Adaptive algorithm optimization control system and method for intelligent energy management

Abstract

The invention discloses a self-adaptive algorithm optimization control system and method for intelligent energy management, and relates to the technical field of energy management, and the method comprises the steps: determining the load change sensitivity according to a load change prediction rate; grouping the devices according to the load change sensitivity and the current load state to obtain a plurality of device load groups; mapping the load change prediction rate and the multi-dimensional dynamic threshold matrix to obtain a target control strategy; determining an initial control parameter according to the target control strategy; performing difference analysis on the operation parameter and the initial control parameter to obtain a control parameter difference value; adjusting the target control strategy according to the control parameter difference value so as to update the initial control parameter; the updated initial control parameters are input into an execution control unit; the adjusted load state of the building equipment is fed back in real time; according to the invention, accurate real-time prediction and control of building equipment load change are realized, and the stability and response efficiency of intelligent building energy management are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management, and in particular to an adaptive algorithm optimization control system and method for intelligent energy management. Background Art

[0002] With the development of the social economy and the acceleration of urbanization, the number of buildings continues to increase, and the energy consumption of building equipment has gradually become a critical issue in urban energy management. Currently, the main functions of building energy management systems are to monitor the operating status and energy consumption of equipment and perform basic regulatory control of the equipment. However, existing technologies have shown significant shortcomings when it comes to meeting the requirements of real-time and precise control under rapidly changing equipment loads and complex operating conditions.

[0003] Existing building energy management systems mostly employ fixed thresholds or periodic adjustment strategies. This control model cannot effectively address the dynamic and uncertain nature of building equipment loads and struggles to quickly respond to drastic load fluctuations, resulting in significant energy waste, poor control accuracy, and reduced equipment operating efficiency. Furthermore, existing solutions often rely on simple prediction methods or single model structures, failing to fully exploit the complex temporal and spatial characteristics implicit in historical data. This makes accurate load forecasting difficult, further impacting the real-time and accuracy of system control strategies.

[0004] On the other hand, existing load sensitivity classification methods are mostly based on simple thresholds or subjective experience, lacking objective quantitative indicators and automated clustering methods for equipment load status. This leads to unclear sensitivity level classification and, consequently, ineffective targeted and refined control. Furthermore, existing technologies for real-time optimization of control parameters and closed-loop regulation generally suffer from poor real-time performance and insufficient stability, resulting in significant deviations between equipment operating parameters and actual requirements, affecting the intelligent level and economic effectiveness of energy management. Summary of the Invention

[0005] The purpose of the present invention is to provide an adaptive algorithm optimization control method and system for intelligent energy management to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides an adaptive algorithm optimization control method for intelligent energy management, comprising: S101: collecting the current load status of multiple devices in the building in real time; and constructing a load change rate prediction model based on historical load data and current load status to obtain the load change prediction rate of the device; S102: Determine the load change sensitivity of the building equipment according to the load change prediction rate; and group all the building equipment according to the load change sensitivity and the current load status to obtain multiple equipment load groups; S103: Mapping the load change prediction rate with a pre-established multi-dimensional dynamic threshold matrix to obtain target control strategies corresponding to multiple equipment load groups; and determining initial control parameters for building equipment based on the target control strategies; S104: performing a difference analysis on the operating parameters obtained by real-time monitoring of the building equipment and the initial control parameters to obtain a control parameter difference value; and adjusting the target control strategy according to the control parameter difference value to update the initial control parameters; S105: Input the updated initial control parameters into the execution control unit of the building equipment; and provide real-time feedback on the load status of the building equipment after adjustment, forming a real-time closed-loop control process.

[0007] In a second aspect, the present invention provides an adaptive algorithm optimization control system for intelligent energy management, which is implemented based on the adaptive algorithm optimization control method for intelligent energy management described above, and includes: The acquisition module is used to collect the current load status of multiple devices in the building in real time; and build a load change rate prediction model based on historical load data and current load status to obtain the load change prediction rate of the device; A grouping module is used to determine the load change sensitivity of building equipment according to the load change prediction rate; and group all building equipment according to the load change sensitivity and current load status to obtain multiple equipment load groups; A determination module is used to map the load change prediction rate with a pre-established multi-dimensional dynamic threshold matrix to obtain a target control strategy corresponding to multiple equipment load groups; and determine the initial control parameters of the building equipment according to the target control strategy; An update module is used to perform a difference analysis on the operating parameters obtained by real-time monitoring of building equipment and the initial control parameters to obtain a difference value of the control parameters; and to adjust the target control strategy according to the difference value of the control parameters to update the initial control parameters; The feedback module is used to input the updated initial control parameters into the execution control unit of the building equipment; and to provide real-time feedback on the adjusted load status of the building equipment, forming a real-time closed-loop control process.

