period of time (such as a few minutes or hours), which is very important for understanding short-term load fluctuations.

[0059] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S63 performs dynamic time series principal component gradient attenuation propagation on the sequence of implicit representation vectors of the local time series pattern of the air conditioning load to obtain the second-scale air conditioning load time series pattern implicit representation vector. It should be understood that, considering that different local time series patterns of air conditioning loads in the sequence of implicit representation vectors of the local time series pattern of the air conditioning load have different importance on the entire time scale, that is, as time goes by, the historical data close to the current time point has a more significant impact on the current state. Therefore, in order to gradually weaken the influence of old features and enhance the role of new features, so that the model can better adapt to the dynamic changes of time series data, in the technical solution of the present application, the sequence of implicit representation vectors of the local time series pattern of the air conditioning load is subjected to dynamic time series principal component gradient attenuation propagation to obtain the second-scale air conditioning load time series pattern implicit representation vector.

[0060] Optionally, in one embodiment of the present application, the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load is subjected to dynamic time series principal component gradient attenuation propagation to obtain the second-scale air-conditioning load time series pattern implicit representation vector, including: extracting the sequence of the current air-conditioning load local time series pattern implicit representation vector and the sequence of the historical air-conditioning load local time series pattern implicit representation vector from the sequence of the local time series pattern implicit representation vector of the air-conditioning load, and performing principal component extraction to obtain the sequence of the current air-conditioning load local time series principal component representation vector and the sequence of the historical air-conditioning load local time series principal component representation vector; calculating the time series propagation attenuation entropy absolute factor of each sequence of the historical air-conditioning load local time series principal component representation vector in the sequence of the historical air-conditioning load local time series principal component representation vector relative to the current air-conditioning load local time series principal component representation vector to obtain the sequence of the air-conditioning load time series propagation attenuation entropy absolute factor; based on the sequence of the air-conditioning load time series propagation attenuation entropy absolute factor, dynamically propagating the current air-conditioning load local time series principal component representation vector and the sequence of the historical air-conditioning load local time series principal component representation vector to obtain the second-scale air-conditioning load time series pattern implicit representation vector.

[0061] Optionally, in one embodiment of the present application, the sequence of the implicit representation vector of the local timing pattern of the current air-conditioning load and the sequence of the implicit representation vector of the local timing pattern of the historical air-conditioning load are extracted from the sequence of the implicit representation vector of the local timing pattern of the air-conditioning load, and the principal component extraction is performed to obtain the sequence of the principal component representation vector of the local timing pattern of the current air-conditioning load and the principal component representation vector of the local timing pattern of the historical air-conditioning load, including: extracting the implicit representation vector of the local timing pattern of the current air-conditioning load from the sequence of the implicit representation vector of the local timing pattern of the air-conditioning load, and performing principal component extraction on the implicit representation vector of the local timing pattern of the current air-conditioning load based on The principal component of the eigenvalue is extracted to obtain the principal component representation vector of the current local time series of the air-conditioning load; other local time series pattern implicit representation vectors of the air-conditioning load in the sequence of the local time series pattern implicit representation vector of the air-conditioning load are defined as the local time series pattern implicit representation vectors of the historical air-conditioning load to obtain the sequence of the local time series pattern implicit representation vectors of the historical air-conditioning load; and the principal component of each local time series pattern implicit representation vector of the historical air-conditioning load in the sequence of the local time series pattern implicit representation vector of the historical air-conditioning load is extracted based on the eigenvalue to obtain the sequence of the local time series principal component representation vectors of the historical air-conditioning load.

[0062] Optionally, in one embodiment of the present application, the sequence of each historical air-conditioning load local time series principal component representation vector in the sequence of the historical air-conditioning load local time series principal component representation vector is calculated relative to the current air-conditioning load local time series principal component representation vector to obtain a sequence of air-conditioning load time series propagation attenuation entropy absolute factors, including: calculating the positional division of the corresponding positions of the historical air-conditioning load local time series principal component representation vector and the current air-conditioning load local time series principal component representation vector to obtain the air-conditioning load local time series principal component attenuation vector; calculating the base-two logarithmic function value of the absolute value of each eigenvalue of the air-conditioning load local time series principal component attenuation vector to obtain the air-conditioning load local time series principal component attenuation logarithmic vector; calculating the positional dot product of the historical air-conditioning load local time series principal component representation vector and the air-conditioning load local time series principal component attenuation logarithmic vector, and performing positional multiplication on the obtained dot product vector to obtain the air-conditioning load time series propagation attenuation value; calculating an exponential function with the natural constant e as the base, with the air-conditioning load time series propagation attenuation value as the exponent to obtain the air-conditioning load time series propagation attenuation entropy absolute factor.