[0008] In the above technical solution, the technical effects and advantages provided by the present invention are: The present invention constructs a load change rate prediction model that integrates convolutional-long short-term memory network and adaptive attention mechanism to accurately predict the load change trend of building equipment in real time, realizes efficient perception and response to rapid load changes, overcomes the problems of insufficient prediction accuracy and slow response of traditional models, and effectively improves the accuracy and timeliness of real-time control of equipment load.

[0009] The present invention objectively quantifies and finely divides the sensitivity of equipment load changes through an improved density peak clustering algorithm, effectively avoiding the problems of strong subjectivity and fuzzy level boundaries in the prior art in sensitivity division, significantly improving the scientificity and pertinence of equipment grouping management, and thus laying a solid foundation for the implementation of differentiated and precise control strategies.

[0010] The present invention achieves accurate real-time closed-loop optimization of building equipment control parameters by real-time monitoring of the differences between operating parameters and initial control parameters, and performs real-time feedback adjustment based on dynamic weighted cosine similarity. It effectively solves the problems of insufficient closed-loop regulation stability and untimely optimization of control parameters in existing solutions, and significantly improves the stability and economy of smart building energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0012] Figure 1 This is a flow chart of an adaptive algorithm optimization control method for intelligent energy management according to the present invention; Figure 2 This is a framework diagram of an adaptive algorithm optimization control system for intelligent energy management according to the present invention. DETAILED DESCRIPTION

[0013] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be more comprehensive and complete, and will fully convey the concepts of the example embodiments to those skilled in the art. The accompanying drawings are merely schematic illustrations of the disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures indicate identical or similar parts, and thus any repetitive description thereof will be omitted.

[0014] In addition, the described features, structures or characteristics can be combined in one or more example embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the example embodiments disclosed in this application. However, those skilled in the art will appreciate that the technical solutions disclosed in this application can be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. can be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring the various aspects disclosed in this application.

[0015] Example 1 like Figure 1 As shown, this embodiment discloses an adaptive algorithm optimization control method for intelligent energy management, including: S101: collecting the current load status of multiple devices in the building in real time; and constructing a load change rate prediction model based on historical load data and current load status to obtain the load change prediction rate of the device; In this implementation, in order to achieve real-time and accurate monitoring of various equipment loads in the smart building energy management system, a distributed real-time data acquisition system is first deployed and constructed; Specifically, each building's equipment, such as air conditioning equipment, elevator systems, lighting devices, electric heating devices, etc., are equipped with real-time data collection devices such as load sensors, power monitors, and operating status sensors to continuously monitor various load parameters of the equipment during actual operation, including but not limited to: real-time power, Real-time current, voltage and equipment operating status (switch status), the status is represented by digital signals (0 / 1); For example, the typical data structure of the real-time load status of each device is shown in Table 1: Table 1: Equipment real-time load status data table Timestamp Equipment Number Power (kW) Current (A) Voltage (V) state 2024-05-01 10:01:05 AC-01 7.5 15.2 220 1 2024-05-01 10:01:05 EL-02 3.3 8.9 220 1 In the system, the sampling period for real-time status data of all devices is set to 5 seconds, that is, the latest status data of all devices is collected every 5 seconds; the collected data is packaged in real time in JSON format and sent to the system server via MQTT protocol or HTTP protocol; In implementation, the load change rate prediction model is constructed based on historical load data and current load status, including: Use historical load data to divide the equipment load state into time series, and build an initial load feature set based on the divided sequence characteristics; For example, the system first extracts the continuous operating load data of each building equipment for the past R (e.g., R=3) months from the historical database, recording the data once every minute to form a time series: Assume that the original historical load data sequence of each device is: ; Where: Indicates the load power value of the device at time t, represents the historical load data sequence of the equipment, 𝑛 is the number of historical sequence data points (e.g., 1 per minute for 3 months); Next, we perform windowed time series division on the original sequence. The specific division method is as follows: The sequence data is divided into sliding windows with a step length of 30 minutes and a step length of 5 minutes. Each window contains 30 data points. The following features are calculated and extracted in each window: Average load power value : ; Load power variance : ; Where: N is the number of data points in the sliding window (e.g. 30 points in 30 minutes), is the load power value in the i-th time window; Maximum and minimum load values within the window ; Load change trend indicator: The slope k is obtained by performing linear regression fitting calculation on the load sequence within the window, which represents the load change trend; Through the above process, the initial load characteristic set of each device is formed. The data example is shown in Table 2: Table 2: Initial load characteristic set data table Window Number Average (kW) variance Maximum value (kW) Minimum value (kW) Trend (kW / min) W-01 8.1 0.45 8.7 7.2 0.02 This step provides a clear and sufficient data basis for the input of the load forecasting model; The initial load feature set is input into the convolutional-long short-term memory neural network model for training, and the network structure is dynamically optimized based on the prediction error output by the model; It should be noted that a convolutional-long short-term memory (Conv-LSTM) neural network is used as the basic model for load change rate prediction. This model combines the spatial feature extraction capabilities of a convolutional neural network (CNN) with the sequential data processing capabilities of a long short-term memory network (LSTM). The model input is the initial load feature set described above. The model output is the predicted rate of change of the equipment load for the next 5 minutes (for example, the change per minute). During training, the mean squared error (MSE) is used as the loss function, and the model output error is evaluated after every 10 epochs of training. If the prediction error does not decrease significantly (for example, the error decrease rate is less than 1%), the model structure is dynamically optimized, and the number of network layers or hidden units is adjusted to ensure that the model prediction accuracy gradually improves and converges. Specifically, the initial structure of the model is set as follows: number of convolutional layers: 2 layers, filter size 3×3, number of filters 32; number of LSTM hidden units: 64; fully connected layers: 2 layers, containing 128 and 32 neurons respectively; output layer: 1 linear neuron; Adopting the adaptive attention mechanism to adjust the feature weights during the network training process, a load change rate prediction model with convergence training is obtained; The method of using an adaptive attention mechanism to adjust feature weights during network training includes: Dynamically determine the feature attention weight distribution based on the loss value change trend during the model training process; The dynamically determining feature attention weight distribution includes: Calculate the correlation coefficient between the feature dimension and the model output error after each training; It should be understood that after each epoch of model training, the Pearson correlation coefficient is calculated between each input feature dimension (such as average load, variance, maximum value, etc.) and the current epoch prediction error to quantify the contribution of the feature dimension to the model error. The formula is: ; Where: is the value of the i-th feature dimension on the j-th sample, is the prediction error of the jth sample, and are the mean values of the corresponding variables, 𝑛 is the number of samples; Normalize the correlation coefficient to obtain the initial weight coefficient of the feature dimension; The correlation coefficients calculated above are normalized using the Min-Max method (i.e., the features are normalized to [0, 1]) to obtain the initial weight coefficients of each feature dimension. The weight adjustment gradient is calculated based on the rate of change of the initial weight coefficients from the most recent training. The specific calculation formula is: ; Where: is the weight coefficient at the kth Epoch, Adjust the gradient for the weight of the i-th feature dimension; The gradient descent method is used to dynamically optimize the feature attention weights according to the weight adjustment gradient to obtain a stable attention weight distribution, specifically: ; Where: For the The feature weight after Epoch update, is the feature weight of the t-th Epoch, is the learning rate (typically set to a value between 0.01 and 0.1). This process ensures that the weights are continuously optimized in the direction of reducing the model error; Re-weight the input feature sequence according to the attention weight distribution, and feedback to adjust the model parameters. The specific formula is: ; Where: is the original feature dimension, is the weighted feature sequence; Alternately perform feature weight updates and model parameter adjustments until the attention weight distribution reaches a stable state; For example, the weighted feature sequence is re-input into the Conv-LSTM model for the next round of model parameter training and adjustment; the weight optimization and model parameter adjustment process is repeated until the amplitude of the weight coefficient change for three consecutive times is less than the set convergence threshold (for example, 0.01), that is: , judge that the feature attention weight has reached a stable distribution state and stop adjusting; The training model after determining that the feature attention weight is stable is the final load change rate prediction model; The current load state is input into the trained and converged load change rate prediction model, and the load change prediction rate of the device is output; For example, the system receives the current load status data (real-time power, current, voltage, and operating status) of each device in real time. After processing the current real-time data, it forms a corresponding feature sequence, which is input into the Conv-LSTM model after the training convergence described above. Based on this input, the model performs real-time forward prediction and outputs the predicted load change rate of each device in the future set time period (e.g., the next 5 minutes). The following table shows an example of a typical output result: Table 3: Data table before and after model prediction equipment Current load (kW) Predicted load change rate (kW / min) AC-01 7.5 0.14 EL-02 3.2 -0.08 The above real-time prediction results are used for real-time load control decisions of the subsequent building energy management system, forming a load control closed loop.