[0063] Optionally, in one embodiment of the present application, based on the sequence of absolute factors of the air-conditioning load time series propagation attenuation entropy, the sequence of the current air-conditioning load local time series principal component representation vector and the sequence of the historical air-conditioning load local time series principal component representation vector are dynamically propagated to obtain the second-scale air-conditioning load time series pattern implicit representation vector, including: based on the time span between each historical air-conditioning load local time series principal component representation vector in the sequence of the historical air-conditioning load local time series principal component representation vector and the current air-conditioning load local time series principal component representation vector, each air-conditioning load time series propagation attenuation entropy absolute factor in the sequence of the air-conditioning load time series propagation attenuation entropy absolute factor is modulated in the time dimension to obtain the space The sequence of load time series span modulation propagation attenuation entropy factors is input into the information transmission screening module based on the gating function to obtain the sequence of air-conditioning load time series span modulation propagation attenuation weights; based on the sequence of air-conditioning load time series span modulation propagation attenuation weights, the weighted sum of the sequence of historical air-conditioning load local time series principal component representation vectors is calculated to obtain the historical air-conditioning load local time series principal component significant transfer aggregation representation vector; the positional sum of the historical air-conditioning load local time series principal component significant transfer aggregation representation vector and the current air-conditioning load local time series principal component representation vector is calculated to obtain the second-scale air-conditioning load time series pattern implicit representation vector.

[0064] Optionally, in one embodiment of the present application, based on the time span between each historical air-conditioning load local time series principal component representation vector in the sequence of the historical air-conditioning load local time series principal component representation vector and the current air-conditioning load local time series principal component representation vector, each air-conditioning load time series propagation attenuation entropy absolute factor in the sequence of the air-conditioning load time series propagation attenuation entropy absolute factors is modulated in the time dimension to obtain a sequence of air-conditioning load time series span modulation propagation attenuation entropy factors, including: subtracting the timestamp of the current air-conditioning load local time series principal component representation vector from the timestamp of the historical air-conditioning load local time series principal component representation vector and rounding down to obtain an air-conditioning load time span value; calculating an exponential function with the air-conditioning load time span value as the exponent with the natural constant e as the base to obtain an air-conditioning load time span modulation value; dividing the air-conditioning load time series propagation attenuation entropy absolute factor corresponding to the historical air-conditioning load local time series principal component representation vector by the air-conditioning load time span modulation value to obtain the air-conditioning load time series span modulation propagation attenuation entropy factor.

[0065] Optionally, in one embodiment of the present application, the sequence of the air-conditioning load timing span modulation propagation attenuation entropy factors is input into an information transmission screening module based on a gating function to obtain a sequence of air-conditioning load timing span modulation propagation attenuation weights, including: comparing each air-conditioning load timing span modulation propagation attenuation entropy factor in the sequence of the air-conditioning load timing span modulation propagation attenuation entropy factors with a predetermined threshold to obtain the sequence of the air-conditioning load timing span modulation propagation attenuation weights; wherein, in response to the air-conditioning load timing span modulation propagation attenuation entropy factor being greater than the predetermined threshold, the air-conditioning load timing span modulation propagation attenuation entropy factor greater than the predetermined threshold is input into a sigmoid function; in response to the air-conditioning load timing span modulation propagation attenuation entropy factor being less than or equal to the predetermined threshold, the air-conditioning load timing span modulation propagation attenuation entropy factor less than or equal to the predetermined threshold is set to zero.

[0066] In summary, in the embodiment of the present application, the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load is subjected to dynamic time series principal component gradient attenuation propagation to obtain the implicit representation vector of the second-scale air-conditioning load time series pattern, which can be expressed by the formula:

[0067] X={x1,x2,...,x i ,...,x p}

[0068] v i = PCA(x i )

[0069]

[0070] V={v1,v2,...,v i ,...,v p}

[0071]

[0072]