[0016] S102: Determine the load change sensitivity of the building equipment according to the load change prediction rate; and group all the building equipment according to the load change sensitivity and the current load status to obtain multiple equipment load groups; The step of determining the load change sensitivity of the building equipment according to the load change prediction rate includes: Calculate the fluctuation entropy value of equipment load change based on the load change prediction rate; It should be noted that to calculate the fluctuation entropy of device load changes, the load change rate data obtained from the most recent continuous forecast of each device is first discretized to form multiple discrete intervals. The number of intervals is dynamically determined based on the data range, and the typical number is 5-10. Each interval is recorded as: ; Then, for each interval, count the frequency of the corresponding predicted rate data falling into it and calculate the probability: , where: To fall into the interval The amount of data, is the total amount of data; finally, based on the above probability The information entropy formula is used to calculate the load change fluctuation entropy value of each device: , where: M is the number of discretization intervals, is the logarithmic function (natural logarithm); It should be understood that: the larger the entropy value, the more complex the equipment load fluctuation is and the more sensitive it is to environmental changes; Based on the fluctuation entropy value, the device load change sensitivity is preliminarily divided using the improved density peak clustering algorithm; To further clarify the sensitivity classification, this embodiment adopts an improved Density Peak Clustering (DPC) algorithm to convert the fluctuation entropy value into a clear and distinct preliminary sensitivity classification; The preliminary classification of device load change sensitivity using the improved density peak clustering algorithm includes: Map the load change fluctuation entropy value to the feature vector in the cluster space ,in, is the fluctuation entropy value of the i-th device, and N is the total number of devices; Calculate the density value and relative distance for each eigenvector and determine the initial cluster center; Among them, the density value The formula is: ; Where: is the eigenvector With the eigenvector The Euclidean distance between is the cutoff distance, which is typically set to the first 2% to 5% of the eigenvector distances; Among them, the relative distance The formula is: ; Dynamically adjust the density peak parameters according to the initial cluster center to achieve adaptive merging of clusters; It should be noted that the initial cluster center is determined by multiplying the density value by the relative distance. The first few (typically 3-5) of the clusters are selected as the initial cluster centers. The improved density peak algorithm adaptively merges clusters. It dynamically adjusts the number and boundaries of clusters based on parameters such as the number of devices in the cluster and the density difference between clusters. It uses the inter-cluster density comparison method to adaptively merge clusters with close distances and similar densities to ensure the stability of the clustering results. Output the final stable device sensitivity clustering results; Generate device sensitivity classification levels based on clustering results and perform stability assessment on the classification levels; It should be understood that to obtain a final, clear and reliable load sensitivity classification, the above clustering results are directly mapped to multiple levels (such as "high", "medium", and "low"). To ensure the stability of the classification results, multiple sampling verifications are performed (for example, clustering is repeated 50 times), and the consistency ratio of each device classification is calculated: ,Select the classification level result with the highest stability (i.e., the highest consistency) as the final load change sensitivity level of each device; Finally, the highest stability classification level is determined as the equipment load change sensitivity; For example, all devices are grouped according to the sensitivity level and the current real-time load status data of the device, combined with the load range or type (such as high, medium, and low load), as shown in Table 4 below: Table 4: Equipment group data table Group Sensitivity level Current load range G1 High sensitivity High load (>8kW) G2 Medium sensitivity Medium load (5~8kW) G3 Low sensitivity Low load (<5kW) All building equipment can be grouped into load groups based on load change sensitivity and current load status.

[0017] S103: Mapping the load change prediction rate with a pre-established multi-dimensional dynamic threshold matrix to obtain target control strategies corresponding to multiple equipment load groups; and determining initial control parameters for building equipment based on the target control strategies; The method for constructing the pre-established multi-dimensional dynamic threshold matrix includes: Calculating an error covariance matrix based on an error matrix between historical load prediction rates and actual load change rates; Specific implementation details: Collect the load prediction rate data and the corresponding actual load change rate data of each device over a period of time (for example, the last 6 months); construct an error matrix for the historical prediction error recorded for each device , where the element in row i and column j is the prediction error of the i-th device at the j-th moment ,in, is the predicted rate, is the actual rate, according to the error matrix Calculate the error covariance matrix ,in, represents the error mean vector, n is the number of data samples; The covariance matrix is used to perform principal component analysis on the load forecast error to obtain the principal component error eigenvector, including: Calculate the eigenvalues and eigenvectors of the error covariance matrix; The eigenvectors whose cumulative contribution rate reaches a preset value (such as 90%) are selected as the principal component eigenvectors, which are specifically expressed as follows: ; The principal component error eigenvector is predicted using the autoregressive sliding average algorithm and a dynamic threshold sequence is output; The method of predicting the principal component error eigenvector using an autoregressive sliding average algorithm includes: Perform the first-order difference of the principal component error eigenvector to obtain a stationary sequence. The specific formula is as follows: ; Where: is the first-order difference value of the i-th principal component error eigenvector at time t, is the original value of the ith principal component error eigenvector; Determine the initial order of the model based on the autocorrelation function of the stationary series; Using stationary series Calculate the autocorrelation function (ACF) and partial autocorrelation function (PACF), and preliminarily determine the initial autoregressive order p and sliding average order q of the model through the function curve; The model order is optimally selected using the Akaike Information Criterion to obtain the autoregressive moving average model parameters; The optimal selection of the model order by the Akaike Information Criterion includes: Set multiple alternative order parameter combinations; For example, set several alternative order parameter combinations: (p,q), typical alternative combinations such as (1,1), (2,2), (3,1), (1,3), etc. Calculate the Akaike Information Criterion value of the model corresponding to each alternative parameter combination; For each set of alternative parameter combinations (p,q), the corresponding ARMA model is fitted using historical data, and the Akaike Information Criterion value is calculated: ; Where: is the number of model parameters, L is the maximum likelihood estimate of the model; The parameter combination that minimizes the Akaike Information Criterion value is selected as the optimal order of the model; The final autoregressive moving average model is constructed using the optimal order parameter combination: ; Where: Model parameters , By using the least squares method to estimate, represents the residual term; The model with determined parameters is used to predict the error characteristic vector values at multiple moments in the future, and the prediction results are output as a dynamic threshold sequence; Use the above ARMA model to predict the changes in the error feature vector at multiple moments in the future (such as the next 30 minutes) and output the error feature prediction sequence: ; Further reverse difference operation is restored to dynamic threshold sequence: ; Constructing a multi-dimensional dynamic threshold matrix according to the dynamic threshold sequence; Combine the predicted values of each principal component error eigenvector into a multidimensional dynamic threshold matrix: ; A matrix is obtained at each moment and dynamically updated for accurate mapping of subsequent control strategies.