[0073] Where X is the sequence of implicit representation vectors of the local time series pattern of the air conditioning load, x1, x2, ..., x i ,...,x p are the first, second, ..., i-th, ..., p-th implicit representation vectors of the local time series pattern of the air conditioning load in the sequence of the implicit representation vectors of the local time series pattern of the air conditioning load, PCA(x i ) is x i Perform principal component extraction based on eigenvalues, v i For x i The corresponding local time series principal component representation vector of air conditioning load, x iT For x i The transposed vector of the local time series pattern of the air conditioning load is n, and n is the number of eigenvalues ​​in the implicit representation vector of the local time series pattern of the air conditioning load. i is x i The corresponding air conditioning load covariance matrix, U i is x i The corresponding air conditioning load principal component orthogonal matrix, U i T For U i The transposed matrix, v i1 ,v i2 ,…,v im is the principal component vector of each air conditioning load in the sequence of the principal component vector of air conditioning load, Λ i For x i The corresponding air conditioning load diagonal matrix, diag(λ i1 ,λ i2 ,…,λ im ) is the matrix whose diagonal elements are λ i1 ,λ i2 ,…,λ im The diagonal matrix of air conditioning load, λ i1 ,λ i2 ,…,λ im are the weight values ​​of the principal component vectors of each air conditioning load, arg k max(·) is the k value corresponding to the maximum value returned, j is the maximum approximate matching value, V is the sequence of the principal component representation vectors of the local time series of the air conditioning load, v1, v2, ..., v i ,...,v p are the first, second, ..., i-th, ..., p-th local time series principal component representation vectors of the air conditioning load in the sequence of the local time series principal component representation vectors of the air conditioning load, respectively, v is v i The eigenvalues ​​at each position in ps v p The eigenvalues ​​of each position in, L is v i The number of eigenvalues ​​in the equation, log represents the logarithmic function value with base 2, exp(·) represents the exponential function with the natural constant e as the base, d i v i and v p The absolute factor of the decay entropy of the air conditioning load time series propagation between p and t i Respectively represent the timestamps of the pth and ith local time series principal component representation vectors of air conditioning loads, For floor rounding operation, e (i→p) Yes i and v pThe air conditioning load time series span modulation propagation attenuation entropy factor between them, mask(·) is the masking process, sigmoid(·) is the sigmoid function, θ is the predetermined threshold, w i is the time series span modulation propagation attenuation weight of each air conditioning load in the sequence of the air conditioning load time series span modulation propagation attenuation weight, p is the number of vectors in the sequence of the air conditioning load local time series principal component representation vector, v c is the implicit representation vector of the second-scale air-conditioning load timing pattern.

[0074] Specifically, first, the sequence of the implicit representation vector of the local time series pattern of the current air-conditioning load and the sequence of the implicit representation vector of the local time series pattern of the historical air-conditioning load are extracted from the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load, and the principal component extraction is performed to obtain the sequence of the principal component representation vector of the local time series of the current air-conditioning load and the principal component representation vector of the local time series of the historical air-conditioning load. It should be understood that principal component analysis (PCA) is a commonly used dimensionality reduction technique that can convert high-dimensional data into low-dimensional data while retaining the main features of the data. Therefore, by performing PCA processing on the implicit representation vectors of each local time series pattern, the most important features can be extracted, noise and redundant information can be reduced, and subsequent analysis can be made more efficient and accurate, thereby capturing the changing patterns of features at different time points.

[0075] Next, in order to quantify the difference between the local time series principal component representation vector of the historical air-conditioning load and the local time series principal component representation vector of the current air-conditioning load, so as to intuitively understand the degree of change of the characteristics over time and thus better capture the dynamic characteristics of the load data, the absolute factor of the time series propagation attenuation entropy of each sequence of the local time series principal component representation vector of the historical air-conditioning load relative to the local time series principal component representation vector of the current air-conditioning load is calculated to capture the dynamic change degree of the air-conditioning load at the time point, and obtain the sequence of the absolute factor of the time series propagation attenuation entropy of the air-conditioning load.