[0018] S104: performing a difference analysis on the operating parameters obtained by real-time monitoring of the building equipment and the initial control parameters to obtain a control parameter difference value; and adjusting the target control strategy according to the control parameter difference value to update the initial control parameters; The performing of difference analysis on the operating parameters and initial control parameters obtained through real-time monitoring of building equipment includes: Extract the operating parameter sequence obtained from real-time monitoring of building equipment and calculate the gradient vector of the sequence; Specifically, the operating parameter sequence of building equipment in the most recent continuous period (e.g., the last 10 minutes) is monitored and extracted in real time, for example: ; Where: Represents the operating parameter value (such as equipment power or operating load) obtained by real-time monitoring at time t; Calculate the time-varying gradient vector based on the above operating parameter sequence : ; This gradient represents the real-time changing trend of the equipment operating parameters in the time dimension; Extract the corresponding initial control parameter sequence and calculate its average change gradient vector based on the sliding window method; For example, a sliding window method (e.g., a window length of 5 minutes and a sliding step of 1 minute) is used to calculate the average change gradient vector of the initial control parameter sequence: ; Where: W is the sliding window length, for example, a typical value is 5; The similarity coefficient between the operating parameter change gradient vector and the initial control parameter change gradient vector is calculated using cosine similarity; The method of calculating the similarity coefficient between the operating parameter change gradient vector and the initial control parameter change gradient vector using cosine similarity includes: Get the inner product of the corresponding dimensions of the operating parameter change gradient vector and the initial control parameter change gradient vector: ; Where m represents the vector length (i.e. the number of calculation dimensions); Calculate the modulus of the operating parameter change gradient vector and the initial control parameter change gradient vector respectively: ; Substitute the inner product and the modulus value into the cosine similarity calculation formula to calculate the initial similarity coefficient: ; Where: 、 are the gradient vectors of operating parameters and control parameters respectively, is the vector modulus; The dynamic weighted smoothing algorithm is used to smooth the initial similarity coefficient to obtain the final similarity coefficient; Since data fluctuations may occur in a real-time environment, in order to improve the stability of the similarity coefficient, this embodiment discloses the implementation of a dynamic weighted smoothing algorithm. The specific implementation process is as follows: Assume that the initial similarity coefficient sequence of the most recent consecutive moments (for example, the last 3 minutes) is: ; According to the variance of the last three initial similarity coefficients , calculate the dynamic smoothing weight: ; Apply the smoothing weights for weighted averaging to obtain the final smoothed similarity coefficient: ; Determine the control parameter difference value according to the similarity coefficient, and output the determined control parameter difference value, including: like A first similarity coefficient threshold is preset (e.g., 0.9), and the control parameter difference value is determined to be "small", that is, the current operating state is close to the control parameter, and there is no significant difference; If the second similarity coefficient threshold is preset (such as 0.7) If the first similarity coefficient threshold is preset (e.g. 0.9), the difference value is determined to be "medium" and may need to be adjusted appropriately; If the second similarity coefficient threshold is preset (such as 0.7) If a second similarity coefficient threshold is preset (e.g., 0.7), the difference value is determined to be “large” and requires significant adjustment; S105: Input the updated initial control parameters into the execution control unit of the building equipment; and provide real-time feedback on the load status of the building equipment after adjustment, forming a real-time closed-loop control process; The step of inputting the updated initial control parameters into the execution control unit of the building equipment includes: Determine the real-time control response delay of equipment based on the execution characteristics database of building equipment; It should be understood that before the equipment is put into operation, the actual response time of each device to control instructions, namely the control response delay, is clearly measured and recorded through field tests and historical data accumulation, and an execution characteristic database is generated based on this. Therefore, during the specific implementation process, the real-time control response delay of the device can be dynamically determined based on the type of device and the type of control action by accessing the execution characteristic database in real time. Determine the time to send the device control signal based on the real-time control response delay and the updated initial control parameters; To precisely control the timing of parameter delivery and ensure that the device response accurately matches the control instruction, this embodiment discloses the following specific implementation: Combine the updated initial control parameters with the above-determined device real-time control response delay to clearly calculate the sending time of the device control signal , the specific calculation formula is: ; Where: The time when the equipment is expected to reach the target load state is determined by the real-time control target of the system; is the real-time control response delay determined above; When the device control signal is sent, the updated initial control parameters are sent to the execution control unit; Collect the load status of building equipment in real time and feed it back to the control decision unit of the system; Specifically, building equipment uses built-in load sensors to collect its own adjusted load status information (such as power, load current, temperature and other parameters) in real time; the load status data collected by the sensor is clearly sampled at a high frequency (for example, the sampling period is 1 second); the collected data is fed back to the system's control decision unit in real time, and the data communication transmission process is clarified. For example: the sensor collected data is uploaded to the control decision unit in real time through wireless transmission technology (such as LoRa or ZigBee).

[0019] Example 2 like Figure 2 As shown, the parts not described in detail in this embodiment are as shown in Example 1. This embodiment discloses an adaptive algorithm optimization control system for intelligent energy management, including: Acquisition module 201 is used to collect the current load status of multiple devices in the building in real time; and build a load change rate prediction model based on historical load data and current load status to obtain the load change prediction rate of the device; The grouping module 202 is configured to determine the load change sensitivity of the building equipment according to the load change prediction rate; and group all the building equipment according to the load change sensitivity and the current load status to obtain a plurality of equipment load groups; Determination module 203 is used to map the load change prediction rate with a pre-established multi-dimensional dynamic threshold matrix to obtain target control strategies corresponding to multiple equipment load groups; and determine initial control parameters of building equipment according to the target control strategies; An updating module 204 is configured to perform a difference analysis on the operating parameters obtained by real-time monitoring of building equipment and the initial control parameters to obtain a control parameter difference value; and to adjust the target control strategy based on the control parameter difference value to update the initial control parameters; The feedback module 205 is used to input the updated initial control parameters into the execution control unit of the building equipment; and to provide real-time feedback on the load status of the building equipment after adjustment, forming a real-time closed-loop control process.