[0076] Then, based on the sequence of absolute factors of the air-conditioning load time series propagation attenuation entropy, the sequence of the principal component representation vector of the current air-conditioning load local time series and the sequence of the principal component representation vector of the historical air-conditioning load local time series are dynamically propagated to obtain the implicit representation vector of the second-scale air-conditioning load time series pattern. Specifically, in order to better extract and analyze the long-term trends under different time spans, so as to ensure that even the differences between nodes that are far apart in time are considered to be within an acceptable range of variation, the time span factor is introduced to modulate the sequence of absolute factors of the air-conditioning load time series propagation attenuation entropy in the time dimension. Afterwards, the information transmission and screening module based on the gating function is used to perform feature screening on the sequence of the air-conditioning load time series span modulation propagation attenuation entropy factors to dynamically control the transmission and screening of information, highlight important features, suppress irrelevant or noise information, and obtain the sequence of air-conditioning load time series span modulation propagation attenuation weights. Then, based on each weight, the sequence of historical air-conditioning load local time series principal component representation vectors is weighted aggregated to effectively integrate the long-term time series trend and the short-term time series fluctuation, and the historical air-conditioning load local time series principal component significant transfer aggregation representation vector is obtained. Finally, the historical air-conditioning load local time series principal component significant transfer aggregation representation vector and the current air-conditioning load local time series principal component representation vector are summed by position to obtain the second-scale air-conditioning load time series pattern implicit representation vector, thereby more accurately reflecting the time series change pattern of the air-conditioning load.

[0077] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S64 is to fuse the implicit representation vector of the first-scale air conditioning load time series pattern and the implicit representation vector of the second-scale air conditioning load time series pattern to obtain the implicit representation vector of the multi-scale air conditioning load time series pattern. It should be understood that considering that the implicit representation vector of the first-scale air conditioning load time series pattern and the implicit representation vector of the second-scale air conditioning load time series pattern respectively express the time series characteristics obtained by aggregating the global trend and short-term trend of the air conditioning load. Based on this, in order to better capture the long-term trend and short-term fluctuation of the load and improve the accuracy of the prediction, in the technical solution of the present application, the implicit representation vector of the first-scale air conditioning load time series pattern and the implicit representation vector of the second-scale air conditioning load time series pattern are fused to obtain the implicit representation vector of the multi-scale air conditioning load time series pattern. In particular, in an embodiment of the present application, the implicit representation vector of the multi-scale air conditioning load time series pattern can be obtained by adding the implicit representation vector of the first-scale air conditioning load time series pattern and the implicit representation vector of the second-scale air conditioning load time series pattern by position or weighted fusion.

[0078] In the above-mentioned non-intrusive air conditioning load monitoring method based on Kalman filtering, the S65 obtains the predicted value of the air conditioning load operation at the next moment based on the implicit representation vector of the multi-scale air conditioning load time series pattern. Optionally, in one embodiment of the present application, the predicted value of the air conditioning load operation at the next moment is obtained based on the implicit representation vector of the multi-scale air conditioning load time series pattern, including: inputting the implicit representation vector of the multi-scale air conditioning load time series pattern into the air conditioning load prediction module based on the decoder to obtain the predicted value of the air conditioning load operation at the next moment. That is, the multi-scale air conditioning load time series pattern implicit representation vector obtained by fusing the implicit representation vector of the first scale air conditioning load time series pattern and the implicit representation vector of the second scale air conditioning load time series pattern is decoded and processed to intelligently predict the air conditioning load operation value at the next moment. In this way, there is no need to rely on a clear feature model, and complex and diverse load conditions can be handled more flexibly, and load characteristics can be automatically learned and identified, thereby improving the accuracy of load identification. At the same time, by capturing load characteristics at different scales, key features can be obtained, effectively suppressing the impact of noise on load identification, thereby effectively solving the limitations of conventional deep learning models in dealing with noise interference and complex load data, and improving the noise resistance and prediction accuracy of load identification.

[0079] Optionally, in one embodiment of the present application, the decoder is used to decode and regress the multi-scale air-conditioning load time series pattern implicit representation vector using the following decoding formula to obtain the air-conditioning load operation prediction value at the next moment, wherein the formula is Wherein, V is the implicit representation vector of the multi-scale air-conditioning load time series pattern, Y is the predicted value of the air-conditioning load operation at the next moment, W is the weight matrix of the decoder, Represents matrix multiplication.

[0080] Preferably, the multi-scale air-conditioning load time series pattern implicit representation vector is passed through a decoder-based air-conditioning load prediction module to obtain the air-conditioning load operation prediction value at the next moment, including:

[0081] Respectively calculating a first characteristic sum value of all eigenvalues ​​of the implicit representation vector of the multi-scale air-conditioning load time series pattern, a second characteristic sum value of all eigenvalues ​​of the implicit representation vector of the first-scale air-conditioning load time series pattern, and a third characteristic sum value of all eigenvalues ​​of the implicit representation vector of the second-scale air-conditioning load time series pattern;

[0082] After dividing the first characteristic sum value by the second characteristic sum value and the third characteristic sum value respectively, the square root of the sum of the squares of the two quotients is calculated to obtain an implicit distribution probability value of the multi-scale air conditioning load time series pattern;

[0083] Calculate the product of the first multi-scale air-conditioning load time series pattern implicit characteristic value of the multi-scale air-conditioning load time series pattern implicit representation vector and the multi-scale air-conditioning load time series pattern implicit distribution probability value to obtain the first multi-scale air-conditioning load time series pattern implicit characteristic distribution probability product;

[0084] Calculate the product of the second multi-scale air-conditioning load time series pattern implicit characteristic value of the multi-scale air-conditioning load time series pattern implicit representation vector and the absolute value of the difference between one and the multi-scale air-conditioning load time series pattern implicit distribution probability value to obtain the second multi-scale air-conditioning load time series pattern implicit characteristic inverse distribution probability product;

[0085] Calculate the quotient of the first multi-scale air-conditioning load time series pattern implicit characteristic value divided by the absolute value of the difference between one and the multi-scale air-conditioning load time series pattern implicit distribution probability value to obtain the first multi-scale air-conditioning load time series pattern implicit characteristic inverse distribution probability quotient;

[0086] Calculating the quotient of the second multi-scale air-conditioning load time series pattern implicit characteristic value and the multi-scale air-conditioning load time series pattern implicit distribution probability value to obtain the second multi-scale air-conditioning load time series pattern implicit characteristic distribution probability quotient;

[0087] After adding the product of the implicit characteristic distribution probability of the first multi-scale air-conditioning load time series pattern and the product of the implicit characteristic inverse distribution probability of the second multi-scale air-conditioning load time series pattern, subtracting the difference between the implicit characteristic inverse distribution probability quotient of the first multi-scale air-conditioning load time series pattern and the implicit characteristic distribution probability quotient of the second multi-scale air-conditioning load time series pattern to obtain the implicit correction characteristic value of the multi-scale air-conditioning load time series pattern;

[0088] For each first multi-scale air-conditioning load time series pattern implicit eigenvalue and second multi-scale air-conditioning load time series pattern implicit eigenvalue of the multi-scale air-conditioning load time series pattern implicit representation vector, perform matrix multiplication of a multi-scale air-conditioning load time series pattern implicit correction feature matrix composed of the multi-scale air-conditioning load time series pattern implicit correction eigenvalues ​​and the multi-scale air-conditioning load time series pattern implicit representation vector to obtain a corrected multi-scale air-conditioning load time series pattern implicit representation vector; and

[0089] The corrected multi-scale air-conditioning load time series pattern implicit representation vector is passed through the decoder-based air-conditioning load prediction module to obtain the air-conditioning load operation prediction value at the next moment.

[0090] The correction process of the implicit representation vector V of the multi-scale air-conditioning load time series pattern is expressed as:

[0091]

[0092] m i,j ∈M,vi ∈V,v j(j≠i) ∈V

[0093]

[0094] Wherein, V represents the implicit representation vector of the multi-scale air conditioning load time series pattern, v i represents the eigenvalue of the i-th position of the implicit representation vector of the multi-scale air conditioning load time series pattern, v j represents the eigenvalue of the jth position of the implicit representation vector of the multi-scale air-conditioning load time series pattern, p represents the implicit distribution probability value of the multi-scale air-conditioning load time series pattern, m i,j represents the implicit correction eigenvalue of the multi-scale air-conditioning load time series pattern, M represents the implicit correction characteristic matrix of the multi-scale air-conditioning load time series pattern composed of the implicit correction eigenvalue of the multi-scale air-conditioning load time series pattern, represents matrix multiplication, V ′ An implicit representation vector representing the corrected multi-scale air conditioning load time series pattern;

[0095] and,

[0096]

[0097] Among them, V1 is the implicit representation vector of the first scale air conditioning load time series pattern, f 1i represents the eigenvalue of the ith position of the implicit representation vector of the first scale air conditioning load time series pattern, and V2 is the implicit representation vector of the second scale air conditioning load time series pattern, f 2i represents the eigenvalue of the ith position of the implicit representation vector of the second-scale air-conditioning load time series pattern, V represents the implicit representation vector of the multi-scale air-conditioning load time series pattern, v i represents the eigenvalue of the ith position of the implicit representation vector of the multi-scale air-conditioning load time series pattern, and p represents the implicit distribution probability value of the multi-scale air-conditioning load time series pattern.