[0020] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters, weights and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0021] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium accessible by a computer, or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, or magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0022] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. An adaptive algorithm optimization control method for intelligent energy management, characterized in that: include: S101: real-time collection of the current load status of multiple devices in the building; A load change rate prediction model is constructed based on historical load data and current load status to obtain the load change prediction rate of the equipment; S102: Determine the load change sensitivity of the building equipment according to the load change prediction rate; and group all the building equipment according to the load change sensitivity and the current load status to obtain multiple equipment load groups; S103: Mapping the load change prediction rate with a pre-established multi-dimensional dynamic threshold matrix to obtain target control strategies corresponding to multiple equipment load groups; and determining initial control parameters for building equipment based on the target control strategies; S104: performing a difference analysis on the operating parameters obtained by real-time monitoring of the building equipment and the initial control parameters to obtain a control parameter difference value; and adjusting the target control strategy according to the control parameter difference value to update the initial control parameters; S105: Inputting the updated initial control parameters into the execution control unit of the building equipment; And provide real-time feedback on the load status of building equipment after adjustment, forming a real-time closed-loop control process.

2. The adaptive algorithm optimization control method for intelligent energy management according to claim 1 is characterized in that: The load change rate prediction model is constructed based on historical load data and current load status, including: Use historical load data to divide the equipment load state into time series, and build an initial load feature set based on the divided sequence characteristics; The initial load feature set is input into the convolutional-long short-term memory neural network model for training, and the network structure is dynamically optimized based on the prediction error output by the model; Adopting the adaptive attention mechanism to adjust the feature weights during the network training process, a load change rate prediction model with convergence training is obtained; The current load state is input into the trained and converged load change rate prediction model, and the load change prediction rate of the device is output.

3. The adaptive algorithm optimization control method for intelligent energy management according to claim 2 is characterized in that: The adaptive attention mechanism is used to adjust the feature weights during the network training process, including: Dynamically determine the feature attention weight distribution based on the loss value change trend during the model training process; The dynamically determining feature attention weight distribution includes: Calculate the correlation coefficient between the feature dimension and the model output error after each training; Normalize the correlation coefficient to obtain the initial weight coefficient of the feature dimension; The weight adjustment gradient is calculated based on the rate of change of the initial weight coefficients from the most recent training. The specific calculation formula is: ； Where: is the weight coefficient at the kth Epoch, Adjust the gradient for the weight of the i-th feature dimension; The gradient descent method is used to dynamically optimize the feature attention weights according to the weight adjustment gradient to obtain a stable attention weight distribution, specifically: ； Where: For the The feature weight after Epoch update, is the feature weight of the t-th Epoch, is the learning rate; Re-weight the input feature sequence according to the attention weight distribution, and feedback to adjust the model parameters. The specific formula is: ； Where: is the original feature dimension, is the weighted feature sequence; Alternately perform feature weight updates and model parameter adjustments until the attention weight distribution reaches a stable state; The training model after determining that the feature attention weight is stable is the final load change rate prediction model.

4. The adaptive algorithm optimization control method for intelligent energy management according to claim 3 is characterized in that: Determining the load change sensitivity of the building equipment according to the load change prediction rate includes: Calculate the fluctuation entropy value of equipment load change based on the load change prediction rate; Based on the fluctuation entropy value, the device load change sensitivity is preliminarily divided using the improved density peak clustering algorithm; The preliminary classification of device load change sensitivity using the improved density peak clustering algorithm includes: Map the load change fluctuation entropy value to the feature vector in the cluster space ,in, is the fluctuation entropy value of the i-th device, and N is the total number of devices; Calculate the density value and relative distance for each eigenvector and determine the initial cluster center; Among them, the density value The formula is: ； Where: is the eigenvector With the eigenvector The Euclidean distance between is the cutoff distance, which is typically set to the first 2% to 5% of the eigenvector distances; Among them, the relative distance The formula is: ； Dynamically adjust the density peak parameters according to the initial cluster center to achieve adaptive merging of clusters; Output the final stable device sensitivity clustering results; Generate device sensitivity classification levels based on clustering results and perform stability assessment on the classification levels; Finally, the classification level with the highest stability is determined as the equipment load change sensitivity.

5. The adaptive algorithm optimization control method for intelligent energy management according to claim 4 is characterized in that: The method for constructing the pre-established multi-dimensional dynamic threshold matrix includes: Calculating an error covariance matrix based on an error matrix between historical load prediction rates and actual load change rates; The covariance matrix is used to perform principal component analysis on the load forecast error to obtain the principal component error eigenvector, including: Calculate the eigenvalues and eigenvectors of the error covariance matrix; Select the eigenvector whose cumulative contribution rate reaches a preset value or above as the principal component eigenvector; The principal component error eigenvector is predicted using the autoregressive sliding average algorithm and a dynamic threshold sequence is output; A multidimensional dynamic threshold matrix is constructed based on the dynamic threshold sequence.

6. The adaptive algorithm optimization control method for intelligent energy management according to claim 5, characterized in that: The method of predicting the principal component error eigenvector using an autoregressive sliding average algorithm includes: Perform the first-order difference of the principal component error eigenvector to obtain a stationary sequence. The specific formula is as follows: ； Where: is the first-order difference value of the i-th principal component error eigenvector at time t, is the original value of the ith principal component error eigenvector; Determine the initial order of the model based on the autocorrelation function of the stationary series; The model order is optimally selected using the Akaike Information Criterion to obtain the autoregressive moving average model parameters; The model with determined parameters is used to predict the error feature vector values at multiple moments in the future, and the prediction results are output as a dynamic threshold sequence.

7. The adaptive algorithm optimization control method for intelligent energy management according to claim 6, characterized in that: The optimal selection of the model order by the Akaike Information Criterion includes: Set multiple alternative order parameter combinations; Calculate the Akaike Information Criterion value of the model corresponding to each alternative parameter combination; The parameter combination that minimizes the Akaike Information Criterion value is selected as the optimal order of the model; The final autoregressive moving average model is constructed using the optimal order parameter combination: ； Where: Model parameters , By using the least squares method to estimate, represents the residual term.

8. The adaptive algorithm optimization control method for intelligent energy management according to claim 7, characterized in that: The performing of difference analysis on the operating parameters and initial control parameters obtained through real-time monitoring of building equipment includes: Extract the operating parameter sequence obtained from real-time monitoring of building equipment and calculate the gradient vector of the sequence; Extract the corresponding initial control parameter sequence and calculate its average change gradient vector based on the sliding window method; The similarity coefficient between the operating parameter change gradient vector and the initial control parameter change gradient vector is calculated using cosine similarity; Determine the control parameter difference value according to the similarity coefficient, and output the determined control parameter difference value, including: like A first similarity coefficient threshold is preset), then the control parameter difference value is determined to be "small", that is, the current operating state is close to the control parameter; If the second similarity coefficient threshold is preset The first similarity coefficient threshold is preset, and the difference value is determined to be "medium"; If the second similarity coefficient threshold is preset A second similarity coefficient threshold is preset (eg, 0.7), and the difference value is determined to be "large".

9. The adaptive algorithm optimization control method for intelligent energy management according to claim 8, characterized in that: The method of calculating the similarity coefficient between the operating parameter change gradient vector and the initial control parameter change gradient vector using cosine similarity includes: Get the inner product of the corresponding dimensions of the operating parameter change gradient vector and the initial control parameter change gradient vector: ； Where m represents the vector length; Calculate the modulus of the operating parameter change gradient vector and the initial control parameter change gradient vector respectively: ； Substitute the inner product and the modulus value into the cosine similarity calculation formula to calculate the initial similarity coefficient: ； Where: 、 are the gradient vectors of operating parameters and control parameters respectively, is the vector modulus; The dynamic weighted smoothing algorithm is used to smooth the initial similarity coefficient to obtain the final similarity coefficient.

10. An adaptive algorithm optimization control system for intelligent energy management, implemented based on the adaptive algorithm optimization control method for intelligent energy management according to any one of claims 1 to 9, characterized in that: include: Acquisition module, used to collect the current load status of multiple devices in the building in real time; A load change rate prediction model is constructed based on historical load data and current load status to obtain the load change prediction rate of the equipment; A grouping module is used to determine the load change sensitivity of building equipment according to the load change prediction rate; and group all building equipment according to the load change sensitivity and current load status to obtain multiple equipment load groups; A determination module is used to map the load change prediction rate with a pre-established multi-dimensional dynamic threshold matrix to obtain a target control strategy corresponding to multiple equipment load groups; and determine the initial control parameters of the building equipment according to the target control strategy; An update module is used to perform a difference analysis on the operating parameters obtained by real-time monitoring of building equipment and the initial control parameters to obtain a difference value of the control parameters; and to adjust the target control strategy according to the difference value of the control parameters to update the initial control parameters; A feedback module, configured to input updated initial control parameters into an execution control unit of the building equipment; And provide real-time feedback on the load status of building equipment after adjustment, forming a real-time closed-loop control process.

Images