[0098] That is, when the implicit representation vector of the first-scale air-conditioning load timing pattern and the implicit representation vector of the second-scale air-conditioning load timing pattern respectively represent the implicit correlation features of air-conditioning load timing patterns of different scales determined based on the forward LSTM model and the 1D-CNN model and the feature timing dynamic propagation network, when multi-scale fusion is performed on them, the multi-scale air-conditioning load timing pattern implicit vector will also cause fusion diversity due to the difference in scales of receptive fields of different timing features, thereby affecting the consistency of decoding mapping iterations and reducing the accuracy of the decoding results.

[0099] Therefore, the applicant of the present application aims at the region-boundary integral relationship of the high-dimensional feature manifold of the implicit representation vector of the multi-scale air-conditioning load time series pattern, and defines the spatial distribution probability representation of the feature set of the implicit representation vector of the multi-scale air-conditioning load time series pattern as a manifold constraint boundary relative to the first-scale air-conditioning load time series pattern implicit representation vector and the second-scale air-conditioning load time series pattern implicit representation vector, so as to approximate the high-dimensional manifold of the feature set of the multi-scale air-conditioning load time series pattern implicit representation vector, which is composed of a single connected region representation of eigenvalue pairs, so as to avoid the ambiguity of the mapping of the local manifold representation of the multi-scale air-conditioning load time series pattern implicit in the diversified feature representation of the vector to the high-dimensional feature decoding regression convergence space, improve the execution iteration consistency of each local feature distribution in the mapping task, improve the feature decoding regression convergence effect, and improve the numerical accuracy of the air-conditioning load operation prediction value of the next moment obtained by the decoder-based air-conditioning load prediction module. In this way, there is no need to rely on a clear feature model, and it is possible to more flexibly handle complex and diverse load conditions, automatically learn and identify load features, and thus improve the accuracy of load identification. At the same time, by capturing load characteristics at different scales, key features can be obtained, effectively suppressing the impact of noise on load identification, thereby effectively solving the limitations of conventional deep learning models in dealing with noise interference and complex load data, and improving the noise resistance and prediction accuracy of load identification.

[0100] In summary, according to the embodiment of the present application, a non-intrusive air-conditioning load monitoring method based on Kalman filtering is explained, which uses a non-intrusive terminal to collect historical air-conditioning load data to form a one-dimensional time series. By setting the delay time and the embedding dimension, the time series is reconstructed in phase space to obtain a high-dimensional space trajectory. The trajectory is slidingly sampled using a predetermined time window, and the local average of each sampling window is calculated to form a local average sequence. Based on this sequence, the air-conditioning load at the next moment is predicted, and the Kalman gain is calculated in combination with the actual measurement data, and the predicted value is corrected to obtain an optimized prediction result. In this way, the air-conditioning load can be estimated more accurately and the prediction error can be effectively corrected, thereby improving the accuracy and real-time performance of load identification.

[0101] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0102] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0103] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0104] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0105] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the device claim can also be implemented by one unit through software or hardware.

[0106] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A non-intrusive air conditioning load monitoring method based on Kalman filtering, characterized in that: include: Obtain the historical load data of the monitored air conditioner through the non-intrusive measurement terminal; Constructing the historical load data of the monitored air conditioner into a one-dimensional time series of air conditioner load; Based on the delay time and the embedding dimension, the one-dimensional time series of the air-conditioning load is reconstructed in phase space to obtain a high-dimensional space trajectory of the air-conditioning load series; Sliding sampling is performed on the high-dimensional space trajectory of the air-conditioning load sequence based on a predetermined time window to obtain a sequence of sampling windows of the air-conditioning load sequence trajectory; Calculating the local average of each air conditioning load sequence trajectory sampling window in the sequence of air conditioning load sequence trajectory sampling windows to obtain a sequence of local average values; Based on the sequence of the local average values, calculating the predicted value of the air conditioning load operation at the next moment; Calculating the Kalman gain between the predicted value of the air conditioning load operation at the next moment and the actual measurement data; The air conditioning load operation prediction value at the next moment is corrected based on the Kalman gain to obtain an optimized air conditioning load operation prediction value at the next moment.

2. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 1 is characterized in that: Based on the delay time and the embedding dimension, the one-dimensional time series of the air-conditioning load is reconstructed in phase space to obtain a high-dimensional space trajectory of the air-conditioning load series, including: Determine an initial delay time, and perform delay processing on the one-dimensional time series of the air-conditioning load based on the initial delay time to obtain a one-dimensional delayed time series of the air-conditioning load; Calculating the mutual information between the one-dimensional time series of the air conditioning load and the one-dimensional delayed time series of the air conditioning load, and selecting a first minimum value as the delay time based on the mutual information; constructing a phase space based on the one-dimensional time series of the air conditioning load and the delay time, and determining the embedding dimension based on false neighboring points of the phase space; The phase space of the one-dimensional time series of the air-conditioning load is reconstructed based on the delay time and the embedding dimension to obtain a high-dimensional space trajectory of the air-conditioning load series.

3. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 2 is characterized in that: Based on the sequence of the local average values, the air conditioning load operation prediction value at the next moment is calculated, including: Performing global trend recognition on the sequence of the local average values ​​to obtain an implicit representation vector of the time series pattern of the air-conditioning load at the first scale; Performing local trend recognition on the sequence of local average values ​​to obtain a sequence of implicit representation vectors of the local time series pattern of air-conditioning load; The sequence of the implicit representation vector of the local time series pattern of the air-conditioning load is subjected to dynamic time series principal component gradient attenuation propagation to obtain the implicit representation vector of the time series pattern of the air-conditioning load of the second scale, including: extracting the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load and the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load and the historical local time series pattern of the air-conditioning load from the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load, and performing principal component extraction to obtain the sequence of the local time series principal component representation vector of the current air-conditioning load and the sequence of the local time series principal component representation vector of the historical air-conditioning load; calculating the absolute factors of the time series propagation attenuation entropy of each sequence of the local time series principal component representation vector of the historical air-conditioning load relative to the local time series principal component representation vector of the current air-conditioning load to obtain the sequence of the absolute factors of the time series propagation attenuation entropy of the air-conditioning load; based on the sequence of the absolute factors of the time series propagation attenuation entropy of the air-conditioning load, dynamically propagating the sequence of the local time series principal component representation vector of the current air-conditioning load and the sequence of the local time series principal component representation vector of the historical air-conditioning load to obtain the implicit representation vector of the time series pattern of the air-conditioning load of the second scale; Fusion the first-scale air-conditioning load time series pattern implicit representation vector and the second-scale air-conditioning load time series pattern implicit representation vector to obtain a multi-scale air-conditioning load time series pattern implicit representation vector; Based on the implicit representation vector of the multi-scale air-conditioning load time series pattern, the air-conditioning load operation prediction value at the next moment is obtained.

4. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 3 is characterized in that: Extracting the sequence of the implicit representation vector of the local time series pattern of the current air-conditioning load and the implicit representation vector of the local time series pattern of the historical air-conditioning load from the sequence of the implicit representation vector of the local time series pattern of the air-conditioning load, and performing principal component extraction to obtain the sequence of the principal component representation vector of the local time series of the current air-conditioning load and the principal component representation vector of the local time series of the historical air-conditioning load, including: Extracting the implicit representation vector of the local time series pattern of the current air-conditioning load from the sequence of the implicit representation vectors of the local time series pattern of the air-conditioning load, and performing principal component extraction based on the eigenvalue on the implicit representation vector of the local time series pattern of the current air-conditioning load to obtain the principal component representation vector of the local time series of the current air-conditioning load; The other implicit representation vectors of local time series patterns of air-conditioning loads in the sequence of implicit representation vectors of local time series patterns of air-conditioning loads are defined as implicit representation vectors of local time series patterns of historical air-conditioning loads to obtain the sequence of implicit representation vectors of local time series patterns of historical air-conditioning loads, and principal component extraction based on eigenvalues ​​is performed on each implicit representation vector of local time series patterns of historical air-conditioning loads in the sequence of implicit representation vectors of local time series patterns of historical air-conditioning loads to obtain the sequence of principal component representation vectors of local time series patterns of historical air-conditioning loads.

5. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 4 is characterized in that: Calculating the absolute factors of the time series propagation attenuation entropy of each sequence of the local time series principal component representation vectors of the historical air conditioning loads in the sequence of the local time series principal component representation vectors of the historical air conditioning loads relative to the local time series principal component representation vector of the current air conditioning loads to obtain a sequence of absolute factors of the time series propagation attenuation entropy of the air conditioning loads, including: Calculate the positional division of the corresponding positions of the local time series principal component representation vector of the historical air conditioning load and the local time series principal component representation vector of the current air conditioning load to obtain the local time series principal component attenuation vector of the air conditioning load; Calculate the base-two logarithmic function value of the absolute value of each eigenvalue of the local time series principal component attenuation vector of the air conditioning load to obtain the local time series principal component attenuation logarithmic vector of the air conditioning load; Calculate the positional point product of the local time series principal component representation vector of the historical air conditioning load and the attenuation logarithm vector of the local time series principal component of the air conditioning load, and perform positional point multiplication on the obtained point product vector to obtain the time series propagation attenuation value of the air conditioning load; An exponential function of the air-conditioning load time series propagation attenuation value with the natural constant e as the base is calculated to obtain the air-conditioning load time series propagation attenuation entropy absolute factor.

6. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 5 is characterized in that: Based on the sequence of absolute factors of the air-conditioning load time series propagation attenuation entropy, the sequence of the current air-conditioning load local time series principal component representation vector and the sequence of the historical air-conditioning load local time series principal component representation vector are dynamically propagated to obtain the second-scale air-conditioning load time series pattern implicit representation vector, including: Based on the time span between each historical air-conditioning load local time series principal component representation vector in the sequence of the historical air-conditioning load local time series principal component representation vector and the current air-conditioning load local time series principal component representation vector, each air-conditioning load time series propagation attenuation entropy absolute factor in the sequence of the air-conditioning load time series propagation attenuation entropy absolute factor is modulated in the time dimension to obtain a sequence of air-conditioning load time series span modulated propagation attenuation entropy factors; Inputting the sequence of the air-conditioning load time series span modulation propagation attenuation entropy factors into the information transmission screening module based on the gating function to obtain the sequence of the air-conditioning load time series span modulation propagation attenuation weights; Based on the sequence of the air-conditioning load time series span modulation propagation attenuation weights, calculating the weighted sum of the sequence of the historical air-conditioning load local time series principal component representation vectors to obtain the historical air-conditioning load local time series principal component significant transfer aggregation representation vector; The position-wise sum of the historical air-conditioning load local time series principal component significant transfer aggregation representation vector and the current air-conditioning load local time series principal component representation vector is calculated to obtain the second-scale air-conditioning load time series pattern implicit representation vector.

7. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 6 is characterized in that: Based on the time span between each historical air-conditioning load local time series principal component representation vector in the sequence of the historical air-conditioning load local time series principal component representation vector and the current air-conditioning load local time series principal component representation vector, each air-conditioning load time series propagation attenuation entropy absolute factor in the sequence of the air-conditioning load time series propagation attenuation entropy absolute factor is modulated in time dimension to obtain a sequence of air-conditioning load time series span modulated propagation attenuation entropy factors, including: The timestamp of the local time series principal component representation vector of the current air conditioning load is subtracted from the timestamp of the local time series principal component representation vector of the historical air conditioning load, and then rounded down to obtain the air conditioning load time span value; an exponential function with the air conditioning load time span value as the exponent is calculated with the natural constant e as the base to obtain the air conditioning load time span modulation value; The air conditioning load time series propagation attenuation entropy absolute factor corresponding to the historical air conditioning load local time series principal component representation vector is divided by the air conditioning load time span modulation value to obtain the air conditioning load time series span modulation propagation attenuation entropy factor.

8. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 7 is characterized in that: Inputting the sequence of the air-conditioning load time series span modulation propagation attenuation entropy factors into the information transmission screening module based on the gating function to obtain the sequence of the air-conditioning load time series span modulation propagation attenuation weights, including: Comparing each air conditioning load time series span modulation propagation attenuation entropy factor in the sequence of the air conditioning load time series span modulation propagation attenuation entropy factors with a predetermined threshold value to obtain a sequence of the air conditioning load time series span modulation propagation attenuation weights; Among them, in response to the air-conditioning load timing span modulation propagation attenuation entropy factor being greater than the predetermined threshold, the air-conditioning load timing span modulation propagation attenuation entropy factor greater than the predetermined threshold is input into the si gmod function; in response to the air-conditioning load timing span modulation propagation attenuation entropy factor being less than or equal to the predetermined threshold, the air-conditioning load timing span modulation propagation attenuation entropy factor less than or equal to the predetermined threshold is set to zero.

9. The non-intrusive air conditioning load monitoring method based on Kalman filtering according to claim 8, characterized in that: Based on the implicit representation vector of the multi-scale air-conditioning load time series pattern, the air-conditioning load operation prediction value at the next moment is obtained, including: inputting the implicit representation vector of the multi-scale air-conditioning load time series pattern into an air-conditioning load prediction module based on a decoder to obtain the air-conditioning load operation prediction value at the next moment.

Images